Your browser asks you to confirm each download.
This preview condenses Turning AI Potential into Enterprise Value by Erlend Hollebosch into one argument and one method. It is written for executives who sponsored an AI pilot that worked and are still waiting for it to reach the P&L.
The sequence that follows is not a preference but arithmetic: design for coherence first, then amplify. Read it in order, or follow the route for your seat at the table.
Somewhere in your organisation an AI pilot worked. Processing time fell, quality rose, the board was pleased. Six months later it is still a pilot: the gain is invisible in the P&L, the licence costs are not.
Ask the executive table why, and you hear nine sincere answers. The CFO sees licences before evidence, the COO an operation that cannot pause, the CIO shadow AI and agents without owners, HR a fear no communication plan reaches. Procurement watches lock-in signed before value, Supply Chain a forecast nobody trusts, and business units lose their best people while the benefits are claimed centrally.
Each explanation is locally true — so each fix is only partly effective. Finance tightens funding and pilots slow; IT tightens governance and shadow AI goes underground.
Successful pilots stall because the organisation around them cannot absorb them.
Decisions wait, work keeps changing hands, priorities compete for the same people.
Coherent enterprises compound value; fragmented ones compound confusion.
Five properties, a fifteen-day diagnostic and five gates with named owners.
Three horizons, five measurement rules and a 1.5× funding threshold.
Ask why the pilot never scaled and every executive gives a sincere, accurate answer — from where they sit.
Each explanation is locally true — so each fix is only partly effective. Finance tightens funding and pilots slow; IT tightens governance and shadow AI goes underground. HR launches engagement and the fear stays, because fear was never the root.
The root cause is not AI, data, culture or leadership. It is incoherence: the parts of the enterprise each work reasonably well, but the connections between them — decisions, handovers, priorities, information — no longer form a working whole.
Many people are involved; no one owns the trade-off.
Every department does its part; the shared outcome stays out of reach.
Everything is priority one, so nothing is.
None of this was created by AI. It accumulated over decades of local fixes — governance added after incidents, systems bought per department, reorganisations that moved boxes but not decision rights. Each layer solved a local problem; together they built an enterprise too complex to change quickly.
Coherence is not a slogan but an operating condition. It rests on five properties — none exotic, rare only in combination.
One name owns each decision — in one second.
Missing → decisions keep waitingEvery outcome that matters has one owner for the whole.
Missing → work keeps changing handsOne ranked portfolio, owned by the executive team as a team.
Missing → priorities keep competingThe data a decision needs reaches it without negotiation.
Missing → work waits for dataEvery person and every agent knows its authority and its never-do rules.
Missing → all three, amplified by agentsIts absence is already on your bill. The author puts it at three to eight percent of revenue — before AI. Finance can price it in ten working days, from minutes, calendars and one workflow walk.
This one-page incoherence statement is also your AI baseline: whatever AI promises, these three costs discount it. The question is never whether you can afford coherence — you are already paying for its absence.
Two confusions mislead programmes. Coherence is not alignment, the offsite where everyone nods; it is what remains three weeks later. Nor is it standardisation: the mature position is to standardise the backbone and differentiate the edge. McKinsey's 7S lens makes the gaps discussable — and agentic AI stresses all seven elements at once.
Following Jeanne Ross of MIT CISR, the book argues that most enterprises are not too small or too old to be digital — they are too complex. The remedy is an operational backbone: the shared core the enterprise runs on, surrounded by an edge where it competes on difference.
Standardise the backbone ruthlessly; differentiate the edge deliberately. Most enterprises do the reverse: a fragmented core drowning in variants.
No new AI capability enters a workflow until its redundant variants are named and scheduled for removal.
Simplification imposed by IT alone fails; the business knows which variants carry real customer value.
McKinsey's State of AI research keeps finding that the companies capturing value redesign workflows rather than deploy the most models. The workflow is the unit of transformation, redesigned in four moves — always in this order.
One sentence, with a number in it.
Agent alone, agent prepares, or human only.
Authority limits, escalations, never-do rules, audit trail.
Or the old workflow quietly returns.
Every redesign brings a performance dip. It is announced in advance, bounded, measured weekly and exited on criteria. What kills programmes is not the dip but the surprise of it.
AI brings no way of working of its own. It accelerates the one it finds.
For decades incoherence was expensive but survivable, because competitors carried the same burden. AI ends that truce. Agentic AI raises the stakes from advice to action: an agent that places orders or moves money must know who owns the outcome, where its authority ends and what it may never do. That is why agentic initiatives stall between demo and production — the technology is ready, but the organisation has not yet decided what it is allowed to decide.
Human in the loop is a design decision, not a slogan. Each workflow step sits in one of three positions. An agent starts supervised and graduates; it never starts free.
Irreversible, high-value or regulated actions
Reversible, bounded actions on the backbone
High-volume, low-risk micro-decisions on verified data
Every agent needs what every process needs: an owner, a documented scope, guardrails and auditability. Agentic AI is cognitive capital, not cheaper labour — treat it as ten-percent-cheaper labour and you get ten-percent outcomes.
The method is a rhythm, not a programme. Each step is timeboxed and closes with a gate held by a named executive. A gate can pass, adjust or stop — never drift.
Costs, heatmap, gap map, one selected workflow
A verdict per domain: proceed, fix in parallel, fix first
A North Star AI vision and a portfolio sequenced on impact and feasibility
Four redesign moves, the dip managed, value measured by Finance
Governance into the line; funding only at 1.5× verified value
Five domains are scored against the three signals on a heatmap. Every cell needs an artefact: a red without evidence is a grievance; an amber without evidence is a hope. The reddest, costliest cell is where you start.
Illustrative scoring. Your heatmap comes from your own artefacts: decision logs, workflow walks, data lineage, allocation maps.
A companion gap map makes three recurring gaps undeniable: the ownership gap, the boundary gap and the capacity gap.
An industrial company introduced AI demand forecasting. The forecast improved within weeks — and the business did not move. The book presents it as a composite of several organisations.
No owner for the chain, no shared data, no agent boundaries.
The book promises a mechanism, not a number. Finance fills the model with its own figures; what the model fixes is the shape of value.
Every claim must name its ledger line — cost avoided, cost reduced, revenue enabled or risk reduced. A claim that cannot name its type is not yet a claim.
Transformations rarely fail on technology. They fail on seven governance signals, each visible to someone long before it reaches the board.
| Signal | Counter-move |
|---|---|
| Hidden veto | Transfer decision rights in writing; make starved cases visible |
| Mispriced value | Price the case before it starts |
| Stalled champion | Train a second team; document the roles |
| Skipped training | No one signs a judgement untrained |
| Automation theatre | Redeploy freed hours; retire the old checks |
| Gravity | Re-run the heatmap on a fixed cadence |
| False negative | No agent enters a process whose heatmap is red |
Each week of postponement is not caution; it is the cost of incoherence, still accruing at three to eight percent of revenue. Day fifteen does not deliver a transformation. It delivers an honest picture of your coherence, priced choices and a decision actually taken.
Diagnose first. Fix the workflow before the technology touches it. Price the case before you build it.
Twelve chapters and three appendices take each idea further: the contribution and the gate each executive role signs; the capability curriculum of progressive exposure; the twenty-one conversations that move culture; a reflection for the executive team at the close of every chapter; and reference appendices with the nine delivery roles, a glossary and the instruments at a glance.
Erlend Hollebosch works on enterprise coherence and AI transformation, with a background in Agile, Lean and process optimisation.
grow-faster.be
Turning AI Potential into Enterprise Value — an executive preview of the short book for the executive team.