Tools · Interactive · EU edition
The productivity-to-P&L simulator.
You get pitched a big AI saving. This shows how much of it actually reaches the ledger - and where the rest leaks.
How to use it: ① Pick the scenario closest to yours. ② Nudge the few sliders to match your reality. ③ Read the plain-English result on the right. That's it - the deeper tabs are optional.
Step 1 · Pick a starting scenario
Step 2 · Your setup
EU overlay: AI Act compliance for a high-risk system ≈ €193–330k setup + ~€71k/yr (CEPS); an SME deployer ≈ €5–100k; €0 if not applicable.
Step 3 · The four levers that matter most
① How much faster the task gets30%
The speed-up on the one task you're targeting. Demos show 15–50%+ - but that's one task, on a good day.Dell'Acqua RCT
② How many people really use it45%
Share of the team that genuinely adopts it in daily work. Buying more seats doesn't help - real, embedded use does.Guerrini–Rice; Hajikhani
③ Are you ready? (data · skills · process)50%
Without the right data, experienced people and a redesigned workflow, the firm-level gain is often zero - not just smaller. This is the gate most companies fail.Fan; HBS; Aldasoro
④ Do you turn saved time into money?
Saved time isn't saved money until you decide what to do with it. The base rate: 83% of firms change nothing - the time just becomes slack.OECD; Nguyen–Doan; Jiang
▸ Show the other 4 gates (advanced)
⑤ How much of the job is that task?25%
A big gain on a small slice of someone's week barely moves their total output; freed time flows to their other tasks.Freund & Mann
⑥ Pilot vs real life60%
Pilots run on clean data and eager volunteers; production is messier. Keep ~50–70% of what the pilot showed.Aldasoro (16%→4%)
⑦ How mature is the rollout?100%
100% = steady state. Gains take ~2–3 years to arrive - drop this to see an early year. (The timeline is on the "pays back" tab.)Hajikhani; HBS; Guerrini
⑧ Do you keep the gain?50%
In competitive markets a gain passed to customers as lower prices leaks away. But an internal cost-out saving is yours to keep - so picking "Cut cost" above raises this automatically; a moat raises it further.Johnston–Makridis; Carreño
💡Try dragging ③ Are you ready? down to 20% - watch the realized number collapse. That single gate is the article in one move.
The result
You were pitched€0task gain × adoption
Actually reaches the P&L€0after all 8 gates
Where the pitched number goes
–Payback (years)
–3-yr net ROI
€0Realized / person / yr
Optional - the intuition behind Gate ⑤. A workflow is a chain of steps. AI usually speeds only some of them. Drag the dividers to set how long each step takes, toggle which ones AI helps, and watch why a big task gain becomes a small process gain.
Your workflow - drag the dividers
Drag a divider to move time between steps.
What actually happens to the whole process
Best single-step speed-up0%the number on the box
Whole-process time saved0%what the workflow gains
0%of time AI touches
0%effective job-level gain
Optional - the timeline. The cost lands now; the benefit ramps in over a 2–3 year lag. That shape - loss first, payoff later - is the productivity J-curve. This shows when (and whether) your case turns positive.
Investment profile
One-off cost (€)150,000
Annual tool cost (€)60,000
Annual benefit at steady state (€)
Pulled from the leak tab, or type your own.
Years until full benefit (the lag)3 yr
~3 years is the evidence-based default.Hajikhani; HBS; Guerrini
Horizon5 yr
Discount rate10%
The J-curve (cumulative discounted cash flow)
€0NPV over horizon
–Breakeven year
€0The valley (deepest point)
Every lever is anchored to a study
| Gate | Plain meaning | Default | Evidence anchor |
|---|
The maths: Pitched = people × cost × ① task gain × ② adoption. Realized = Pitched × the remaining six gates (each a 0–1 multiplier). Because they multiply, a chain of individually reasonable discounts collapses the pitched number by one to two orders of magnitude - which is the whole point. These are planning heuristics, not forecasts. The studies span different settings (elite consultants, Chinese and EU/US firms, pre-LLM Finnish data), most are working papers, and several proxy "AI" indirectly. Use the tool to find your biggest leak and pressure-test a business case - not to predict a number to the euro. Full detail is in the companion "Evidence Synthesis & Skeptical Assessment". EU context: 20% of EU firms use AI (Eurostat 2025); Europe has ~closed the adoption gap with the US (37% vs 36% gen-AI) but not the deployment gap (55% vs 81% multi-activity, EIB). The newest 2026 firm data - St. Louis Fed (+0.07% utilization-adjusted TFP) and McKinsey (only 39% see any EBIT impact despite 88% adoption) - confirms the conversion gap persists on today's models; only Gate ① is model-sensitive.
Defaults anchored to the graded literature. Planning heuristics, not investment advice.