-
Notifications
You must be signed in to change notification settings - Fork 4
First 90 Days
Not the theoretical start. The Monday morning start.
Everyone who has read this far is holding a version of the same question: where do I actually begin?
The sequence below is deliberately unglamorous. The goal is not to launch a transformation. It is to finish three specific things that tell you, with evidence rather than intuition, whether the organisation is ready and what to build first.
Day 90 ends with a decision memo, not a demo.
Week 1 · Score your own readiness. Complete the NORTH scorecard for the organisation. Get input from frontline supervisors, not only from the leadership team, because the two will disagree, and the disagreement is the finding.
Week 2 · Walk the floor. Shadow three people in your highest-volume operational area. Listen for the Sigh Test. Document what you observe, not what you are told.
Week 3 · Inventory the automation you already have. Classify every bot and tool as Workhorse, Zombie or Upgrade Candidate. Most organisations discover more zombies than they expected.
Week 4 · Score your top five candidates. Rank them. Pick the highest scorer your team can actually build in eight weeks, not simply the highest scorer.
Week 5 · Begin Proof of Value. Sponsor named and briefed. Baseline established. Do not write a line of code yet.
Week 6 · Confirm data access. If you hit a wall here, that wall is your real first priority, not the agent. More programmes stall on this week than on any technical problem.
Week 7 · Run the pre-mortem with the full team. Find your most likely failure modes before you build. Write the mitigations down.
Week 8 · Begin Proof of Competence. One developer, one domain expert, one product owner. Eight weeks maximum. If it cannot be demonstrated in eight weeks, the scope is too large. That is information, not a setback.
Weeks 9 to 10 · Build the golden dataset. The domain expert builds it. The developer does not participate in this step. The person who defines what good looks like cannot be the person optimising against it.
Week 11 · Demonstrate against the golden dataset. Do not demo to the steering committee until you have an accuracy score to present. A demo without a number is a performance.
Week 12 · Make a deliberate decision. Proceed, redesign, or park. Write a one-page memo with the data that drove it.
Day 90 · Present the memo, not the demo.
The thing you are proving on day 90 is not that the agent works. It is that your programme has discipline. That is what gets the second one funded.
The most practical thing a leader can do in ninety days costs nothing and takes five minutes per conversation.
| Stop saying | Start saying |
|---|---|
| "We're implementing AI" | "We're redesigning how our team spends its time" |
| "The agent will handle that" | "The agent handles the routine part; our team handles the judgement part" |
| "We're automating the process" | "We're automating the keystroke work so people can do the thinking work" |
| "This improves efficiency by a third" | "This frees the team from a third of the work they say they hate most" |
| "We're in the pilot phase" | "We're running a controlled experiment with clear success criteria" |
| "We need to move faster on AI" | "We need to prove the first one works before we build the second" |
| "The technology is ready" | "The technology is ready. Let's assess whether our processes and people are" |
These are not euphemisms and the difference is not tone. Every sentence on the right is falsifiable and every sentence on the left is not. The right column also happens to be what people hear anyway, so saying it out loud costs nothing you had.
The efficiency row is the one that matters most. "Improves efficiency by a third" is heard, instantly and at every level, as "cuts a third of us."
Programmes fail when the steering committee meets to receive status. They work when it meets to decide. That requires a different agenda.
| Minutes | What |
|---|---|
| 10 | Three numbers only. Production accuracy against baseline. Cost per transaction against target. Open risk items needing a decision |
| 15 | One decision. Not a discussion. A decision, with a recommendation and the data behind it |
| 10 | Portfolio. What is in each gate. What is blocked, and what unblocks it |
| 10 | People. Adoption. Training completed. Reskilling progress for affected roles |
No meeting without at least one decision made and recorded. If there is nothing to decide, cancel it and give everyone forty-five minutes back.
The fourth block is the one that gets dropped when the meeting overruns, which is how a programme discovers its adoption problem two quarters late.
By day 90 every candidate should sit in one of these. This becomes the twelve-month roadmap and the defensible answer when the board asks what you are building.
| Low complexity | High complexity | |
|---|---|---|
| High value | QUICK WINS · start here | MOONSHOTS · earn these |
| Low value | INTERN TASKS · later | MONEY PITS · kill now |
Quick wins. Rule-based, high-volume, clear procedures. Your first three production agents. Build them with the full governance discipline so they become the proof that the discipline is affordable. Examples: payment reconciliation in finance; order intake in logistics.
Moonshots. High value, high complexity. These need the factory operating at real maturity. Do not start here. Come back once the quick wins have built trust and capability. Examples: end-to-end case adjudication; multi-party contract negotiation support.
Intern tasks. Real opportunities that do not move a core metric. Year two, with citizen developers, once the platform exists. Examples: internal FAQ assistants; standard report distribution.
Money pits. High complexity, low value. These feel exciting in a workshop and devastating in a post-mortem. Examples: predictive analytics with no decision attached to the prediction; sentiment analysis with no action pathway.
The money-pit quadrant is the one with a name for a reason. Its occupants are almost always the projects an executive is personally attached to, which is why the quadrant needs to exist on paper before the conversation happens.
Adapted from the AI CoE and Agent Factory Playbook, based on The Augmented Enterprise framework. Vertical specifics have been generalised and all examples replaced.
Back to Home · The three proof gates · NORTH and the trust ladder
The thinking
Operating model
Frameworks
Governance
Playbooks
Value and people
Reference
In the repository
The courses
Reviewed 2026-08.