You bought tools.You ran pilots.the numbers look the same.
I take one part of your business, end to end, and work on it until the improvement shows up in the numbers you already track.
30 minutes · an honest first read · no pitch
2–3
Weeks
to the first answer on where it pays.
1
Process
taken end to end, not ten experiments.
0
Day rates
fixed fee per step, stop after any of them.
How it works
you decide after every step

Stage 01
Find where it pays.
- Input
- The numbers you already track, and a walk through how the business actually runs.
- Activity
- I map the handful of processes it runs on and estimate what better technology would be worth, and how hard it would be to get there.
- Deliverable
- A value map, and one process picked together. Two to three weeks.
- Decision gate
- If nothing credible shows up, we stop here.

Stage 02
Plan it with your team.
- Input
- The chosen process, and the people who run it and know where it hurts.
- Activity
- We define how it should run, what to build, what it costs, and what it should return.
- Deliverable
- A plan with numbers attached. You could execute it without me.
- Decision gate
- You decide whether to build, and with whom.

Stage 03
Build it, get it used, measure it.
- Input
- The plan, step by step.
- Activity
- Each step has a fixed price and a named result, built and put into real use.
- Deliverable
- A working system and an improvement visible in your operating numbers.
- Decision gate
- After every step you decide whether to continue.
What an engagement produces
No client names appear on this site until there is written permission to use them. Instead, this is the kind of artifact every engagement is built around — shown here with illustrative numbers.
Process economics map
Transformation brief
- Baseline
- 6.4h manual review per case, 9% rework
- Intervention
- Structured intake + assisted review, humans on exceptions
- Return
- ≈€9k/month at current volume, breakeven month 4
- Depends on
- Clean case history, one process owner with time
- Stop if
- Pilot misses agreed threshold after step two
Implementation scorecard
| Metric | Before | Now |
|---|---|---|
| Adoption | — | 84% of staff, weekly |
| Cycle time | 11.2 days | 6.9 days |
| Rework rate | 9% | 3% |
| Cost per case | €41 | €26 |
Built for companies that need results, not another AI programme.
Builder, not slideware
I build things myself.
I design, ship, and run production AI systems. That changes the advice: I will not put anything on your roadmap that I would not want to build and maintain. And when a vendor tells you something takes six months, I can tell you whether that is true.
No open-ended day rates
Fixed prices, small steps.
Every engagement is a fixed fee with a concrete deliverable. You can stop after any step and keep everything produced so far.
Even when it costs me
An honest recommendation.
If the answer after the first step is that AI will not move your numbers right now, I will say so and stop there. I would rather lose an engagement than a reputation.
Built and running today
Products
Kogsy Check-In
checkin.kogsy.careA daily voice check-in for older people still living at home. It calls at the time the client picks, turns the conversation into a plain note for the care manager, and brings a person in if anything seems off. I built it, and I run it.
How it works
- 1Scheduled call, at the time the client picks
- 2Conversation
- 3Plain note to the care manager
- 4A person steps in if anything seems off
Kogsy GO
gokogsy.comMost home care agencies cannot say what documentation costs them. GO measures it, with the agency’s own numbers, and the agency keeps the numbers whether they buy anything further or not.
How it works
- 1The agency’s own numbers
- 2Documentation cost, measured
- 3The agency keeps the numbers, either way
Building and running these taught me what this work takes in the real world: consent, reliability, and software that people with no time for software will actually use. That experience is in every roadmap I write.
About
Both sides of the work.

I’m Henrik Grünfeld. I have spent the last years building AI and voice systems and putting them into one of the hardest environments there is: home care, where the users are older adults and the staff have no patience for bad software. Kogsy is my firm for both sides of that work, the consulting and the products.
Operating experience
- AI and voice systems, designed, shipped, and run in production
- Complex systems in demanding operations, including healthcare and security
- Consent, reliability, and adoption with non-technical users
- Process measurement: what work costs, in the operator’s own numbers
Delivery
I work from Germany, with clients in Europe and the US. Engagements are delivered through HG Connect UG (haftungsbeschränkt).
Start with30 minutes.
Bring what you have tried, or bring nothing. You leave with an honest first read on where your business could profit from this, and where it will not.
A good fit if
- You run a real operation with numbers you already track
- You want one process improved end to end
- The people who run that process can be in the room
- You want fixed prices and the option to stop after any step
Probably not if
- You want a strategy deck without implementation
- You want ten experiments running at once
- You need an AI initiative for its own sake, results aside
