IntelliBuddies®
Home Platform
Solutions
Resources
Perspective Pricing About Us
Free Trial
IntelliBuddies®
Home Platform
P2P Solutions O2C Solutions Reconciliation Insurance Processing
Blog Events Newsroom Videos Impactful Areas Stories of Impact
Perspective Pricing About Us
Free Trial
All Solutions
P2P Solutions Procure-to-Pay end-to-end automation
  • Finance
  • Healthcare
  • Operations
  • Manufacturing
O2C Solutions Order-to-Cash process automation
  • Retail
  • Manufacturing
  • SaaS
Reconciliation Account & financial reconciliation
  • Banking
  • Enterprise
  • Fintech
Insurance Processing Claims, policy & underwriting automation
  • Claims
  • Underwriting
  • Policy
Learn
Blog Latest news, tips, and insights
Events Webinars and conferences
Newsroom Press releases and announcements
Videos Demos, tutorials, and success stories
Impactful Areas Explore automation by industry
Stories of Impact Real-world success stories
Developer Resources
Documentation Guides, API references, and more
Tutorials Step-by-step developer guides
FAQ Frequently asked questions
IB-X™ · Agentic automation

AI agents that turn requests into completed work.

One platform that unifies conversational AI, workflow automation and RPA, so work finishes and people stay in charge of the decisions that matter.

Book a demoSee how it works

500+ connectors · SaaS or self-hosted

  1. Request capturedConversational Agent
  2. Intent understood and validatedConversational Agent
  3. Routed for approvalWorkflow Agent
  4. Systems updatedRPA Agent
  5. Work done, with a full audit trailAI Command Center
Trusted byBlackstone Talent GroupHyundaiJeejaySeedsTeknaThird Wave CoffeeTiavikTrimacVipraAtomberg
Our manifesto

Enterprises do not need more agents. They need work that gets finished.

  1. 1Work should get finished.
  2. 2Value multiplies, it does not add.
  3. 3Authority stays human.
Read the manifesto
The future is not more agents. It is agents that behave like one, for people who can trust them.Ravi Krishnamurthy, Managing Director & CEO, IntelliBuddies®
The IB-X™ platform

Three specialised agents. One platform.

Conversational Agent

Starts the work in plain language, in chat or voice, and understands what is being asked.

Explore →

Workflow Agent

Routes, approves, escalates and tracks SLAs across people, systems and other agents.

Explore →

RPA Agent

Updates the systems that have no API, on desktops, VMs and on-prem servers.

Explore →

How the three work together →

Solutions

Start from the process you want fixed.

Procure-to-Pay

From purchase request to paid invoice, every step checked and explained.

See P2P →

Order-to-Cash

An order should reach cash without anyone retyping it.

See O2C →

Reconciliation · SmartRecon

Match every rupee to the record it belongs to, with a plain account of month one and how it improves.

See SmartRecon →

Insurance Processing

One platform for every step of a claim, policy and underwriting file.

See Insurance →

Industries: BFSI · Healthcare and Life Sciences · Manufacturing and Supply Chain

Customers

Results from customers in production.

88,000

invoices processed a year

Automotive group

75%+

of service bookings automated

Dealership network

40%

faster work orders

Manufacturing enterprise

All 10 customer stories →

Perspective

From the CEO

All essays →
The Multiplier · 4 min read

Don't Add Tools. Multiply Outcomes.

Read the essay →
The Multiplier · 5 min read

The Seam Is Where Value Dies

Read the essay →
The Multiplier · 4 min read

Why Pilots Shine and Production Disappoints

Read the essay →
Security and role-based accessControl who can do what
Immutable audit trailWho decided what, and when
SaaS or self-hostedYour data stays where you need it
API and connector ecosystemFits the systems you already run

Turn your manual processes into automated outcomes.

Book a demoStart free trial
Perspective

How enterprise work gets finished.

Essays by Ravi Krishnamurthy on getting work finished, with people in charge of the decisions that matter.

The manifesto

Multiply, don't add.

Enterprises do not need more agents; they need work that gets finished, by systems that behave as one, in service of the people who do that work.

Read the manifesto
The future is not more agents. It is agents that behave like one, for people who can trust them.Ravi Krishnamurthy, Managing Director & CEO, IntelliBuddies®
From the CEO

Essays from the CEO

The Multiplier

The Multiplier · 4 min read

Don't Add Tools. Multiply Outcomes.

Read the essay →
The Multiplier · 5 min read

The Seam Is Where Value Dies

Read the essay →
The Multiplier · 4 min read

Why Pilots Shine and Production Disappoints

Read the essay →

Finished Work

Finished Work · 4 min read

The Industry Sold Answers. Enterprises Need Outcomes.

Read the essay →
Finished Work · 4 min read

Measure Finished Work, Not Activity

Read the essay →
Finished Work · 4 min read

Start With a Conversation, Scale Into a System

Read the essay →

Authority

Authority · 4 min read

An LLM Should Not Hold the Keys to the Ledger

Read the essay →
Authority · 3 min read

The Decisions That Must Stay Human

Read the essay →
Authority · 4 min read

Autonomy Without Accountability Is a Liability

Read the essay →

People First

People First · 3 min read

What We Owe the People Who Do the Work

Read the essay →
People First · 3 min read

Honest Curves: Why Month One Looks Worse Than Month Twelve

Read the essay →
People First · 3 min read

The Next Phase of Enterprise Automation Is Unification

Read the essay →
Also from IntelliBuddies

Product and engineering notes

Blog

Latest posts from the team

Platform components such as IntelliTrainer and IntelliFlow, and how generative AI changes RPA and BPA.

Open the blog →
Videos

Demos and tutorials

Product walkthroughs and success stories.

Watch →
Events

Webinars and conferences

Where we speak and exhibit.

See events →

Get new essays by email

New essays from the CEO, delivered to your inbox.

Subscribe
The IB-X Manifesto

Multiply, don't add.

Enterprises do not need more agents; they need work that gets finished, by systems that behave as one, in service of the people who do that work.

  1. Belief 1 of 8

    We believe work should get finished.

    Answers, alerts and dashboards are not outcomes. An outcome is a request understood, a decision routed, a system updated and a record kept. Anything less leaves a person to carry the work across the gap.

    Read the essays on this belief: The Industry Sold Answers. Enterprises Need Outcomes. · Measure Finished Work, Not Activity

  2. Belief 2 of 8

    We believe value multiplies, it does not add.

    Finishing work takes understanding, orchestration and execution at once. Understanding without execution is a conversation that changes nothing. Execution without orchestration is speed without direction. Orchestration with no way in and no hands is a diagram. In a product, if one factor is zero the result is zero.

    Read the essays on this belief: Don't Add Tools. Multiply Outcomes. · Why Pilots Shine and Production Disappoints · The Next Phase of Enterprise Automation Is Unification

  3. Belief 3 of 8

    We believe the seam is where value dies.

    Every handoff between disconnected tools loses context, breaks the audit trail and blurs who is accountable. Fewer seams beat more features.

    Read the essays on this belief: The Seam Is Where Value Dies

  4. Belief 4 of 8

    We believe judgment belongs to AI and repeatability belongs to everything else.

    Use AI where interpretation is needed. Keep what must be repeatable, explainable and auditable deterministic. Intelligence should sit at the edge of a process, not hold the keys to it.

    Read the essays on this belief: An LLM Should Not Hold the Keys to the Ledger

  5. Belief 5 of 8

    We believe authority stays human.

    Agents act. People decide what carries consequences: disputes, write-offs, exceptions of high value. A good system knows where its authority ends and hands over cleanly.

    Read the essays on this belief: The Decisions That Must Stay Human · Autonomy Without Accountability Is a Liability

  6. Belief 6 of 8

    We believe automation exists for people.

    The point is to take tedious, high-volume, low-value work off people so they have time and space for work that matters. Better performance, lower attrition and greater contribution follow from that, not the other way round.

    Read the essays on this belief: What We Owe the People Who Do the Work

  7. Belief 7 of 8

    We believe in starting smart and scaling faster.

    Think big about the outcome, start with one real use case, and grow without starting over. A platform that makes every next step a new integration project has already failed the test.

    Read the essays on this belief: Start With a Conversation, Scale Into a System

  8. Belief 8 of 8

    We believe trust is earned with honest curves.

    Systems that learn from data and from human decisions get better with use. We will say what the first month looks like as readily as the twelfth, and we will not promise perfection.

    Read the essays on this belief: Honest Curves: Why Month One Looks Worse Than Month Twelve

Why we call it X

X is the multiplier.

X is the multiplier: RPA, BPM and AI working as one system at the leading edge of enterprise automation, so the whole is worth more than the sum. It is also the X factor: the quality that is hard to name but easy to recognize when work simply gets finished.

The future is not more agents. It is agents that behave like one, for people who can trust them.

Ravi Krishnamurthy, Managing Director & CEO, IntelliBuddies®

In short

What we stand for, who we work for, and what changes

Stands for Finished work over answers. One system over stitched tools. Human authority over blind autonomy. Honest progress over promised perfection.
Works for The people who do the work, the leaders accountable for it (finance and technology), and the enterprises that need to scale it.
Benefits Time back for meaningful work, clearer accountability, growth without re-platforming, and automation people can trust.
Read the CEO essaysTalk to us about how you would apply this

How we put this into practice: the IB-X platform and the company.

Home / Perspective / From the CEO
The Multiplier

Don't Add Tools. Multiply Outcomes.

By Ravi Krishnamurthy, CEO, IntelliBuddies® and VIT Infotech4 min read

Walk into almost any large company today and you will find an AI initiative. Ask what work it has finished, and the honest answer is often a demonstration. The assistant answers questions well. The automation clicks through the screens well. The process diagram looks good on the wall. Yet the invoice still waits, the order still stalls, and somebody still copies a number from one window into another.

My argument is simple. Enterprises do not get value by adding tools. They get it by multiplying three things, and when any one of them is missing, the result is not smaller. It is zero.

Finished work needs three things at once

The first is understanding: reading a request or a messy document and working out what is being asked. The second is orchestration: deciding who approves, what happens when the numbers do not match, who is told, and how long anyone has to respond. The third is execution: doing the work inside the systems where it lives, including older ones that have no clean interface.

Take a supplier invoice that arrives by email. Someone, or something, has to read it. Someone has to decide who approves it and what happens if the amount differs from the order. Someone has to post it and schedule the payment. Remove any one of those and the invoice does not get paid correctly. The three are not options on a menu. They are factors in the same equation.

UnderstandingRead the request or document
×
OrchestrationWho approves, who is told, how long
×
ExecutionDo the work in the systems where it lives
=
Finished work
Understanding × orchestration × execution = finished work. If any factor is zero, so is the result.

A product punishes a missing part

If value added up, a missing part would cost you a little and the rest would carry on. It does not work that way. Picture an assistant that can answer every question about the invoice but cannot start the approval. Picture a script that posts entries flawlessly but cannot be started reliably or told when to stop. Picture a process map that is perfectly drawn and connected to nothing. Each of these looks impressive in a demonstration. None of them finishes the work.

In a sum, a missing part costs you something. In a product, it costs you everything. That is why a company can own several good tools and still see little finished work. Strength in two factors does not make up for zero in the third.

If value added up

  • 3 + 3 + 0 = 6
  • The missing part costs you a little, and the rest carries on

If value multiplies

  • 3 × 3 × 0 = 0
  • The missing part costs you everything
Illustrative arithmetic: one part missing costs a little in a sum and everything in a product.

The seam is where value dies

Most organizations do not lack the three factors. They own them separately, bought at different times from different places, and joined by hand. The joins between them, the seams, are where the trouble starts.

When something goes wrong with that invoice, who owns the exception? The conversation tool does not know where the process stands. The process tool does not know what the automation did. The automation does not know why it was asked. So a person copies the context across, and the record of who decided what is scattered over three systems. Every seam between tools loses context, weakens the audit trail and blurs accountability. Adding another tool adds another seam.

Joined by hand
Conversation tool
→
A person copies the context
→
Process tool
→
A person copies the context
→
Automation
Connected as one
Understanding
Orchestration
Execution
One record · one set of controlsContext, state and ownership travel with the work
Joined by hand, every handoff is a seam. Connected as one, the record and the controls are shared.

Why this keeps happening

Nobody sets out to build seams. Each purchase makes sense on its own. One team needs to answer customers faster, so it buys a conversational tool. Another wants to stop re-keying data, so it buys automation. A third needs approvals tracked, so it buys a process tool. Every decision is sensible and every demonstration works. The cost appears later, in the gaps, and nobody owns the gaps because nobody bought them. Measuring each tool by its own results hides the problem, because every one of them can be succeeding while the outcome still fails.

The strongest objection

The best answer to all this is a good one: pick the strongest tool for each layer and integrate them. That can be the right call. When the layers change at different speeds, or when a team has real integration skill, a specialist can be better at its part than anything built to work as a whole.

But the question I would ask is not which part is best. It is who owns the finished outcome, and what each handoff costs. Every integration has to be maintained whenever either side changes, and the single record that an auditor wants has to be built by the customer. If you choose to assemble, count the seams as part of the price, and name one person who is accountable for the whole result.

The principle

Value multiplies. It does not add. Judge any automation not by how clever its parts are, but by whether understanding, orchestration and execution are connected strongly enough that work actually finishes.

That changes what we should measure. Counting questions answered, tasks automated or workflows drawn describes activity. The number that matters is how many pieces of work finished correctly, with a record of who decided what.

What this means for you

If you lead finance

Ask who owns a finished outcome from request to result, and whether you can see one trail that shows it.

If you lead technology

Count the seams in a process before you count the features, and ask what happens to context at each handoff.

If you invest in or partner with this market

Expect it to move from collections of tools toward systems whose parts were built to work as one, and judge companies by how they handle the seams.

I hold this belief as part of what we stand for at IntelliBuddies.

Read the principle: Value multiplies, it does not add →

If you disagree, tell me where. Send a reply

RK
Ravi Krishnamurthy

CEO, IntelliBuddies® and VIT Infotech. More than three decades in software. Writes about how enterprises can get work finished, with people in charge of the decisions that matter. Read the manifesto.

Next essay

The Seam Is Where Value Dies

Get the next essay by email

New essays from the CEO, delivered to your inbox.

Subscribe
Home / Perspective / From the CEO
The Multiplier

The Seam Is Where Value Dies

By Ravi Krishnamurthy, CEO, IntelliBuddies® and VIT Infotech5 min read

Ask a technology team what keeps it busy, and a large part of the answer is often keeping tools talking to each other. A process that looks clean on a diagram can fail at the exact point where one system hands the work to the next. Nobody notices the handoff until something goes wrong, and then nobody can say where the work went.

My advice to technology leaders is simple: count the seams before you count the features. A seam is the point where work passes from one tool, system or team to another. It is where context is lost, the record breaks and nobody owns the outcome.

What actually crosses a seam

When work passes between tools, data usually crosses. That is the part integrations are built for. Three other things often do not. The first is context: why the request was made and what has already been tried. The second is state: where the work stands right now. The third is ownership: who is accountable if it goes wrong.

Picture a customer disputing an invoice. The conversation starts in a chat tool, becomes a ticket, is checked in a spreadsheet and ends as an adjustment in the finance system. The amount travels the whole way. The rest does not. The reason for the dispute and the promise already made to the customer stay behind in the chat, so context is lost. Nobody downstream can tell whether the spreadsheet check is finished or still waiting, so state is lost. And when finance questions the adjustment, nobody knows who owns the answer, so ownership is lost.

Usually crosses

  • Data: the amount, the fields, the document

Often stays behind

  • Context: why the request was made, what was tried
  • State: where the work stands right now
  • Ownership: who is accountable if it goes wrong
Data usually crosses a seam. Context, state and ownership often do not.

Seams cost you in three ways

The first is rework. When context does not travel, someone asks the customer again, re-keys the data or searches old messages to piece the story together.

The second is the record. If an auditor asks who approved an exception and why, the answer is spread across several systems. Someone rebuilds it from email after the fact, and the rebuilt version is only as good as their memory.

The third is ownership. A problem that arises at a seam belongs to the system on neither side of it. The chat team says the ticket is wrong, the finance team says the request was incomplete, and the exception waits.

Why seams stay hidden

Seams rarely show up in a demonstration. Each tool works well on its own, and the architecture diagram shows tidy arrows between boxes. The arrows hide what a handoff really involves: a mapping that someone maintains, a failure that someone has to notice, and a person who knows why a field is formatted a certain way. Seams also have no budget line or owner, because each tool has a team and the space between them has none. So they surface as symptoms: delays, rework, audit findings, and a vague sense that the process is slower than its tools should make it.

Seams grow faster than tools

There is also arithmetic. Three tools that might need to talk to each other have up to three connections between them. Five tools have up to ten. Eight have up to twenty-eight. Not every pair needs a link, but each link you build is something to design, monitor, secure and repair whenever either side changes. A new tool can be easy to buy and costly to live with, because the cost arrives as connections.

3 tools3
5 tools10
8 tools28
Up to how many connections tools may need: 3 tools, up to 3; 5 tools, up to 10; 8 tools, up to 28.

The strongest objection

The best objection is that modern integration platforms and standard interfaces have made seams cheap. There is a lot in that. Good integration moves data reliably and removes a great deal of manual work, and I would not argue against using it.

But moving data is the easy half of the problem. A reliable connector does not decide who owns an exception, does not carry the reason a request was made, and does not create the single record an auditor asks for. Those need design, not plumbing. So where you cannot remove a seam, give it an owner. Write down what crosses it and what does not, and name who handles an exception that happens in-between.

The principle

Fewer seams beat more features. At each handoff in a process, ask three questions. What gets lost here? Who owns an exception at this point? Where is the record of what happened? A product with impressive features and many seams will often finish less work than a plainer one whose parts hand over cleanly.

A simple test: pick one outcome, such as paying a supplier correctly, and follow it from request to result. Count the tools it touches and the points where a person copies something between them. That count is a fair first measure of your seams.

This matters more as agents take on more of the work. An agent that acts across several tools inherits every seam between them, and it follows from the idea in my last article: value multiplies, and a seam is where one factor can drop toward zero.

What this means for you

If you lead finance

Ask for one record per outcome, from request to result, and notice how many systems you have to open to reconstruct a single decision.

If you lead technology

Run a seam audit on one process. List every handoff and answer the three questions for each. The longest list of unanswered questions shows where to start.

If you invest in or partner with this market

Look at how a company treats the seams between its own parts. Whether they were designed together says more than the length of the feature list.

I hold this belief as part of what we stand for at IntelliBuddies.

Read the principle: The seam is where value dies →

If you disagree, tell me where. Send a reply

RK
Ravi Krishnamurthy

CEO, IntelliBuddies® and VIT Infotech. More than three decades in software. Writes about how enterprises can get work finished, with people in charge of the decisions that matter. Read the manifesto.

Previous essay

Don't Add Tools. Multiply Outcomes.

Next essay

Why Pilots Shine and Production Disappoints

Get the next essay by email

New essays from the CEO, delivered to your inbox.

Subscribe
Home / Perspective / From the CEO
The Multiplier

Why Pilots Shine and Production Disappoints

By Ravi Krishnamurthy, CEO, IntelliBuddies® and VIT Infotech4 min read

The pilot review goes well. The demonstration works, the sponsor is pleased, and the project is approved to scale. A few months later the same system is struggling, the team is working longer hours than before, and someone is asking what changed.

What changed is that the pilot was quietly supplying something the system could not: people. A pilot often works because missing factors are filled in by hand, and production takes that help away. My advice is to judge a pilot not by its demonstration, but by what it will need when nobody is watching.

A pilot runs on borrowed support

A pilot is usually run with the best people, close attention and a friendly scope. The cases are chosen, the data is cleaner than normal, the builders are in the room, and any problem is fixed the same day. All of that is reasonable. It is also support that production does not have.

Take an invoice pilot. It runs on one supplier's invoices in one format. When something looks odd, a builder notices and corrects it before anyone is affected. The pilot appears to understand, orchestrate and execute. In truth, people are doing part of each, out of sight.

Nobody is being dishonest. The builders want the pilot to succeed, and fixing a problem quietly is what good engineers do. But the cost of that help appears nowhere in the results.

UnderstandingThe system does it
OrchestrationDone by people, out of sight
ExecutionDone by people, out of sight
In a pilot the system understands. People quietly supply orchestration and execution, out of sight.

Production brings the cases the pilot never saw

Production means every supplier and every format. It means the invoice with no order number, the partial delivery, the approver on leave and the screen that changed after an update. Each is an exception, and each needs an owner and a path. In a pilot they were rare enough to handle by hand. At full volume they are the work.

The exceptions were not absent from the pilot. They were rare enough that someone handled them quietly, and a person who handles an exception quietly does not appear in the pilot's results.

A pilot runs on

  • The best people and close attention
  • One supplier, one format, cleaner data
  • Builders in the room
  • Problems fixed the same day

Production must supply

  • Every supplier and every format
  • An owner and a path for each exception
  • Cover when the approver is on leave
  • Resilience when a screen changes
  • Operation without the builders
What a pilot borrows, and what production has to provide on its own.

Demonstrations reward understanding

Reading a document or answering a question is the part that demonstrates well, so it gets the attention. That is not a criticism of understanding, because nothing starts without it. But it is the factor a demonstration shows best and production tests least. Production is decided by the less visible factors. Who is told when something is wrong? What happens when the approver does not respond? How does the process keep running when a system changes? Where does the record live? A pilot that shows only understanding has shown one of the three factors, and the other two are where production succeeds or fails.

What gets measured

Pilots are often reported on items handled, time saved per item or the accuracy of the reading step. Each of those is measured on the cases the pilot happened to see. None tells you how many pieces of work finished correctly without help, or what the exceptions cost. A system can post an impressive handling figure and still leave a large share of the real work to people.

The strongest objection

The obvious objection is that a pilot is meant to be small. That is right, and I am not arguing against starting small. A narrow start is how you learn cheaply. The mistake is not the size of the pilot. It is leaving out what production will need.

So make the small thing representative. Include one real exception and a named owner for it. Run it on the foundation you intend to scale on, so you do not have to start again. And decide before you begin how you will measure finished work, not how many items the pilot handled.

A second objection is that production problems can be fixed after launch. Some can, and iterating is healthy. But a hidden factor is not fixed in an afternoon. The owner, the exception path and the record all have to be designed, and by then the pilot has set expectations that production cannot meet.

The principle

Value multiplies, and a pilot can hide a factor at zero by filling it in with people. Pilots are worth running, but a good one is a rehearsal of production, not a performance for the sponsor. Before you scale, ask four questions. Who is quietly doing work the pilot does not show? What happens on the first exception, and who owns it? Could this run next month without its builders? Are we measuring finished work, or items handled? If the answers are unclear, the pilot has proved the demonstration, not the process.

What this means for you

If you lead finance

Ask what the pilot's results exclude. Count the hours people spent supporting it, and ask who owns the outcome once the builders step back. A pilot's savings are only real if they survive that count.

If you lead technology

Make the pilot representative: one real exception, a real owner, and the platform you intend to scale on.

If you invest in or partner with this market

Ask how a company's customers move from pilot to production, and what changes when they do. The answer shows whether the product finishes work or only demonstrates it.

I hold this belief as part of what we stand for at IntelliBuddies.

Read the principle: Value multiplies, it does not add →

If you disagree, tell me where. Send a reply

RK
Ravi Krishnamurthy

CEO, IntelliBuddies® and VIT Infotech. More than three decades in software. Writes about how enterprises can get work finished, with people in charge of the decisions that matter. Read the manifesto.

Previous essay

The Seam Is Where Value Dies

Next essay

The Industry Sold Answers. Enterprises Need Outcomes.

Get the next essay by email

New essays from the CEO, delivered to your inbox.

Subscribe
Home / Perspective / From the CEO
Finished Work

The Industry Sold Answers. Enterprises Need Outcomes.

By Ravi Krishnamurthy, CEO, IntelliBuddies® and VIT Infotech4 min read

Ask a buyer what they hoped to get from their last automation purchase, and the honest reply is often finished work, not better answers. Yet for the past few years much of the technology industry has competed on how well its tools answer. Better answers, faster answers, answers in plain language. Answers are easy to show and easy to like. They are rarely what an enterprise was short of.

My argument is that an answer is not an outcome. An answer tells you something. An outcome changes something: an invoice paid, an order shipped, a record corrected. Enterprises should buy, build and measure for outcomes.

The difference between an answer and an outcome

A customer asks where their refund is. An assistant can look up the status and reply. That is an answer, and it is useful. An outcome is different. The request is understood, approved if it needs approval, the refund is issued, the ledger is updated, the customer is told, and a record shows who decided what.

The two begin with the same question and end in very different places. An answer stops at the edge of the conversation. An outcome carries on through approvals and into the systems where the work is actually done.

An answer
Question
→
Status looked up
→
Reply sent
An outcome
Request understood
→
Approved if needed
→
Refund issued
→
Ledger updated
→
Customer told
→
Record kept
The same question, "Where is my refund?", ends in very different places.

Why the industry sold answers

Answers demonstrate well. A conversation interface can be shown in minutes, it feels intelligent, and it needs no change to the systems underneath. Outcomes are harder. They require routing, approvals, exceptions, and actions inside systems that were never designed to be automated. They are slower to build and harder to show on a stage.

Answers are also easy to count. Questions answered, conversations handled and tasks completed all rise quickly. Buyers tend to buy what they can see and vendors tend to sell what they can show, so the market drifted toward the part that demonstrates best.

None of this makes the answer-first approach foolish. It was the part of the problem that technology could solve quickly, and it opened the door. The risk is stopping there.

What an outcome requires

Turning an answer into an outcome takes the three factors from my first article working together. Something must understand the request, something must route and govern it, and something must act in the systems where the work lives. If any one is missing, the customer gets an answer and a person still has to do the rest.

It also takes people, in the right places. Some outcomes need a decision that belongs to a person, such as a refund above an agreed limit. A good system finishes the routine cases itself, hands the larger ones to a person with the full case in front of them, and carries on once that person has decided.

Within the agreed limitThe system finishes the case and records it
Above the limitA person decides with the full case in front of them
After the decisionThe work resumes without anyone restarting it
The system finishes routine cases; larger ones go to a person with the full case, and work resumes afterwards.

How to buy for outcomes

In a demonstration, ask to see one request go all the way through. Not the conversation alone, but the request arriving, the approval, the action in the system, and the record afterwards. Then ask three things. What happens when the request is unusual? Where in the flow does a person decide, and what do they see when they do? And how would we know, a month later, that the work finished correctly? A vendor who built for outcomes will answer these quickly. One who built for answers will return to the conversation.

The strongest objection

The obvious objection is that answers are valuable, and that is true. For many tasks, an answer is the outcome. If an employee wants to know a policy, a good answer is the end of the job, and a knowledge assistant is a sensible and inexpensive way to provide it.

The difference matters when the question is really a request. "Where is my refund?" may be asking for information, but "Please refund this" is asking for something to happen. So for each use case, ask whether the goal is that someone knows something or that something is done, and buy accordingly. Treating the second kind as if it were the first is how an organization ends up with a fluent assistant and an unchanged process.

The principle

Work should get finished. Judge any automation by whether the thing the person asked for actually happened. After the system responds, ask what has changed in the world. If the answer is "the person now knows something", you have bought an answer. If it is "the refund is issued and the record is kept", you have bought an outcome.

A useful habit is to write the outcome as a sentence before choosing a tool, such as "every refund request is closed correctly, with a record." If you cannot write it, you are not yet ready to buy.

This matters more as agents take on more of the work. An agent that talks well but cannot act simply moves the work to a person's inbox in a more polite form.

What this means for you

If you lead finance

Ask what each automation closes, pays or collects, not how many questions it answers. The outcome is where the money moves.

If you lead technology

Name the outcome and the systems it must change before you choose a tool. A tool chosen for its conversation may have no way into those systems.

If you invest in or partner with this market

Expect the weight to shift from assistants that talk to systems that finish, and look for the share of requests that end with the thing done.

I hold this belief as part of what we stand for at IntelliBuddies.

Read the principle: Work should get finished →

If you disagree, tell me where. Send a reply

RK
Ravi Krishnamurthy

CEO, IntelliBuddies® and VIT Infotech. More than three decades in software. Writes about how enterprises can get work finished, with people in charge of the decisions that matter. Read the manifesto.

Previous essay

Why Pilots Shine and Production Disappoints

Next essay

Measure Finished Work, Not Activity

Get the next essay by email

New essays from the CEO, delivered to your inbox.

Subscribe
Home / Perspective / From the CEO
Finished Work

Measure Finished Work, Not Activity

By Ravi Krishnamurthy, CEO, IntelliBuddies® and VIT Infotech4 min read

Open almost any automation dashboard and the numbers are going up. Tasks automated, questions answered, workflows built, hours saved. A finance leader looks at the screen and asks a fair question: where did the money go? The numbers measure motion. They do not tell you how much work was finished.

My argument is that the unit to count is a piece of work that finished correctly, without help, and with a record. Everything else is activity.

Activity is easy to count and easy to inflate

Tasks automated, questions answered and workflows built are all simple to count, and all can rise while the amount of finished work stays flat. Imagine a dashboard that shows every invoice the system has touched. The figure is impressive. But if a large part of those invoices come back as exceptions that people repair, the team may be spending as many hours as before, and the dashboard still shows progress.

Hours saved is a harder case. It is a reasonable idea, but it is usually estimated per task, on the assumption that everything else about the process stays the same. Real processes do not hold still. The saving on one step can reappear as rework on the next.

What dashboards often count

  • Tasks automated
  • Questions answered
  • Workflows built
  • Estimated hours saved

What finance should count

  • Finished correctly
  • Finished without help
  • Time from request to result
  • Exceptions, and who owned them
Activity can rise while finished work stays flat.

Four measures of finished work

The first is finished correctly: the request reached its result, whether that is a payment made, an order shipped or a record fixed, and someone with authority agrees it is right.

The second is finished without help: no person had to repair, redo or chase the work for it to count. This separates a system that finishes work from one that finishes work with people in the background.

The third is time from request to result. Measure it across the whole process, not per step, because a quick step inside a slow process changes nothing the customer or the business can feel.

The fourth is exceptions and who owned them: how many went to people, how long they waited, and who was responsible. A rising number here is not a failure on its own. It tells you where to improve, which is the point.

Behind all four sits the record. If you cannot show who decided what, the work is not finished in a way an auditor would accept.

From activity to cost

Finance can turn these measures into one figure: the cost per finished outcome. Take the cost of the system, add the people time spent on exceptions and support, and divide by the pieces of work that finished correctly. The number is honest because the people time is inside it, including the time that was easy to leave out in a pilot. It also lets you compare a pilot, a manual process and a live system on the same basis.

System cost
+
People time on exceptions and support
÷
Pieces of work finished correctly
=
Cost per finished outcome
Cost per finished outcome puts people time inside the number.

How to measure it well

Define "correctly" with the process owner before you start, and write it down. Measure on live work, not on a pilot, because a pilot hides the help it receives. And have someone review a sample of finished work, so the count does not rest on the system's own judgment of itself.

What you will find when you start

The first measurement is usually humbling. The share of work that finished without help is lower than the handled count suggested, and the gap is the work people were doing quietly. That is not a reason to stop. It is the most useful figure you will get, because it shows where the exceptions come from. Fix the most common cause and the number moves.

Who owns the number

Someone must own the measure, and it should not be the team that built the system. It sits best with the process owner, the person accountable for the outcome, such as the head of accounts payable, who has every reason to want the figure to be true.

The strongest objection

The strongest objection is practical. Hours saved fits neatly into a business case, volumes are easy to report, and finished work takes effort to define and measure. That is fair. A business case needs a number, and a number that is hard to produce will lose to one that is easy.

My answer is to produce the harder number once, then reuse it. Cost per finished outcome is a financial measure. It can sit in the same business case as hours saved, and it will tell you more, because it still holds when the process changes around it. It also tells you what to do next, which a volume figure rarely does. There is a quieter benefit as well. When you define "correctly" with the process owner, you often find the process had no clear definition before. That is a finding in itself.

The principle

Work should get finished, so count it. Ask four questions of any automation. How many pieces of work finished correctly? How many finished without help? How long did it take from request to result? Who owned the exceptions?

What this means for you

If you lead finance

Replace the activity dashboard with cost per finished outcome, and insist that people time is part of it.

If you lead technology

Record the outcome in the process itself, not separately in each tool, so the measures do not depend on stitching reports together.

If you invest in or partner with this market

Ask a vendor for finished-work measures from live use, with the evidence behind them, and treat volume figures on their own with care.

I hold this belief as part of what we stand for at IntelliBuddies.

Read the principle: Work should get finished →

If you disagree, tell me where. Send a reply

RK
Ravi Krishnamurthy

CEO, IntelliBuddies® and VIT Infotech. More than three decades in software. Writes about how enterprises can get work finished, with people in charge of the decisions that matter. Read the manifesto.

Previous essay

The Industry Sold Answers. Enterprises Need Outcomes.

Next essay

Start With a Conversation, Scale Into a System

Get the next essay by email

New essays from the CEO, delivered to your inbox.

Subscribe
Home / Perspective / From the CEO
Finished Work

Start With a Conversation, Scale Into a System

By Ravi Krishnamurthy, CEO, IntelliBuddies® and VIT Infotech4 min read

Many automation programs begin with a conversation. A team builds an assistant that answers questions, because it is the easiest thing to build and the easiest thing to show. Then someone asks it to do more. Please approve this. Please update that. Please post this entry. The team finds that each request needs something the assistant was not built to do, and each addition becomes a new project. Nobody did anything wrong. The first step was sensible.

My argument is that starting small is right, and that the question to ask is whether the first step leaves a path to the whole or a dead end that must later be replaced.

Why a conversation is a good place to start

A conversation is the front door. It has the lowest barrier to entry, people can try it without training, and it shows you what they actually ask for. Picture an internal helpdesk assistant that answers policy questions. Within weeks you learn which questions are really requests for action: reset my access, approve my leave, update my details. That learning is valuable and cheap, and it tells you where to go next.

It also builds confidence. People see something working, and that makes it easier to ask for the next step. And it is a low-risk way to learn which of your data and systems are in good enough shape to automate.

Where it gets stuck

Requests for action need approvals and execution. If the first tool cannot extend to those, every step forward means a new tool and a new connection. You add a workflow tool, then an automation tool, and with each one you add seams, of the kind I described earlier, where context is lost and ownership is unclear. Some of what you built in the first step must also be rebuilt, because the new tools do not share its logic. At that point the team is choosing between two poor options: stretch the first tool beyond what it was built for, or start again.

The cost is not only money. It is time, and the momentum of a program that was going well.

Test the path, not the first step

When you choose a starting point, test three things. First, can the conversation hand work to a workflow without custom glue? Second, can the workflow act in the older systems where the work really happens? Third, do one record and one set of controls cover all three?

You can run these tests on day one, before anything is built. Imagine the later steps: first the conversation, then an approval step, then posting the entry in a system that has no modern interface. If each step adds capability to the same foundation, the path exists. If each one needs something new bolted on, it does not.

3 · Post the entry in a system with no modern interfaceExecution added
2 · Add an approval stepOrchestration added
1 · Start with the conversationUnderstanding, the front door
One foundationSame record · same controls · same connections
Each step adds capability to the same foundation, so the path exists.

What a good first use case looks like

A good first use case has a real owner, real volume, a real exception and a visible result. It is small enough to finish and large enough to matter. It should ask the conversation to hand work onward, even if only one step, so the handoff is tested early and not discovered later.

Why teams skip the path

Teams skip this because the first step is urgent and the later steps feel distant. There is a sponsor to please and a demonstration to give, and planning the path feels like delay. In practice it is a small amount of design that can save a great deal of rework.

The strongest objection

A fair objection is that technology changes quickly, and that committing to a foundation risks lock-in. A lightweight tool is cheap, replacing it later is acceptable, and a mix of specialist tools may serve better than one platform. All of that can be true.

But the cost to count is not only the price of the first tool. It is the seams you add at each step and the work you rebuild. And lock-in is something you can test for. Ask whether your data and logic can move if you ever need to leave, and whether each step adds seams or removes them. A foundation that is open to exit and closed to seams is a better bet than either extreme.

The principle

Think big, start smart, scale faster. Think big about the finished outcome you want. Start with one real use case, including a real exception and a named owner. Then scale by adding capability on the same foundation, so each step makes the next one easier.

Scaling faster does not mean doing everything at once. It means each new use case starts from what already exists: the same record, the same controls, the same connections. Speed comes from reuse. This is the opposite of a pilot built to impress. A pilot designed this way is the first stage of the real thing.

What this means for you

If you lead finance

Fund the path in stages, and tie each stage to finished work, so each decision to continue rests on evidence.

If you lead technology

Design the path before the first step. Sketch the three stages and test the handoffs between them at the start.

If you invest in or partner with this market

Look at how a customer's second and third use cases arrive. If they require a new product each time, the platform is not doing its job.

I hold this belief as part of what we stand for at IntelliBuddies.

Read the principle: Start smart, scale faster →

If you disagree, tell me where. Send a reply

RK
Ravi Krishnamurthy

CEO, IntelliBuddies® and VIT Infotech. More than three decades in software. Writes about how enterprises can get work finished, with people in charge of the decisions that matter. Read the manifesto.

Previous essay

Measure Finished Work, Not Activity

Next essay

An LLM Should Not Hold the Keys to the Ledger

Get the next essay by email

New essays from the CEO, delivered to your inbox.

Subscribe
Home / Perspective / From the CEO
Authority

An LLM Should Not Hold the Keys to the Ledger

By Ravi Krishnamurthy, CEO, IntelliBuddies® and VIT Infotech4 min read

There is a temptation, when a technology is as capable as large language models, to hand it the whole job. It reads the invoice, picks the account, approves the amount and posts the entry. In a demonstration it looks like the future. In a ledger it is a risk that nobody has priced.

My argument is that intelligence belongs at the edge of a process, not at its core. Use AI where interpretation is needed, and keep anything that must be repeatable, explainable and auditable deterministic.

What each is good at

Language models are good at judgment of a particular kind. They can read a messy document, work out what a customer is asking, classify a request, summarize a long case and draft a reply. These tasks have no single correct answer written down in advance, and a good-enough interpretation is valuable.

Ledgers, payments and approvals are different. Posting an entry, applying a policy limit, calculating an amount and releasing a payment all have a right answer that must come out the same way every time the same facts appear. Here the value is not in cleverness. It is in repeatability.

Where AI helps: interpretation

  • Read a messy document
  • Work out what a customer is asking
  • Classify a request
  • Summarize a long case
  • Draft a reply

Where repeatability matters

  • Post an entry
  • Apply a policy limit
  • Calculate an amount
  • Release a payment
Judgment for AI, repeatability for the rest.

Why mixing them without a boundary is risky

A language model is probabilistic. Given the same input it can give a different answer on a different day, and when it is wrong it is often wrong in a confident and plausible way. That is acceptable when a person reads a draft. It is not acceptable when the output moves money.

Three problems follow. The first is variability: the same invoice can be treated differently from one run to the next. The second is explanation: when an auditor asks why an entry was posted, "the model decided" is not an answer. The third is the asymmetry of errors. A mistaken summary costs a moment. A mistaken payment costs money, and may be hard to reverse.

A concrete boundary

Consider coding an invoice. A model reads the document, extracts the supplier and the amounts, and proposes an account. Deterministic rules then check the proposal against the chart of accounts, the supplier record and the approval limits. If everything agrees, a standard step posts the entry. If anything does not, the case goes to a person with the proposal and the reason it was stopped. Every step is recorded: the input, the proposal, the rule applied and who approved.

The model did the part that needs interpretation. The rules did the part that needs to be right every time. Neither could replace the other.

Invoice coding
A model reads the invoiceExtracts supplier and amounts, proposes an account
→
Rules check the proposalChart of accounts, supplier record, approval limits
↓
Everything agreesA standard step posts the entry
Anything does notA person sees the proposal and the reason it was stopped
Every step recordedInput · proposal · rule applied · who approved
A model proposes; deterministic rules check; a person answers for exceptions; every step is recorded.

The strongest objection

The objection is that this wastes what the technology can do. Models are improving quickly, rigid rules are brittle when inputs are messy, and an agent that acts end to end is more useful than one that keeps stopping to ask. There is truth in all of it. Capability is rising, and rules alone cannot read a crumpled scan.

But reliability on average is not the same as repeatability on a given case. A system that is right nearly every time still needs a way to catch the occasional miss before it reaches the ledger, and accountability cannot be delegated to a model. The boundary is also not fixed. It can move as evidence builds. Let the system act on one category of cases, measure the results, and extend only when the record supports it. Autonomy should be earned, case by case.

The principle

Judgment for AI, repeatability for the rest. Put judgment where interpretation is needed, at the front of the process to understand and at the back to explain. Keep the decision to post, pay or approve inside rules that behave the same way every time, with a person answering for the exceptions. Ask of any proposal to use AI: is this a task with a right answer that must repeat, or one that needs interpretation? Then place it accordingly.

This matters more as agents take on more of the work. An agent that can act needs a boundary around what it may decide, and the boundary should be set by what the task needs, not by what the model can do.

What this means for you

If you lead finance

Ask where in each process a model decides and where rules decide, and insist on a record of each.

If you lead technology

Draw the boundary in the design. Treat the model's output as a proposal to be checked, not as an instruction to be followed.

If you invest in or partner with this market

Look for companies that can say precisely where AI is used in their product and where it is deliberately not.

I hold this belief as part of what we stand for at IntelliBuddies.

Read the principle: Judgment for AI, repeatability for the rest →

If you disagree, tell me where. Send a reply

RK
Ravi Krishnamurthy

CEO, IntelliBuddies® and VIT Infotech. More than three decades in software. Writes about how enterprises can get work finished, with people in charge of the decisions that matter. Read the manifesto.

Previous essay

Start With a Conversation, Scale Into a System

Next essay

The Decisions That Must Stay Human

Get the next essay by email

New essays from the CEO, delivered to your inbox.

Subscribe
Home / Perspective / From the CEO
Authority

The Decisions That Must Stay Human

By Ravi Krishnamurthy, CEO, IntelliBuddies® and VIT Infotech3 min read

Every automation program eventually reaches the same question: what do we let the system decide? Answer too boldly and you create risk. Answer too cautiously and you keep people doing work a system could do. The useful move is to stop asking it in general and ask it decision by decision.

My argument is that some decisions carry consequences that must be answered for by a person, and that good automation takes everything else away from people so they can make these decisions well.

A ladder of decisions

Decisions are not all alike. At the bottom are routine ones with a clear rule, such as matching a payment that agrees exactly with an invoice. A system can decide these and record that it did. Above them sit decisions that need review, such as a match with a small difference, where the system can propose and a person confirms. At the top are the decisions that stay with people: writing off a difference, or raising a dispute with a payment provider.

Stays with peopleWrite off a difference. Raise a dispute with a payment provider.
Needs reviewA match with a small difference: the system proposes, a person confirms.
RoutineA payment that agrees exactly with an invoice: the system decides and records it.
Some decisions the system makes, some it proposes, and some stay with people.

What puts a decision at the top

A few tests help. Is the consequence hard to reverse? Does it spend money or create an obligation above an agreed limit? Does it involve fairness or a judgment about people? Are the rules unclear or in conflict? And would someone have to answer for it to an auditor, a regulator or a board? Any one of these is a reason to keep a person in charge. Several together make it close to certain.

Why accountability cannot be delegated

A system cannot be accountable. It cannot explain itself to a regulator, take responsibility in front of a board or be held to a standard of care. When something goes wrong, a person answers, whether or not they made the decision. So the question is not whether people remain accountable. They do. The question is whether they decide with the full picture, or inherit the outcome of a choice they never saw.

Design the handoff

The quality of a human decision depends on what the person is shown. A bare alert saying that approval is needed forces them to dig through several systems. A good handoff gives the full case in plain words, what the system suggests and why, the options and what each would do, and a place to record the reason. Once the person decides, the work resumes without anyone restarting it. The system does the preparation. The person keeps the decision.

Decision needed
The case, in plain wordsWhat happened and why it was stopped
What the system suggests, and whyA proposed action with its reason
The options, and what each would doConsequence of every choice
Your reasonRecorded with the decision
ApproveChoose another optionRefer on
The system does the preparation. The person keeps the decision.

Setting the lines

Set the lines by value, reversibility and risk, write them down, and start cautiously. Then review them as evidence accumulates. If the system's proposals for a category of cases are confirmed again and again, that is a reason to consider moving the category down a step. If a decision keeps being overturned, that is a reason to keep it where it is.

The strongest objection

The common objection is that human approvals are a bottleneck, and that automation should remove them. Approvals can indeed slow work, and long queues are a real cost. But the answer is to make them better, not to remove them. Clear the routine cases so that only the real decisions reach a person. Give them the full case so each takes minutes, not hours. Measure how long decisions wait, and move the thresholds as the evidence allows. A person who decides fewer things, with better information, is a stronger control than a system that decides everything without anyone looking.

The principle

Authority stays human. Agents act, and people decide what carries consequences. A good system knows where its authority ends and hands over cleanly.

What this means for you

If you lead finance

List the decisions in a process that carry the most consequence, and ask who makes each one today and what they are shown when they do.

If you lead technology

Build the handoff as a first-class step of the process, not as an email that interrupts it.

If you invest in or partner with this market

Ask what the system will refuse to decide. A clear answer shows that someone has thought about where authority ends.

I hold this belief as part of what we stand for at IntelliBuddies.

Read the principle: Authority stays human →

If you disagree, tell me where. Send a reply

RK
Ravi Krishnamurthy

CEO, IntelliBuddies® and VIT Infotech. More than three decades in software. Writes about how enterprises can get work finished, with people in charge of the decisions that matter. Read the manifesto.

Previous essay

An LLM Should Not Hold the Keys to the Ledger

Next essay

Autonomy Without Accountability Is a Liability

Get the next essay by email

New essays from the CEO, delivered to your inbox.

Subscribe
Home / Perspective / From the CEO
Authority

Autonomy Without Accountability Is a Liability

By Ravi Krishnamurthy, CEO, IntelliBuddies® and VIT Infotech4 min read

Autonomy is the word of the moment. Vendors promise agents that run whole processes without anyone touching them, and the benefits can be real. But autonomy is a transfer of authority, and authority without accountability is not an asset. It is a liability that arrives gradually and is noticed late.

My argument is that autonomy is worth having only when someone remains answerable and every action leaves a record. Enterprises, and the investors who back the companies serving them, should treat governance as a condition of autonomy, not as a cost of it.

Where autonomy creates risk

An autonomous agent acts at speed and at scale. Those are its strengths, and they are also how it does harm. A mistake that a person would make once, a system can make many times before anyone notices. Some actions are hard to undo: a payment sent, a message delivered to customers, a record overwritten. And if nobody owns the agent's decisions, the first sign of trouble is often a customer or an auditor, not a dashboard.

Picture an agent that approves a type of refund by a rule that is slightly wrong. The cost is not one bad refund. It is every refund of that type until the problem is found.

What accountability needs

Five things. A named owner who answers for what the agent does. Clear limits on what it may do on its own. A record of each action and the reason for it. A way to stop it quickly. And a regular review of how it is performing. None of these slows a good agent down much. Without them, even a good agent is a risk you cannot manage, because you cannot see what it is doing or change it when you need to.

Autonomy that grows with trust

The way through is governed autonomy. Start with a system that suggests while a person acts. Move to acting after a person approves. Then to acting alone within set limits, with reporting. Extend the scope only when live evidence supports it. Each step needs more in place, not less. This is slower than switching everything on, but the customer can say yes at each step, and that is what lets the system go further.

Suggests

A person acts

  • Named owner
  • Record of each action

Acts after approval

A person approves first

  • Named owner
  • Record of each action
  • Clear limits

Acts alone within limits

With reporting

  • Named owner
  • Record of each action
  • Clear limits
  • A way to stop it quickly
  • Regular review
Illustrative: each step to more autonomy needs more in place, never less.

Why investors and boards should care

As agents take on more of the work, governance becomes something buyers ask about early, especially in finance, healthcare and other regulated settings. A company that can show who is accountable, what the agent may do and how every action is recorded will find it easier to be trusted with more. One that cannot will find its customers keep the agent on a short leash, however capable it is. For an investor, the quality of the governance is part of the product, and part of how far the product can grow inside a customer.

The strongest objection

The objection is about speed. Governance adds work, competitors will ship autonomy faster, and customers want results. Speed matters, and I would not argue for governance that slows everything down. But the speed that matters is how quickly a customer lets the system do more, and that depends on trust. A vendor that can show the record and the limits earns that trust sooner. One that cannot is often held back, by the customer's own risk team if by no one else.

There is also a practical point. Governance built in from the start is cheap. Governance added after an incident is expensive, rushed and public.

The principle

Autonomy is earned. The more a system is allowed to do alone, the more a person must be able to see, limit and stop. Authority stays human even when action does not.

What this means for you

If you lead finance

Before approving any autonomy, ask five questions. Who is accountable? What can the agent do alone? Can I see every action and the reason for it? Can it be stopped? And how are its limits changed?

If you lead technology

Build the record and the stop switch first, not last. They are much harder to add once the system is in use.

If you invest in or partner with this market

Put the same five questions to any company you back, and treat unclear answers as a risk to the business, not only to the customer.

I hold this belief as part of what we stand for at IntelliBuddies.

Read the principle: Authority stays human →

If you disagree, tell me where. Send a reply

RK
Ravi Krishnamurthy

CEO, IntelliBuddies® and VIT Infotech. More than three decades in software. Writes about how enterprises can get work finished, with people in charge of the decisions that matter. Read the manifesto.

Previous essay

The Decisions That Must Stay Human

Next essay

What We Owe the People Who Do the Work

Get the next essay by email

New essays from the CEO, delivered to your inbox.

Subscribe
Home / Perspective / From the CEO
People First

What We Owe the People Who Do the Work

By Ravi Krishnamurthy, CEO, IntelliBuddies® and VIT Infotech3 min read

Every conversation about automation has two audiences, and usually only one is addressed. Leadership asks what it will save. The people who do the work ask what it will do to them. Both questions are fair, and a program that answers only the first tends to find the second answered for it, by rumor.

My argument is that automation exists for people. Its purpose is to take tedious, high-volume, low-value work off them, so they have time and space for work that matters. If that is the purpose, it creates obligations.

Start from the person's work

Ask the people who do a process what they would happily stop doing. The answers are usually specific: re-keying data between systems, chasing missing information, reconciling by hand, copying the same figures into another file. Those are the tasks to take away first, and the people who do them know details that no process map shows. An accounts payable clerk knows which suppliers always send an odd invoice, and why. Involve them in choosing what to automate, and the system will be better for it.

Take off people

  • Re-keying data between systems
  • Chasing missing information
  • Reconciling by hand
  • Copying the same figures into another file

Give people time for

  • Reviewing exceptions
  • Dealing with customers and suppliers
  • Improving the process
  • Analysis and new skills
Take the tedious work away so people have time and space for work that matters.

Be honest about what changes

Some roles will change, and some work will go. Saying so early is kinder and more effective than letting people guess. Explain what will be automated, what will remain and where there will be new work: reviewing exceptions, dealing with customers and suppliers, improving the process. Then back it with training and time to learn it. Silence is not neutral. People fill it with the worst case.

Do not make the person the clean-up crew

One way to fail is to automate the easy cases and leave people with only the hard ones, with no context and no tools. The job becomes a queue of problems with nothing to help solve them. The better design gives them the full case, in plain words, with the suggestion and the options in front of them, and treats them as the experts they are. Their judgment is what the system hands over to, and it should be treated that way.

Give people a say, and a way to grow

People who work with a system every day notice what it gets wrong long before any report does. Give them a channel to say so, and act on it, so that their corrections improve the system and are not lost. Give them a route to new skills too, whether that is analysis, supplier relationships or running the automation itself. A program that makes the system better and the people more capable is one that lasts.

The strongest objection

A skeptic will say that cost reduction is the real reason for most automation, and that talk of people is cover. Cost matters, and I would not pretend otherwise. In some cases the number of people needed will fall.

But the choice that remains is how the benefit is used and how honestly the change is handled. Programs that people trust get adopted faster and used better, because the people who know the exceptions help to find them. Programs that people fear get worked around. The honest approach, saying plainly what changes and what is being done about it, is also the effective one.

The principle

Automation exists for people. Judge a program not only by the work it removes, but by the work it leaves. Is that work more meaningful than before? Do people have the context and the authority to do it well? Did they have a say? These can be measured in the end: the time spent on work that needs a person, and how people say the job has changed.

What this means for you

If you lead finance

Count the time returned to people for work that matters, next to the cost saved. A program that saves cost and leaves the work worse has saved less than it appears.

If you lead technology

Design the exception queue as a job people would want, with the full case and the tools to resolve it.

If you invest in or partner with this market

Ask how a company's customers involve their own staff, because adoption depends on it.

I hold this belief as part of what we stand for at IntelliBuddies.

Read the principle: Automation exists for people →

If you disagree, tell me where. Send a reply

RK
Ravi Krishnamurthy

CEO, IntelliBuddies® and VIT Infotech. More than three decades in software. Writes about how enterprises can get work finished, with people in charge of the decisions that matter. Read the manifesto.

Previous essay

Autonomy Without Accountability Is a Liability

Next essay

Honest Curves: Why Month One Looks Worse Than Month Twelve

Get the next essay by email

New essays from the CEO, delivered to your inbox.

Subscribe
Home / Perspective / From the CEO
People First

Honest Curves: Why Month One Looks Worse Than Month Twelve

By Ravi Krishnamurthy, CEO, IntelliBuddies® and VIT Infotech3 min read

Every vendor demonstration shows a system at its best. Every real deployment shows it at its start. The difference surprises buyers and damages trust, and it is avoidable. A system that learns will look worse in its first month than in its twelfth, and anyone who says otherwise is either describing a system that does not learn or not telling the whole story.

My argument is that trust is earned with honest curves. Ask a vendor for the shape of the improvement, not just the destination, and plan for it.

Why a system starts lower

Three gaps explain most of it. The first is history. A new system has not seen how your business pays, approves or handles exceptions, and much of what makes processing accurate is knowing the patterns of a particular customer or supplier. The second is rare cases. Unusual situations that happen only occasionally have not appeared yet, and when they do, the system has no experience of them. The third is edge cases specific to you: a franchise that uses an unusual account, a customer who sometimes pays without a reference. Nobody could have listed these in advance.

HistoryThe system has not yet seen how your business pays, approves or handles exceptions
Rare casesUnusual situations have not appeared yet
Edge cases specific to youA franchise on an unusual account; a customer who pays without a reference
Three gaps explain most of a lower first month. All of them close with use.

What closes the gaps

Observation. Each time a person resolves an exception, the system records the decision and the pattern behind it. Cases that repeat become rules the system can apply next time. History accumulates. The improvement does not come from tuning settings, or from an expert writing every rule before the first run. It comes from the work itself, which is why it continues after the start.

Month 1 Month 12 More exceptions, more review Patterns learned from human decisions Work finished without help
Illustration of the shape, not measured data.

Plan for the curve

Planning begins with expectations. Agree what the first month should show: the system should already beat the manual baseline, but with more exceptions, and therefore more review time, than later. Budget that review time at the start, knowing it should fall. Track the trajectory, such as the share of work finished without help, month by month. And agree in advance what evidence would let you give the system more to do.

The strongest objection

A buyer might reply that a good system should work from day one, and that a vendor who explains a weak first month is making excuses. There is a fair point inside it. Some value must arrive immediately, and a system that does not beat the manual process from the start should not be bought.

But a promise of perfection should raise a question, not settle one: what happens to the cases it has not seen? The alternative, an expert encoding every rule for you before the first run, can give a strong start. It is slow, it tends to break when things change, and it does not improve by itself.

The principle

Trust is earned with honest curves. A vendor who shows you the first month as honestly as the twelfth is telling you how the system behaves when it meets your business. A curve you can see is one you can manage.

What this means for you

If you lead finance

Ask for the expected first-month result, the review effort it implies and the evidence for improvement, and budget accordingly.

If you lead technology

Build in the feedback loop, so that every human decision is recorded and reused instead of being lost.

If you invest in or partner with this market

Look for companies that can show the shape of the curve from live use, and be wary of day-one perfection.

I hold this belief as part of what we stand for at IntelliBuddies.

Read the principle: Trust is earned with honest curves →

If you disagree, tell me where. Send a reply

RK
Ravi Krishnamurthy

CEO, IntelliBuddies® and VIT Infotech. More than three decades in software. Writes about how enterprises can get work finished, with people in charge of the decisions that matter. Read the manifesto.

Previous essay

What We Owe the People Who Do the Work

Next essay

The Next Phase of Enterprise Automation Is Unification

Get the next essay by email

New essays from the CEO, delivered to your inbox.

Subscribe
Home / Perspective / From the CEO
People First

The Next Phase of Enterprise Automation Is Unification

By Ravi Krishnamurthy, CEO, IntelliBuddies® and VIT Infotech3 min read

Enterprise automation has arrived in waves. First came tools that execute: scripts and software robots that carry out repetitive tasks. Then came tools that orchestrate: process engines that route and govern work. Most recently came tools that understand: assistants and generative AI that read, summarize and converse. Each wave solved a real problem, and each was bought separately.

My argument is that the next phase is unification. Not a bigger tool, but systems whose parts were built to work as one, because the value, as I have argued in this series, multiplies. Buyers are starting to feel it, and the companies serving them will be judged on it.

What buyers are discovering

The lessons of the last few years point in one direction. The seams between tools lose context and blur ownership. Pilots that rely on hidden help do not survive production. Activity counts hide how little work finishes. Governance has to be built in, not added afterwards. Each of these is a symptom of the same cause: three kinds of capability, bought at different times, and joined by hand.

None of this is anyone's fault. Each wave arrived when it did, and each purchase was sensible when it was made. The pattern only shows when you step back and look at the whole process, from request to result.

What unification means, and what it does not

It does not mean one vendor for everything, or a giant tool that does each job poorly. A specialist will often be better in a niche, and many enterprises will keep systems they cannot replace. Unification means that the three things a process needs, understanding, orchestration and execution, share a foundation. There is one record and one set of controls. The handoffs are designed in advance. And there is a path to add capability without starting again. Where parts come from elsewhere, the connections are designed and owned.

Tools that executeScripts and software robots
Tools that orchestrateProcess engines that route and govern
Tools that understandAssistants and generative AI
↓
One systemOne record · one set of controls · handoffs designed in advance · a path to add capability without starting again
Three waves, each bought separately. The next phase joins them on one foundation.

How to judge a platform

Five questions help. Are the three factors designed together? Can a customer start with one and add the rest without re-platforming? Is there one record and one set of controls across all three? Can the company show finished-work measures from live use? And can a customer leave, taking data and logic with it? A company that answers these plainly has probably thought about the seams. One that answers with a list of features has probably not.

  1. Are the three factors designed together?
  2. Can a customer start with one and add the rest without re-platforming?
  3. Is there one record and one set of controls across all three?
  4. Can the company show finished-work measures from live use?
  5. Can a customer leave, taking data and logic with it?
Five questions for judging a platform.

The strongest objection

The objection is that markets fragment and re-bundle in cycles, that best-of-breed specialists connected by open interfaces will keep winning, and that platforms tend to bloat. There is history on that side. Specialists thrive, and no platform does everything well.

I am not arguing against specialists. I am arguing that the connections between them cannot be an afterthought. Whether a customer uses one vendor or several, someone has to own the record, the controls and each seam. Unification is a property of the design, not of the invoice.

The principle

Value multiplies, it does not add. The next phase of enterprise automation will reward whoever makes understanding, orchestration and execution behave as one, with people in charge of the decisions that matter, and will measure the result in work finished.

The future is not more agents. It is agents that behave like one.

What this means for you

If you lead finance

Expect to be asked to fund outcomes, not tools, and ask who owns the whole result from request to finished work.

If you lead technology

Audit the seams before adding another tool, and make unification a selection criterion, whether you buy from one vendor or several.

If you invest in or partner with this market

Judge companies by their seams, their governance and the finished-work evidence from live use, and ask where the design ends and the glue begins.

I hold this belief as part of what we stand for at IntelliBuddies.

Read the principle: Value multiplies, it does not add →

If you disagree, tell me where. Send a reply

RK
Ravi Krishnamurthy

CEO, IntelliBuddies® and VIT Infotech. More than three decades in software. Writes about how enterprises can get work finished, with people in charge of the decisions that matter. Read the manifesto.

Previous essay

Honest Curves: Why Month One Looks Worse Than Month Twelve

Get the next essay by email

New essays from the CEO, delivered to your inbox.

Subscribe
IntelliBuddies®

Platform

  • Conversational Agent
  • Workflow Agent
  • RPA Agent

Resources

  • Impactful Areas
  • Stories of Impact
  • Newsroom
  • Events
  • Videos
  • Blogs

Perspective

  • Manifesto
  • From the CEO

Company

  • Product Documentation
  • About us

Contact

support@intellibuddies.com
(925) 558-8928
Stay up to date with IntelliBuddies® Subscribe to receive updates, news, and announcements
© 2026 IntelliBuddies®. All rights reserved.
Privacy Policy Support Policy EULA