The board wants an AI plan for HR. You want one that won't blow up in your people's faces.
Somewhere between "we need an AI plan" and "let's not break anything," there's a real strategy.
We help CHROs turn what they already know about their people into an HR-AI strategy the board can back, then prove it one use case at a time, in systems your security team signs off on.
30 minutes. No deck. Just a conversation.
Three people asked you this quarter
-
“Where's our AI plan for HR?”
-
“Every function brings an AI line item this year.”
-
“We're standardizing AI tooling. What does HR need?”
The plan is going to get written either way. Better if you're the one holding the pen.
It's a fair question, and someone will answer it. Maybe IT, as part of a wider roadmap. Maybe a consultancy with a very nice deck. Maybe a vendor whose product happens to fit.
You're the one who knows which teams are one resignation away from trouble and which promises the company can't afford to break. That's what the plan gets built on.
You don't have to become technical to write it. It helps to have a way of thinking about where AI genuinely helps a people function. That's what we bring.
If you've been wary of HR AI, you've had good reason.
95%
of enterprise GenAI pilots show no measurable P&L impact
MIT, Project NANDA
You've probably watched one of these up close. Very few had anything to do with the technology.
-
The tool showed up before anyone knew what it was for.
Someone brought in a copilot, then went looking for something to point it at. It was never aimed at a number anyone reported on, so budget season came and there was nothing to defend it with.
-
It reached employee data before anyone trusted it.
The pilot met the security team after the demo instead of before it, and that was that. You may have been the one who slowed it down.
-
No one person owned whether it worked.
IT looked after the software, HR looked after the process, the vendor looked after the roadmap. When the number didn't move, it wasn't clear whose problem that was.
The caution is earned. Most of it is avoidable if a few things get decided up front.
The five questions we'd walk through with you.
We learned them running our own HR AI inside enterprises for two years. All five land on one number: employee lifetime value, how well your people perform and how long they stay.
Most AI frameworks sequence by ROI. In HR, you sequence by trust.
The data is personal, the users are wary, and mistakes travel. That's why these five are in this order.
Impact
Which number does it move?
eLTV
performance × tenure
Trust to earn
How much does it have to earn first?
Readiness
Is your function ready?
Three lenses. Every question runs through one of them.
- 01
Which number do you already report that you'd most like to move?
Impact
Attrition in the roles you can't afford to lose. Time-to-fill. The hours your team spends answering the same forty questions. Start from one of those and the strategy has somewhere to go.
Skip this and you end up automating whatever the demo happened to show.
- 02
Which of those can AI actually help with?
Trust to earn
Ask four things. What's it worth if it works? What does it cost if it's wrong? What data does it need? Will people use it?
This is where the impressive idea gets separated from the useful one.
- 03
In what order?
Readiness
Start with the boring, provable, low-risk things. Each one earns a bit more trust with security, with your people, and with the board, and that trust buys you the next attempt.
This is how the flagship project ends up at the back of the queue for three years.
- 04
What stays human, no matter what?
Trust to earn
The one that gets programs approved. We answer it in full for your security team
Decide early what AI will never decide on its own: hiring, pay, promotion, discipline, termination. Draw those lines first and the rest gets signed off.
Most first security reviews fail here.
- 05
Whose name is on the number?
Readiness
One person, one metric, one date when the number gets looked at. Without that, however good the demo was, what you have is a pilot.
Otherwise the metric belongs to nobody.
That's the whole frame. If your plan already answers all five, take it and go. If it doesn't, working through them for your function is what we do with CHROs, and it takes about a month.
We'll walk the five questions with you on the call.
Run the five questions on one ideaSee the map of HR agents this thinking produces
From the board question to a system that runs.
-
The Strategy
We work through the five questions for your function: your metrics, your use cases filtered and ordered, your guardrails, your business case. Written in your language, so you can present it without us in the room.
You leave with The strategy you walk back into the exec meeting with.
-
The Proof
A working version on test data, in our sandbox or one in your own tenant, so you can watch it do the job before you commit to anything real. No live employee records either way.
You leave with Proof it works on your data, before you spend real money.
-
The Build
Built inside your environment, under your security team's rules. We plan how people will use it and how you'll measure it before we write any code. Your team runs it without us after that.
You leave with A production system, and the number that justifies it.
Sometimes the strategy work says "don't build anything yet." When it does, that's what we'll tell you.
Fixed fee for the strategy month, then a fixed fee per use case. Hosting and model usage are billed to you by your own providers.
The answers your security team will want.
You'll be asked these before anything gets connected, so they sit in one place. Send this page on.
- Where it runs
- Our sandbox on synthetic data, or one in your own cloud tenant on synthetic or de-identified data your security team approves. No live employee records in the proof either way. Production runs inside your environment.
- The model
- It runs where you choose, on the provider your security team approves, with that provider's no-training and retention terms in writing.
- What AI will never decide
- Hiring, pay, promotion, discipline, termination. It can inform, rank, summarize or draft; a person makes the call. You get that list in writing before anything is built.
- Access
- Through your identity and your devices. We hold no production credentials after handover.
- What you keep
- Code, prompts, evaluation sets, documentation and runbooks. No Culturro platform dependency.
- Certifications
- We hold none and claim none. The system runs under your controls and inherits your posture, so what your team assesses is our process.
A month from now.
AI comes up in the exec meeting again. This time you've got a page. Five questions, answered for your function, tied to numbers the CFO already tracks. When the CEO asks why this one first, you have a reason, and it's a short one.
Deep experts in how people and organizations work. Who also build AI systems.
Client engagements, two products of our own
-
2017
In the people domain since
-
30+
Organizations across engagements
-
5,000+
People coached by Merlin, from 40+ companies
-
2 yrs
Running our own HR AI inside enterprises
Culturro started in 2017 as a people consultancy. We built Agnya for employee experience, then Risely for leadership development, whose AI coach is Merlin.
The people domain came first.
Organizational behavior since 2017, long before AI was the question. Culture, managers, how teams actually change. It's why we can tell an AI idea that demos well from one that survives contact with real people.
We built and ran the real thing.
We built Merlin, an AI coach for managers, and have run it inside enterprises for more than two years. Security reviews, adoption problems, renewal conversations, all of it on our own product, with our own name on the line.
You keep everything.
Built in your environment, under your security team's rules, with no-training terms in writing. You get the code, the documentation and the runbooks. Any technical decision we make, we'll explain in plain English.
Walk in with the answer.
Thirty minutes on your function and your numbers. We'll go through the five questions and tell you what we'd do first, or whether we'd do anything at all yet.
No deck. Bring the numbers you already report on.
- Fixed-fee strategy, about a month
- Sandboxed proof before any real data
- You keep the code, the docs, and the plan