How do you make twenty champions?
A 40-year-old animal health company with a ₹2,000 crore valuation didn't have an AI motivation problem. Their leadership was already excited, already buying tools. They had a deployment problem — and the fix wasn't a better demo. It was giving twenty people a page that was already about their own job, and letting them argue with it.
By Rhea Karuturi · a day in Ahmedabad with the Hester team
The problem was never enthusiasm
Hester Biosciences makes animal vaccines and health products out of Mehsana, Gujarat. Forty years old, listed, roughly ₹2,000 crore. I went in expecting the usual adoption story — a sceptical old-economy company that needs convincing about AI.
That's not what I found. Priya, who runs the business, was already sold. The leadership team had already brought several technology solutions into the company. On our first call it was obvious that the will was there, and had been for a while.
What they had instead was subtler and much more common: a start-stop problem. Tools arrive with real momentum at the top, then hit the floor and stall. Not because anyone objects — because nobody in the middle ever became responsible for making them work.
That reframes the job entirely. If the constraint were motivation, you'd build a better pitch. The constraint is ownership distribution, so you have to build something else: more owners.
One champion is a single point of failure
Every stalled tool I've seen has the same shape. Someone senior loves it. That person gets busy. Nothing about the tool survives their attention moving elsewhere, because nobody else ever had a stake in it.
One champion
Leadership buys in, sponsors the tool
Rollout
Demo, training, logins issued
Attention moves
The sponsor gets a new priority
Drift
Nobody owns the gap between tool and work
Stall
The tool is alive in the licence, dead in the day
The failure isn't at the start. It's four steps later, and it's quiet.
Stop demoing at people
The morning was a working dashboard for their Pet Care division — real formats, real numbers, an agent that reads the page. It was genuinely good. It was also the least important thing that happened all day.
A great demo makes spectators, not owners
We built the Pet Care division as a live operating layer: a WhatsApp-style field app where reps get their tour plan, check in against a geofence, record a doctor meeting and leave a voice note — which lands in the dashboard as an order draft, competitor intel and a geographic view of what's moving.
People liked it. They said so. And that reaction is exactly the danger: liking a demo costs nothing. Nobody in the room had anything at stake in it yet. They were an audience, and audiences disperse.
Make the room co-authors before you build the product
So I said it out loud, early: today is not about this dashboard. Today is about going from Priya and her leadership team being the champions for new technology, to every one of the twenty people in this room walking out as a champion for whatever we build.
That reframing does something structural. It moves the room from evaluating my work to specifying theirs — and it makes the rest of the day a working session rather than a pitch.
- Ask what breaks in their week, not what they think of the product
- Name the person, not the department — pages belong to humans
- Treat every objection as a requirement you didn't have yet
The post-lunch page
Over lunch I took the feedback each person had given that morning about their own work, and turned it into twenty-one pages. After lunch, everyone clicked their own name.
Kill the blank page
The standard way to gather requirements is to ask: what would you want your dashboard to show? It sounds respectful and collaborative. It almost never works.
Ask a procurement head that question cold and you get vague, safe answers — “sales figures, maybe some trends” — because you've handed them a blank page and asked them to be an information architect on the spot. That's not their job, and it's genuinely hard.
So we inverted it. Every person got a page that was already a draft of their own job, built from the words they'd used that morning, with their verbatim asks printed as chips across the top so they could see themselves in it.
Twenty-one names. Everyone finds themselves in the first five seconds.
Let them edit the page by talking to it
Each page has a chat beside it, and the chat doesn't just answer — it rewrites the page. Tell it the first chart isn't what you need, or that a section is missing, and the agent changes the layout live: adds a section, drops one, reorders, replaces the KPIs. The changed block highlights and scrolls into view.
Underneath, the page is a spec — an ordered list of sections as data, not code — and the agent emits edit operations against it. That's the whole trick. It means feedback isn't collected for later; it's applied in front of the person who gave it.
- Watching the page change is what converts “sounds nice” into “this is mine”
- The correction is more precise than any answer to an open question
- Reset is always one click away, so nobody is afraid to break it
Procurement — built from your feedback
| PO # | With | Days | Status |
|---|---|---|---|
| PO-4471 | Plant Head | 11 | Awaiting 2nd quote |
| PO-4483 | Finance | 9 | Budget code query |
Shape this page
The agent spotted the gap itself — and the page changed while its owner watched.
Take their documents in the format they already have
The second half of the blank-page problem is the data question: send me your files and we'll see what we can do. That stalls for weeks, because it asks people to prepare something.
Instead, each page takes uploads in whatever shape the work already lives — the monthly expense statement, the distributor stock Excel, a scanned register, an MS Project plan. The spreadsheet parses in the browser and their real columns appear on the page immediately, and the agent then rewires the surrounding sections to say which are now fed by that file.
- Their columns, unchanged — we build the parser around the format, not the reverse
- What can't be opened in a browser says so honestly, and shows exactly which fields get extracted server-side
- The proof is the point: your file, your rows, on your page, within seconds
Capture it, or the day evaporates
The first version of this had a real flaw: all that feedback lived in the browser session. Twenty people spent an afternoon telling the product exactly what they needed, and it would have vanished on refresh.
Now every message, every applied edit and every upload is appended to a log — append-only, so nothing can be quietly edited away — and there's a review page grouped by person, exportable as a document. That log is the requirements spec. It's written in their words, timestamped, with the agent's response beside it.
What changes when the page is already theirs
Blank page
Draft page
If you're running one of these
A tool with one champion dies the moment that person gets busy. A tool with twenty gets built into how the company actually works.
The pages we made that afternoon are demos. Some of the numbers in them are synthetic and clearly labelled as such. None of that mattered as much as the thing that happened in the room: twenty people stopped watching a product and started correcting one.
That's the whole method. Don't ask people to imagine what AI could do for their work. Show them a draft of their own week, get it slightly wrong on purpose, and let them fix it in front of you. They'll tell you more in ten minutes of arguing than in a month of requirement-gathering — and at the end of it, the thing has their fingerprints on it.
This is the shape of the ₹5,00,000 session tier — a customised day with a demo dashboard built for the company beforehand.