Picture the last Thursday of a quarter, 4:15 in the afternoon, a mid-sized conference room. An operations coordinator, not a director, not a consultant, stands at the end of five slides. The room has just seen four documents nobody knew the company had: a map of every AI tool actually in use, an honest readiness score with evidence attached, a vendor business case repriced from spectacular to sober, and one process nominated for a pilot with success and kill criteria already drafted. The chief operating officer leans forward and asks the question the whole quarter was engineered to produce: "What do you need to keep going?" Nothing about that scene is luck. It is the output of a ninety-day plan you can start next Monday with no budget, no title change, and no permission slip. This closing lesson of Level 1 is that plan, hour by hour, artifact by artifact, trap by trap.

The Quarter, the Four Artifacts, and the Honest Budget

Everything you have read in Level 1 converges on one uncomfortable fact: the difference between the 95 percent of GenAI pilots that produced nothing measurable and the few that did was never the model. It was whether anyone in the organization could see clearly, read skeptically, and measure honestly before money moved. You now know how to do all three on paper. This lesson converts that knowledge into a single executable quarter, because a certification you have read is a credential, while a quarter you have run is a career.

The plan ships exactly four artifacts in ninety days. Memorize the list, because every week of the plan exists to produce one of them:

• Artifact 1: the shadow-AI inventory. The amnesty survey and triage grid from Chapter 1, run for real, telling you what is actually in use in your organization today.
• Artifact 2: the readiness baseline memo. The 20-question baseline from earlier in this chapter, scored across people, process, data, and governance, with one line of evidence per question.
• Artifact 3: one skeptical vendor review. Chapter 4's toolkit (the demo-conditions checklist, the 20 due-diligence questions, the reprice worksheet) applied to one live or upcoming vendor conversation.
• Artifact 4: one process nominated and triaged for a future pilot. Not a pilot. A nomination: one process with a real baseline, a sketch of its baseline pack, and draft success and kill criteria.

Now the honest budget, because plans that lie about effort die in week three. This quarter runs on evenings and edges: roughly 3 to 5 hours a week, front-loaded. The heavy fortnights are the first two weeks (designing and running the amnesty survey takes drafting, sponsor conversations, and a follow-up push) and weeks five and six (walking standard operating procedures, the SOPs, against how work actually happens takes hours you cannot fake). Later weeks are lighter: the vendor review rides on a meeting that was going to happen anyway, and the final presentation is assembled from documents you already shipped. Call it 45 to 60 hours across the quarter, total. That is one work-week and change, spread thin, exchanged for a professional identity. Few trades in your career will price this well.

One design decision explains the whole plan: none of the four artifacts requires anyone's signoff to produce. You will inform, invite, and present, but you will never need to ask. That is not an accident of scope; it is the strategy. The moment a plan requires permission, it inherits someone else's calendar and someone else's risk appetite. This one inherits only your evenings.

Days 1 to 30: See Clearly

The first arc produces the two documents that make you the best-informed person in the building on the only question that matters at the start: what is actually happening here?

Weeks 1 and 2: the two inventories

Start with the shadow-AI amnesty survey exactly as Chapter 1 specced it: a written no-consequences commitment from the most senior sponsor you can reach, six questions ending with the champion detector, and a triage grid waiting for the results. In parallel, open the governance lesson's AI system inventory: one spreadsheet, every tool and use you can find, sanctioned or not, with owner, data touched, and decision influenced. The two instruments feed each other; the survey fills rows the official records will never show you.

Do one more thing in week one, and do not skip it because it feels premature: book the week-12 meeting now. Thirty minutes with your manager, or the steering group if you can reach one, titled something plain like "AI readiness: where we actually stand." You will meet the reason for this move in the failure-modes section, but the short version is that an artifact nobody sees is a hobby, and the calendar invite is the cheapest insurance ever sold against a wasted quarter.

Weeks 3 and 4: the baseline, scored with evidence

Run the 20-question readiness baseline from earlier in this chapter, and hold yourself to its evidence rule: a point requires a document, a report, or a named owner you can point to, not a feeling. Most organizations that score themselves honestly for the first time land lower than anyone in leadership would guess, and that gap is not bad news for you; it is your subject matter. Write the result up as a two-page memo: score, top three gaps, one line of evidence per question. Date it. A baseline is only a baseline if it is timestamped.

Artifacts shipped: the shadow-AI inventory and the readiness baseline memo. The political frame for this arc: you are not asking for anything yet. No budget request, no tool proposal, no reorganization. You are producing the document nobody else has, and in most organizations nobody else has it because producing it looks like work and promises no glory. That is precisely why it is available to you. When 63 percent of organizations lack or are unsure of AI-ready data practices, the person holding an evidenced inventory and an honest score is rare by arithmetic, not by talent.

Days 31 to 60: Prove the Lens

The second arc moves you from cartographer to analyst. You have shown what exists; now you show what the readiness lens does when pointed at a decision.

Weeks 5 and 6: the afternoon triage and the one process

Run the afternoon triage from the process-readiness lesson across 8 to 12 candidate processes: stability, documentation, measurability, scored quickly and honestly. Then do the part that takes real hours: for the two or three leaders, walk the SOP against how the work actually happens. Expect the divergences the process-readiness lesson promised you; finding them is the point, not a setback. Your goal is to leave week six holding exactly ONE process with a real baseline: known volume, known cycle time, known error rate, known cost per unit, or a credible two-week path to capturing them. One process, deliberately chosen, beats a portfolio of maybes, because everything in Level 2 and Level 3 builds on a single well-baselined candidate.

Weeks 7 and 8: the skeptical vendor review

Find one live or upcoming vendor conversation. Almost every organization has one somewhere: a renewal, a pitch that landed in a VP's inbox, a demo already scheduled. Attach yourself to it, politely and with your toolkit visible. Run Chapter 4's full sequence: the demo-conditions checklist during the demo, the 20 due-diligence questions sent in advance, the reprice worksheet on whatever business case the vendor presents. Write the output as a two-page vendor review memo: what was claimed, what survived the checklist, what the repriced case looks like, and what decision the evidence supports. Sometimes the memo will say "proceed, with these conditions." Often it will deflate a business case. Either answer is a win, because the memo's real product is the demonstration that claims in this building now get checked.

Artifact shipped: the vendor review memo, or its sharper sibling, the deflated business case. The political frame for this arc: the first time leadership watches a vendor claim get repriced in front of them, your role in the room changes. You stop being the person who takes notes and become the person the CFO glances at when a number sounds too smooth. No title change announces this. It simply starts happening, and it is worth more than the title.

Days 61 to 90: Name the Path

The third arc converts a quarter of diagnosis into a direction, and delivers it to people who can act on it.

Weeks 9 and 10: the pilot nomination

Take your one triaged process and build its nomination pack: a one-page description of the process and its baseline numbers, a sketch of the full baseline pack you would capture before any pilot (the instrument you will build properly in Level 2), and, most importantly, draft success and kill criteria written in value metrics. "Cycle time down 20 percent within 12 weeks or we stop" is a sentence almost nobody in your organization has ever written before a pilot, and writing it now, before any vendor or tool is chosen, previews the stage-gate discipline of the later levels. You are not launching anything. You are demonstrating what a launch-worthy candidate looks like.

Weeks 11 and 12: the five slides

Build and deliver the synthesis presentation in the meeting you booked in week one. Five slides, no more: where we are (the baseline score and top three gaps), what is running unofficially (the shadow-AI inventory's headline findings), what is ready (your nominated process and its numbers), what is not (the gaps that would sink a pilot today), and what I recommend next quarter (a specific, small, evidence-first step). Twenty minutes of content, ten of discussion. You are not pitching a project; you are presenting a diagnosis, and diagnoses are much harder to argue with than proposals because they contain no ask to refuse.

Week 13: the retro and the Level 2 decision

Spend one evening on a personal retrospective: which instruments felt natural, which felt forced, what you would resequence, and what response the five slides produced. Then make the Level 2 decision deliberately. If the quarter worked, Level 2 is where the instruments get sharper. If it stalled, the retro tells you whether the block was the organization or the sequencing, and both are fixable.

Artifact shipped: the next-quarter recommendation, delivered, not filed. And a bookkeeping note that matters: the Level 1 capstone, the Pilot Autopsy and Readiness Baseline memo, is assembled almost entirely from this quarter's artifacts. The baseline memo, the inventory findings, and the vendor review are its raw material. Run the ninety days and the capstone becomes an editing job, not a writing job.

The Artifact: The 90-Day On-Ramp Canvas

This lesson's named artifact is the whole plan on one page. Draw it as a grid, three rows for the three arcs, five columns, and pin it where you will see it on the evenings you are tempted to skip.

And the week-12 deck, fixed at five slides so it cannot bloat into a project pitch:

1. Where we are: the baseline score, the top three gaps, one evidence line each.
2. What is running unofficially: the shadow-AI headline: how many tools, which quadrants, the one champion worth knowing about.
3. What is ready: the nominated process, its baseline numbers, why it survived triage.
4. What is not: the two or three gaps that would sink a pilot launched today, stated without drama.
5. What I recommend next quarter: one small, evidence-first step with draft success and kill criteria already attached.

The Three Traps, and What Not to Do at All

Ninety-day plans rarely die of difficulty. They die of three specific traps, each with a countermove you install in advance.

Trap one: scope creep into fixing things. Somewhere around week five, the SOP walks will expose a broken handoff or an ugly spreadsheet, and every instinct you have as a process professional will scream to fix it. Resist. This quarter is diagnosis. The moment you start repairing processes, you burn the hours budget, you acquire stakeholders, and you convert a clean diagnostic quarter into an unscoped improvement project with your name on it. Note every fixable thing in a parking-lot list; the list itself becomes slide-five material.

Trap two: the permission trap. The quiet voice that says "I should probably clear this with someone first" will visit you in week one and again before the vendor review. Remember the design: none of the four artifacts needs signoff, and that is deliberate. A survey with an amnesty sponsor, a spreadsheet, a scored questionnaire, a memo about a meeting you attended, and five slides in a meeting you booked: nothing on that list requires a committee. People who wait for a mandate to do diagnostic work wait, on average, forever, because nobody hands out mandates for work they do not yet know they need.

Trap three: the visibility trap. The subtlest failure is running the quarter perfectly and shipping four artifacts into a folder nobody opens. Diagnosis that nobody sees changes nothing, including your standing. The countermove was installed in week one: the week-12 calendar invite, booked before you had anything to show. That invite converts the whole quarter into a deadline, and deadlines are how evenings-and-edges work actually finishes.

Alongside the traps, three prohibitions. During these ninety days you buy nothing, you pilot nothing, and you promise nothing. No tool purchase, however small; the quarter's credibility rests on having no vendor's flag in your pocket. No pilot, however tempting the champion's workflow from the amnesty survey looks; a pilot without the baseline pack and stage gates of the later levels is an application for membership in the 95 percent. And no promises of savings, headcount, or transformation; you are building a reputation for evidence, and evidence people do not advertise results they have not measured. The organizations that scrapped most of their AI initiatives in 2025, 42 percent of them by S&P Global's count, almost all started by buying, piloting, or promising before diagnosing. The discipline that separates you from that record starts with sequencing.

Buy nothing, pilot nothing, promise nothing for ninety days: this quarter you produce evidence, and evidence needs no one's permission.

Ninety Days in the Life: The Tomas Herrera Diary

Here is the plan run start to finish, as a composite, fully hypothetical case with illustrative numbers. Tomas Herrera is an operations coordinator at Velarra Distribution, a fictional 250-person wholesale firm. No title, no budget, no team. He finishes Level 1 on a Sunday and starts Monday.

Week 1. Tomas drafts the amnesty survey in two evenings and books the week-12 meeting with the COO's operations review, twelve weeks out, before he has a single result. It feels presumptuous. He books it anyway. Hours this week: five.

Week 2. The survey goes out and lands badly: 12 percent response, all of it bland. Rereading Chapter 1, he sees the miss: he sent it under his own name with no amnesty commitment, so it read as an audit. He asks the COO for one written sentence, "no answer to this survey will be used against anyone, and nothing already in use will be shut off without a replacement," and resends under her signature. Response climbs to 41 percent by Friday. The inventory lands at 23 distinct uses, including a credit-control clerk who has quietly cut dispute-letter drafting from 50 minutes to 15 using a consumer chatbot, on customer data that should never have left the building. Both facts go in the grid: one champion, one red-quadrant risk with a sanctioned-replacement note.

Weeks 3 and 4. The 20-question baseline scores 7 out of 20. Data readiness is the crater: three customer masters, no error-rate measurement anywhere outside finance. The memo runs two pages, dated, with the top three gaps named. He shares it with his manager with no ask attached, exactly as specced. His manager forwards it to the COO unprompted, which Tomas learns about later. Cumulative hours: 19.

Weeks 5 and 6. The afternoon triage covers ten processes. Two SOP walks produce the usual archaeology: the returns-approval SOP describes a three-step flow that is actually seven steps and two workarounds in practice. The winner is customer credit-note processing: 900 a month, a real cycle-time log in the ERP (enterprise resource planning system), a measured 6 percent rework rate, and a defensible $11 cost per credit note. One process, real numbers. Parking-lot list: nine items he does not fix.

Weeks 7 and 8. A vendor is already pitching Velarra's finance director an AI document-processing suite: $220,000 over two years, "six-month payback." Tomas asks to sit in, sends the 20 questions ahead, and runs the demo-conditions checklist live. The demo used the vendor's own clean sample invoices; on Velarra's documents, extraction needs review on roughly a quarter of pages. The reprice worksheet adds verification labor, integration, and a realistic adoption curve, and the six-month payback reprices to roughly 26 months, before the cost of the data cleanup the baseline memo already flagged. His memo recommends postponing the purchase and counter-proposing a four-week paid trial on Velarra's own documents, with the credit-note baseline captured first. The finance director takes the recommendation. Nobody says "good catch" in the meeting. The finance director does, by email, that evening.

Weeks 9 and 10. The nomination pack: credit-note processing, its four baseline numbers, a sketch of the fuller baseline pack, and draft criteria: "rework rate from 6 percent to under 3 percent and cycle time down 30 percent within 12 weeks, or we stop and write up why." Two evenings of work, mostly assembly.

Weeks 11 and 12. Five slides, twenty minutes, the meeting booked eleven weeks earlier. Slide two (23 unofficial AI uses, one champion, one closed risk) gets the most questions. Slide five recommends one thing: baseline the credit-note process properly next quarter and evaluate the postponed vendor's trial against it. The COO asks, "What do you need to keep going?" Tomas, who has read this lesson, does not improvise a budget request. He says: "Four hours a week of sanctioned time, and access to the ERP reporting module." He gets both, in the meeting. Total hours across the quarter: about 54.

Week 13. The retro is one page. What worked: the week-1 invite, the amnesty resend, refusing to fix the returns SOP. What he would change: start the SOP walks a week earlier. Decision: Level 2, starting Monday.

Stand back from the diary and mark where you now are, because it is where Tomas stood on his first Monday. You can see clearly: inventory what actually runs, official and shadow. You can read skeptically: no demo, benchmark, or payback claim reaches you unpriced again. And you can score honestly: baseline before belief, evidence before enthusiasm, a documented kill worth more than a zombie pilot. That is Level 1, complete. Level 2 adds the force multiplier: AI itself as your own assessment instrument, starting with the next lesson on prompting for process professionals, where the tools you have been auditing start working for the auditor.

What to Do Monday Morning

The on-ramp starts with five moves, all inside week one.

1. Print the canvas and date it. Write day 1 as next Monday and day 90 on the calendar. A plan without dates is a mood.
2. Book the week-12 meeting today. Thirty minutes with your manager or steering group, twelve weeks out, titled "AI readiness: where we actually stand." Booked before you have results, because the invite is the countermove to the visibility trap.
3. Draft the amnesty survey and identify your sponsor. Six questions from Chapter 1, plus the one written no-consequences sentence from the most senior name you can get. No sponsor sentence, no send.
4. Open the AI system inventory spreadsheet. Four columns, first ten rows from what you already know. Do not wait for the survey to start the file.
5. Write the three prohibitions where you will see them. Buy nothing, pilot nothing, promise nothing until day 91. When the week-5 urge to fix things arrives, add the item to the parking lot and keep walking.

Key Takeaways

• Run the quarter as three 30-day arcs: see clearly (both inventories plus the evidenced 20-question baseline), prove the lens (the afternoon triage and one skeptical vendor review), and name the path (the pilot nomination and the five-slide synthesis).
• Ship exactly four artifacts: the shadow-AI inventory, the dated readiness baseline memo, the vendor review memo, and the next-quarter recommendation; the L1 capstone is assembled from these, so the quarter writes your capstone for you.
• Budget the effort honestly: 3 to 5 hours a week from evenings and edges, roughly 45 to 60 hours total, front-loaded into survey design and the SOP walks.
• Book the week-12 presentation in week 1, before any results exist, because artifacts nobody sees change nothing and a deadline is what finishes evenings-and-edges work.
• Refuse the three traps by design: no fixing during diagnosis (parking-lot list instead), no waiting for permission (none of the four artifacts needs signoff, deliberately), no invisible shipping (the calendar invite is the countermove).
• Hold the three prohibitions for all ninety days: buy nothing, pilot nothing, promise nothing; the 42 percent of companies that scrapped their AI initiatives mostly sequenced spending before diagnosis, and your discipline starts with the order of operations.
• Pick ONE process out of the 8-to-12 triage and leave the quarter holding its real baseline numbers, because everything in Level 2 and Level 3 builds on a single well-chosen, well-measured candidate.
• Treat the COO's "what do you need to keep going" as the quarter's engineered outcome, and answer it with time and access, not budget: the smallest yes that keeps the evidence engine running.