An AI-native consultancyStrategy and operating model design
From AI ambition to enterprise value.
Most organizations can already name three things AI could do for them. Far fewer can say what those three things are worth, who would own them, or what would have to change for the saving to reach the accounts.
That gap is where the money goes.
Book thirty minutes No obligation.
Photography · Tim van de Koppel
Scroll
01Where I come in
Start from the business, not the technology.
Four things get established before anything is proposed. None of them is a use case list.
-
1
The business problem, and the number attached to it.
Cost, cycle time, capacity, error rate. A target somebody has already committed to. If there is no number, there is nothing to design against.
-
2
What the work actually does today, as opposed to what it is believed to do.
Not the process map. What the work does when nobody is presenting it. The two are rarely the same document.
-
3
Which parts a machine could carry, and which it could not.
Judgment is not one thing. Some of it is pattern matching a machine already does better, and some of it is accountability that cannot be handed to one.
-
4
What has to change around it before either is true.
Ownership, the data underneath, who is allowed to decide what, and where the freed hours are supposed to go. This is the part that gets skipped.
02How the work runs
I make enterprise AI pay back by fixing the organization around it.
-
1
Diagnose
I start with your numbers, not with AI. How long the work takes now, what it costs, where it gets stuck, and which parts of it a machine could realistically carry. Without a before, there is no after.
-
2
Design
Then the organization around it. Who owns the work, who decides what, and what has to change so the technology has somewhere to land. Buying tooling for a process nobody owns only makes the confusion faster.
-
3
Embed
And what makes it hold. What your people need to be able to do, and how you keep seeing whether the value is actually landing. A tool nobody was trained for quietly stops being used.
A plan you can act on, with a number against every step and a name against every owner.
I am independent: no software to sell and no implementation team of my own. The practice was built for this work rather than turned toward it. If you need a build team, I will help you choose and brief them, contracted directly by you, with no fee to me from either side.
03The diagnosis
Most enterprise AI never reaches the P&L.
Budgets are not the constraint. Models are not the constraint. The operating model is.
The adoption gap
- 17%
- of organizations have deployed AI agents
- 60%+
- expect to, within two years
- 40%+
- of agentic projects forecast to be canceled before end of 2027
Gartner, 2026, on deployment and intent. Cancellation forecast: Gartner, June 2025.
AI stalls where nobody funded the change past the pilot, where the work was never redesigned, where data has no owner, and where nobody decided what a machine may decide. None of that is fixed by a better model.
Ambition is not the constraint. The capability to carry it is.
04The pattern
The same three failures, in different industries.
What recurs in regulated environments. None of it is a model problem.
-
1
No end-to-end owner
The work crosses four departments and belongs to none of them, so nobody can decide where an agent may act.
Every escalation goes sideways instead of up, and the decision that would unblock it is never anyone’s to make.
-
2
Governance written by the builder
The team that built the solution also defined the check on it.
It validates the build, not the risk. The first genuinely novel failure walks straight through it.
-
3
Value never baselined
Nothing was measured before, so the after-number cannot be proven.
Funding stops at pilot, not because the pilot failed but because nobody can show that it worked.
None of these are model problems. Every one of them is an operating model problem.
Observed across public sector, insurance, pharma and semiconductor engagements.
05Where I work
Almost everyone sells the 10 and the 20.
The 70 decides whether any of it returns money. That is the part I am hired for.
You cannot ship your way out of a missing operating model.
The 10/20/70 split is BCG’s. The observation that almost nobody sells the 70 is mine.
06What actually changes
The work does not disappear. The decision points move.
One process, three states. Illustrative: an exception inside a finance operations team.
TodayNo AI in the process
- person
- person
- person
- person
- person
One person carries the whole case. Deciding it and doing the paperwork are the same job, so the queue is the only lever you have.
AssistedAI as assistant
- person
- machine
- person
- person
- person
The machine prepares the case. The person still decides and still types it in. Faster per case, same cost base, and the saving lands in the backlog.
AgenticAssistant plus bounded agents
- machine
- machine
- bounded
- bounded
- machine
The machine acts inside a bounded space and is checked by something it did not write. The person moves up: setting the bounds, judging the edge cases.
This is not a productivity story. Roles move up, roles shrink, and both belong in the business case.
The check in JUDGE is written by someone other than whoever built the agent. That separation is the whole design.
07Where value actually leaks
Freed capacity has three destinations. Most programs name none.
How much a machine takes varies per step, and at some steps it is most of the step. What happens to what it frees is where the P&L is won or lost.
-
1
Redeployed
People move onto work that never got done because nobody had the hours. Usually higher judgment, usually closer to the customer.
Capacity added. Cost base flat.
-
2
Orchestrated
The same people run several automated flows and hold the exceptions. Fewer hands per unit of output, and a materially different job.
More output per head.
-
3
Released
Roles that were mostly routine get smaller, and some of them go. That is a real number and it belongs in the business case, not in a footnote.
Cost base actually falls.
- 100% of work
- Automated work The percentage depends on the work.
- Released capacity Capacity is created. Decisions determine its fate.
- Value realization Value becomes visible and repeatable.
- Measurable P&L impact
The design covers both halves: what the machine takes, and where the capacity it frees ends up.
08The capacity trap
Freed hours are not a saving until someone decides where they go.
Every level of the organization has a backlog waiting to absorb them. That is not a failure of the technology.
Photography · Tim van de Koppel
09Built for it, not retrofitted
AI built as a capability in its own right, not bolted onto the ones you already have.
Readiness is not a badge and it is not an audit you pass. It is a design premise: a business built so a machine is a normal input to how it runs.
Eight dimensions of the operating model. They are not independent and they are not equally urgent.
Which one you are actually standing on is the finding, and it is rarely the one the organization is currently working on.
Capability is not a course you run afterwards. It is part of the design, or the design does not land.
10The value model
Your baseline, not my benchmark.
I do not import someone else’s percentage. I build your number in week one, from your systems, across all three kinds of value.
-
1
Baseline
Volume, cycle time, rework, escalation rate and cost to serve.
Taken from your systems, before anything is built.
-
2
Where the work is
Routine volume, judgment calls, and the knowledge work that never gets done because nobody has the hours.
Not every hour in the work is the same hour.
-
3
Value at stake
Cost released where the work was routine. Coverage where you used to sample. Capability added where people can reach work they could not.
Sized against your P&L, with the assumptions written down.
The output is a number your CFO can challenge, and a roadmap sequenced against it.
What I will not do: quote a percentage from someone else’s engagement and let you assume it is yours.
11How I engage
Three ways to bring me in.
Diagnosis first. Longer only where the design still needs an owner.
2 to 3 weeks
AI Value Assessment
- Starts from your business agenda, not a use case list
- Your baseline built in week one, from your systems
- Maturity and sequence across the eight dimensions
- Prioritized roadmap and board-ready readout
Fixed scope, fixed price.
6 to 10 weeks
Operating model design
- Target operating model across the eight dimensions
- Decision rights, guardrails and escalation thresholds
- Use cases qualified, sequenced and owner-assigned
- A handover pack your team builds against
Fixed scope. Fixed end date. Scoped and priced during the Assessment.
Ongoing, light
Advisory alongside your team
- Sparring for the AI or transformation lead
- Review of strategy, roadmap and governance
- Independent challenge on vendor choices
- No delivery capacity, just judgment
A standing half-day, at a fixed cadence. From € 2,000 per month.
The assessment carries a delivery guarantee: no prioritized, board-ready roadmap, no invoice.
No open-ended delivery capacity, and no build team of my own. If you need one, I will help you select and brief it, contracted directly by you and with no fee to me from either side. · Available direct, or through your preferred contracting party.
The operating model is the difference
Between AI ambition and enterprise value.
A useful first conversation
Tell me which part of the business has to change next year, and I will tell you which of four things is standing in the way: a process problem, a data problem, a governance problem, or a sponsorship problem. That is usually enough to know whether it is worth working together.
Thirty minutes. No obligation.
Bring the target you have to hit next year.