Three binary gates that kill or pass an idea. Three multipliers that rank what survives. If any gate is red, the idea stops — no matter how exciting the boardroom demo was.
At a closed-door roundtable — 12 enterprise AI leaders from pharma, chemicals, tech, and financial services — I asked what killed their last AI initiative. Not a single person said technology. The answers: no named owner, legal blocked it, "we're waiting for the platform team," and the most honest: "the person who championed it left." These three gates exist because of those answers.
Revenue-tied or >$500K/yr cost reduction vs. unclear efficiency gain
Process-driven department with a champion vs. freestyle culture
Working POC in 6 weeks vs. 12+ months to first output
Internal ops, customer experience, or new revenue? 70% of wins start internal — then use credibility for customer-facing moves.
Where the most money gets wasted. We see companies spend $2M building what a configured GPT-4 pipeline does out of the box. The winning pattern: assemble — best-of-breed tools, configured for your workflow.
Kills more initiatives than bad technology. Winning model: thin central team (2-3 people) sets standards — execution in business units.
Start with guardrails, not policies. First 2-3 POCs surface 80% of governance questions. Separate "internal" (lightweight) from "customer-facing" (rigorous).
Adoption (week 4) → Efficiency (week 8) → Business Impact (3-6 months). No adoption by week 4? Stop.
Only 20% of pilots scale. It's a change management problem, not technology. Scale through process-driven departments first.
At a closed-door session, two heads of AI from a specialty chemicals firm and a medical devices company independently described the same pattern: teams skipping to Decision 2 ("let's pick a tool!") without making Decision 1 ("where do we play?"). Result: 12 disconnected experiments, none with enough focus to reach production. One reorganized around this sequence and went from 12 pilots to 3 funded programs — all three shipped.
Why 6 weeks: Short enough that organizational antibodies can't kill it. Long enough to build something real with actual data. Matches procurement — most departments can approve without board sign-off.
One use case, one owner, one metric. Define kill criteria: "if X doesn't happen by Week 4, we stop." Max 3 people in the room.
Connect to real data — not sample, not synthetic. Build the simplest version that touches real data. 80% data is enough; the POC surfaces which 20% matters.
Functional prototype, 2 rounds of user testing with 3-5 actual users (not stakeholders). This is where you set accuracy expectations: "wrong 20-30% of the time, here's the human fallback."
Run parallel with existing process for 5 days. Hard numbers: time saved, errors caught, cost vs. value. 2 hours/week × 10 people = 1,000 hours/year = €75-150K.
Scale, Pivot, or Kill. One room, one owner, one hour. Killing in 6 weeks for €50K is a win — you saved €500K.
An industrial equipment manufacturer had engineers spending 3-5 hours per inspection report — writing 14-15 page technical documents by hand. We built a voice-to-report system: engineers speak while they inspect, AI generates the full report in 10 minutes. Trained against thousands of documented failure modes, cross-references spoken observations against known issues, prompts for checks the engineer might miss. 1.5 months discovery, 2 months POC, live in production. Over 80% of stakeholders adopted it — despite knowing the data had quality gaps — because the human-in-the-loop was designed in from day one.
| Traditional | 6-Week Blueprint | |
|---|---|---|
| Time to first result | 6-18 months | 6 weeks |
| Investment before learning | €200K-€2M | €30-75K |
| Kill cost if it fails | Catastrophic | Manageable |
1-2 people. C-level sponsor + someone who actually understands AI. Set portfolio, allocate budget, make kill decisions. If this group meets weekly, something is broken.
2-3 people. Not a department. Evaluate tools, set standards, build context infrastructure (the layer most companies miss — your AI needs CRM data, brand guidelines, competitive positioning, or it produces generic output nobody approves). Train users continuously, not one-off workshops. Companies spend 10× more on AI platforms than training. A $50/month Copilot license with zero training is a $50/month waste.
In the business units. Squad model: 1 owner + 1 builder + 2-3 users. Execute POCs, operate scaled solutions, feed learnings back. The people closest to the problem build the solution.
Each win unlocks the next department through peer proof.
Process-driven, data-rich
Internal proof
Two wins → confidence
Highest resistance, peer proof
Board sees pattern
At our AGX session, someone said: "If you want to drain a lake, don't ask the frogs." That was about governance and compliance functions — they block AI because it threatens their gatekeeping role. The Domino Strategy exists because of this reality: start where resistance is lowest, build proof, then use that proof to unlock the departments with the most organizational antibodies. When supply chain saves 1,000 hours/year, operations asks to be next. That's organic scaling — more powerful than any top-down mandate.
Not abstract — your data, your org structure, your use cases.
Start with 30 minutes: I'll run Framework 01 on your specific situation and tell you what the first move is.
Let's Look at Your Top 3 Use Cases →