The last session of a mid-sized industry conference, two hundred operations and supply-chain people in the room, and a panel of four: a vendor's VP of product, a consultancy partner, a professor with a good dataset, and a moderator. Forty minutes of confident material, none of it wrong, none of it usable. Then a woman near the back takes the microphone and asks the only question in the building that matters. "How many people here have personally taken an AI use case from assessment, through pilot, into scaled daily operation, and measured what it actually did?" Three hands go up out of two hundred, and none of them is on the stage. The panel spends its last ten minutes being interesting; the three with raised hands say nothing. This lesson is about that asymmetry, and about the fact that you are now, quite probably, one of the three.

The Room Where Nobody Has Done It

The public conversation about AI readiness is loud, well-funded, and structurally missing its most useful voice. Vendors describe their products, so they cannot publish the deployments that went badly without describing their own product going badly. Consultancies describe frameworks, useful until you need a specific answer, and abstracted beyond falsification precisely because the client cannot be named. Researchers describe aggregates, and those are excellent: MIT's roughly 95 percent of enterprise generative AI pilots producing no measurable profit-and-loss (P&L) return, with only about 5 percent of custom tools crossing into production; S&P Global's 42 percent of companies scrapping most of their AI initiatives in 2025; McKinsey's 88 percent using AI regularly against roughly 39 percent reporting any earnings impact; Gartner's forecast that over 40 percent of agentic AI projects will be canceled by the end of 2027. All true, and none of it tells you what happened at one company, on one process, in the eleven weeks that decided it.

The people who know that part mostly say nothing, for four reasons, three of them respectable. Confidentiality is real: your baselines, error rates, and vendor terms belong to your employer. Modesty is real in a specific form: the practitioner assumes their own transformation was unremarkable, messy and half luck, which is wrong, because that is what everyone's looks like. Time is real, since the person who has done the work has the least free calendar. The fourth reason is the least respectable and the most common: publishing invites scrutiny, and silence looks safe.

The result is a strange market condition. In most fields the bar for a credible contribution is high; in AI readiness it is remarkably low. A practitioner who can describe one real transformation honestly, including the part that failed and why, is immediately more useful than most of what gets published here: not more eloquent, just more useful, in the sense that a reader can do something differently on Monday. That gap will close, and whoever fills it while it is open acquires something later arrivals will work much harder for.

Why This Is a Program Asset, Not a Vanity Project

The phrase "thought leadership" has been ruined by people who had no thoughts and led nothing, so be exact. External authority does five things for the program you are running right now.

• It recruits practitioners. The scarcest input in a readiness program is not budget or models, it is people who have taken something from assessment to scaled operation, and they ignore recruiter messages while reading method pieces by someone who knows what a baseline is.
• It changes vendor behavior. A vendor negotiating with a buyer whose audience would hear about a bad experience negotiates differently: you have added a reputational term to the contract without writing one.
• It unlocks peer data. Benchmarks that do not exist commercially get shared privately between leaders who trust each other, which is how you learn whether your 14-week time-to-value is good or embarrassing.
• It gives the program air cover. A method your steering committee has heard twelve times lands differently when a peer at another company mentions having read it.
• It creates career optionality. This one is yours rather than the program's, and it is not a reason to be shy about it.

Authority built from real evidence is a program asset and a personal one at once, produced by the same act, which is why register matters. Treat it as a professional discipline, in the family of writing a good standard operating procedure (SOP), not as personal branding: the audience you want can tell within two paragraphs whether you have ever run a real baseline.

The Evidence-to-Authority Path

Here is this lesson's artifact: the Evidence-to-Authority Path, a four-stage progression from internal artifact to external contribution, with the control that makes each stage safe. The stages are ordered, and the order is the point: almost everyone who fails at this fails by starting at stage three.

Stage One: Internal Writing First

This is the stage everyone skips, which is why so much practitioner writing is thin. Four documents constitute the practice, and you already have to produce three: the decision memo (written at the moment of decision, covering the options, the evidence, what you expect, and what would tell you that you were wrong), the honest quarterly (the report that includes what did not work, deductions itemized rather than buried), the postmortem and its kill memo (written to teach rather than to defend), and the method documentation (scorecard definitions, intake criteria, baseline procedure).

Now the point most people miss: internal writing is not preparation for external writing, it is the practice itself. A leader whose internal writing is vague will produce vague external writing, because vagueness is a thinking problem rather than a style problem, and publishing does not fix a thinking problem, it broadcasts one. If your quarterly says "adoption continues to improve across the pilot cohort," you have a measurement issue wearing a phrasing costume.

The discipline is a single instruction: write the internal version as though a smart outsider will read it. Four habits carry it. Define your own jargon on first use, including internal system names. State the baseline and how it was measured every time you state a change. Say what you would have concluded if the number had gone the other way, the sentence that separates analysis from advocacy. And name the alternative explanation you cannot rule out, because volume, seasonality, and the other three concurrent programs do not stop existing because they are inconvenient.

This is not a tax: your own chief financial officer's (CFO's) analyst reads the way a hostile outsider reads. Hand the draft to a colleague two functions away and treat every question they ask as a hole.

Stage Two: The Peer Circle

The highest-value stage and the least visible: five to eight practitioners in comparable roles at organizations that do not compete, meeting on a regular cadence with genuine candor about what actually happened.

Why it works is a matter of what can be said where. A published case study cannot contain the sentence "our document-extraction pilot failed because master data had roughly 14 percent duplicate supplier records with no owner, and we did not find out until week nine." That sentence is specific enough to be actionable and specific enough to be embarrassing. The circle is the only forum where failure stories carry enough specificity to be useful; everything else you read about failure is anonymized into uselessness or aggregated into a statistic.

How to form one. Three sources, in ascending order of quality: industry associations (the working groups and benchmarking committees, not the keynote hall), alumni networks, where a prior relationship absorbs the awkwardness of the first ask, and direct outreach to people whose problems you recognize, the best source and the least used. Not a generic connection request describing yourself: reference the problem you heard them describe, offer the specific thing you have that bears on it, and propose a small bounded commitment. "You mentioned your intake process is drowning in requests you cannot triage. We built a scoring rubric for that and could walk you through what it gets wrong. Forty-five minutes, video, no agenda beyond that?"

The operating rules are few and load-bearing. No attribution: nothing leaves the room with a name or company attached, stated aloud at every session rather than assumed. Reciprocity, the rule with teeth: you get what you bring, and the person who only listens is not invited twice. A circle where two are candid and four are extracting value collapses within three meetings. A standing agenda that forces specifics, one number that moved and one thing that failed per member. And a size cap, because above roughly eight the candor falls off a cliff. Cross-industry membership is a feature, because readiness problems are process problems and the patterns transfer: non-competing is the requirement, and comparable means role and scale, not sector.

Stage Three: Selective Public Contribution

Talks, articles, panels, case studies. This is where most people start, which is why most contributions read as opinion: they are. Enterprise AI has an oversupply of opinion and an undersupply of evidence.

One principle separates a contribution from a promotion: publish the method and the failure modes, not the results. Results require the most sanitization and teach the least. Your 44 percent cycle-time reduction is a fact about your process, baseline, staffing, and flavor of legacy system, not about the world. A reader elsewhere cannot use, reproduce, or check it, and it invites the wrong question: "is that true?" rather than "how would I do that here?" Method and failure modes transfer, disclose almost nothing commercially, and last: your results change next quarter, while the reason the pilot broke will still be the reason pilots break in 2029.

Three formats carry the value.

1. The honest case study, indexed. One real transformation with figures ranged or indexed rather than absolute, and a genuine failure paragraph. That paragraph is the credibility engine, not a humility gesture: it is the part a reader cannot get from a vendor, and what makes them believe the rest.
2. The method piece that hands over an artifact. A scorecard structure, an intake rubric, a kill-criteria template. This is the strongest available signal that you did the work, because real artifacts are full of small awkward decisions nobody would invent: the exclusion rule, the tie-break, the field everyone forgets. Frameworks can be faked; templates cannot.
3. The myth-correction piece grounded in the public failure record. This program's anchors are public and any practitioner can build on them: MIT's 95 percent and the roughly 5 percent that reached production, S&P Global's 42 percent scrapped, McKinsey's 88 against 39, Gartner's 40 percent-plus agentic cancellations. What you add is the mechanism: the aggregate says pilots fail, and you can say how the failure looked from inside, the missing half of every one of those studies.

Then the credibility rule: never publish a claim you could not defend to your own CFO's analyst. The practical form is a four-part check on every public number. What was the baseline, and who signed it? Over what window? What was excluded, and why? What confounder can you not rule out? If any of the four has no answer, the number does not go in.

Publish the method early and the results late. Method improves under scrutiny; results only get more accurate with time.

Stage Four: Standing Authority

Standing authority is a specific state: you are consulted rather than pitching, and the symptoms are concrete. Practitioner recruitment goes inbound, and one good hire outweighs a year of contribution costs. Vendor negotiations change shape: the account team arrives prepared and finds flexibility on terms previously called standard. Peer organizations share benchmarks that are not purchasable at any price. And the internal program acquires legitimacy no internal communication can manufacture, because your executives hear your method described from outside the building.

It carries three obligations, which are what stop the asset becoming a liability.

• Accuracy, actively maintained. When you are wrong, correct it publicly and quickly, as visibly as the claim. A prompt correction is a credibility deposit; a quiet edit is the opposite, and people notice.
• Disclosure of interests. Advisory relationships, equity, paid travel, vendor sponsorship of the event: disclose in one plain line before anyone asks. The undisclosed interest that surfaces later destroys more credibility than the interest itself ever earned.
• Not becoming a person who talks about work rather than doing it. The real occupational hazard, and a gradual one: when the flow reverses, the asset decays invisibly, because invitations keep arriving for a year or two after the substance has gone.

The practical guardrail is a time budget, set before the invitations start, while the number is a policy rather than a negotiation with your own ego. A workable shape: two days a quarter, one speaking commitment and one written piece a quarter as a ceiling rather than a target, and a rule that external work comes out of your calendar and not the program's. Add an annual test: if your last two contributions contained no new evidence from your own work, stop until you have some.

The Confidentiality Discipline

This is the part that decides whether any of the above is possible for you. Confidentiality handling is a professional skill, closer to redacting a contract than to keeping a secret. Most practitioners treat it as binary ("can I publish this or not?") and default to no. It is not binary: four transformations turn real work into publishable material, each preserving the lesson while removing what should not leave the building.

Abstraction: describe the process class, not the process

Move up one level of description until the identifying detail dissolves but the operational lesson survives. "A high-volume exception-handling workflow in a distribution business, roughly ten thousand items a month, three handoffs, a two-person approval queue" rather than "our supplier-invoice mismatch queue in the Rotterdam distribution center." The test is directional: would a competitor gain something operationally actionable? Not "could they guess who we are," because in a small sector they usually can. Your queue design, error taxonomy, and vendor pricing are actionable; the shape of exception handling in distribution is not. Go one step above the identifying detail and no further, because done badly this produces mush.

Indexing: report ratios, not absolutes

Report changes as percentages, ratios, indexes, or ranges rather than absolute figures. "Cycle time fell 44 percent, from a baseline we can characterize as multi-day" preserves the whole lesson and discloses nothing material about volume, cost, or capacity. Set the baseline to 100 and report against it. Do the same with money and people: "a mid-six-figure annual run rate" and "a team in the low double digits" carry the scale without the disclosure. Indexing also forces you to state the baseline, because a percentage is meaningless without one.

Compositing: build the labeled archetype

Combine patterns across several engagements, use cases, or organizations into one archetypal account, labeled clearly as a composite. This is the technique most consultancies use constantly and explain almost never. It lets you tell the story that is true of five situations without exposing any one, and it often teaches better than a single case. The rule is the label: "the following is a composite drawn from four deployments across three organizations, with numbers representative rather than actual" costs one sentence and buys the whole technique. An unlabeled composite is a fabrication; a labeled one is a teaching instrument.

Clearance: ask early, and ask for a standing agreement

The unglamorous one, and the one that actually blocks people. Your employer's communications and legal functions decide what you may say publicly about your work, and most practitioners never ask, assume the answer is no, and are wrong: it is frequently "yes, with conditions."

Two rules make it workable. First, ask early rather than after drafting: a communications lead handed a finished article they did not know was coming will find problems with it, while the same person consulted at outline stage becomes a collaborator who tells you which claims are fine before you spend a weekend on the ones that are not. Second, and more valuable: negotiate a standing agreement while nothing is pending, covering four things. Which topics are pre-cleared (typically method, general practice, commentary on public research). Which always require review (anything with company figures, anything naming a vendor, anything touching a customer). Who reviews, and on what turnaround. With nothing on the table, that is a ten-minute conversation about principles; three weeks before a conference, it is a negotiation about your specific slides, and you will lose it.

The sector caution: ask, do not guess

Some sectors are not ordinary, and there the mistake is not embarrassing but dangerous. In a listed company, material non-public information (MNPI: information a reasonable investor would consider important that has not been publicly disclosed) is governed by securities law, and selective disclosure of it is a regulatory event, not a faux pas. A cost-reduction figure large enough to move an earnings line can qualify, and quiet periods narrow it further. Regulated industries add layers: patient information, customer data, export-controlled material. The rule is the opposite of what confident people do: ask rather than guess. A five-minute question to your general counsel is answered from experience; a guess is answered by an investigation.

Two Years at Norvik: A Worked Example

Norvik Group is the 2,400-person industrial distributor this level has followed: a scrapped year of ungoverned pilots, then a governed rebuild, now a program with a quarterly readiness scorecard and a value roll-up that survives contact with finance. The following two years of its transformation director's work are illustrative, with hypothetical figures.

Months 1 to 6: the internal writing gets sharper. She changes nothing about her external presence and everything about her documents. The quarterly report is rewritten to an outsider's standard: system names defined, every stated change carrying its baseline and measurement window, a paragraph per section headed "what would have changed our conclusion," and a standing line naming the confounders she cannot rule out. Two kill memos are circulated, on the weekly reporting assistant and a supplier-classification tool that never beat the existing rules engine, both written to be read by someone who was not in the room, so they explain the decision instead of defending it. None of it is external, and all of it becomes the raw material for everything else.

Months 7 to 12: the peer circle. She forms a circle of six, two from an industry association benchmarking subcommittee and four from direct approaches to people whose problems she recognized, spanning six sectors. Quarterly, two hours, video, no slides, no attribution, one number that moved and one thing that failed per member.

The most valuable exchange in two years happens in the third meeting, and it is somebody else's failure. A member describes a customer-service triage agent that had quietly been reclassifying roughly 9 percent of incoming cases into a queue nobody monitored, for six weeks, found only when a customer escalated. No dashboard showed it, because the agent's accuracy metric counted a classification as successful whenever it was made confidently. Norvik has an agent pilot in intake at that moment, and it ships with monitoring on queue-destination distribution, a weekly human sample of reclassified cases, and a scope cut, none of which were in the original brief. That anecdote, unpublishable anywhere in that form, is worth more than the year's conference budget.

Months 13 to 18: the first public contribution, deliberately a method piece. She publishes the structure of Norvik's readiness scorecard: the scoring rubric, the evidence standard, and a template readers can use. It carries the public anchors (MIT's 95 percent, S&P's 42 percent, Gartner's agentic findings) as the argument for why the scoring is severe, and contains, deliberately, zero Norvik results. Clearance takes nine days rather than nine weeks, under the standing agreement negotiated four months earlier when nothing was pending.

Months 19 to 24: the talk, built on the kills. A conference session titled around the two documented kills rather than the wins, with figures indexed to a baseline of 100, the process described one level up ("a mid-volume classification workflow in industrial distribution"), and one slide showing the deductions that reduced the naive value claim. The questions afterwards run forty minutes long, which is the tell that the material was real.

Outcomes at two years, all illustrative. Three inbound practitioner candidates, one hired into a process-lead role that had been open five months. One vendor negotiation in which the account team arrives visibly better prepared, brings two same-sector references unprompted, and concedes a support term previously called non-negotiable. Two peer organizations sharing benchmarks Norvik could not have purchased. One advisory approach from a private equity operating partner, declined, and valued because knowing it exists changes how the next conversation about her scope feels. Total time cost: roughly eleven days across two years.

The Failure Story: The Premature Case Study

Now the other path, illustrative and common. An ambitious program lead at a 1,100-person specialty manufacturer publishes a detailed case study at month seven: well written, confident, four named use cases, real absolute figures, a claim of 2.3 million in annualized savings.

Three things are wrong with it and none is visible on publication day. The number is the naive roll-up: shared enabling benefits counted in more than one use case, no counterfactual, no persistence check. The controls this level's measurement lesson teaches would have cut it by a fifth or more before anyone outside saw it. Second, the program is seven months old, younger than the evidence requires, and over the following year two of the four described use cases are killed, correctly. Third, nobody in finance saw the figures before publication.

The bill arrives in three installments. A competitor's analyst cites the piece the following year in a note about buyers overstating early results, and it is now the first thing that surfaces when anyone searches his name beside his company's. The two kills become known in the sector the way these things do, so the people who read the case study first hear about the kills second, which is the worst possible order. And finance responds to having its numbers used without clearance in the only rational way available: all future external material routes through a review that adds six weeks, a control that is reasonable, permanent, and fatal to every timely method piece he might have written.

His credibility takes the hit in both directions at once: externally with an audience that later learned what happened, internally with the function whose numbers he used without asking. Note what he did not do wrong: he was not dishonest, not lazy, and the program was fine. He published the results early and the method never, which is precisely backwards.

The Career Economics, Honestly

Level 1 argued that this work is hired on evidence of having done it rather than on credentials, and that the artifact portfolio is what gets people hired. External contribution makes that portfolio visible. A portfolio nobody can see is something you describe in an interview, competing with everyone else's description; one that is partly public arrives before you do, already evaluated by someone who has read your method piece and formed a view of how you think.

The second effect is optionality. Senior readiness practitioners increasingly have paths that were not a category five years ago: fractional transformation leadership, board and investor advisory, expert-network work, and program-design engagements for organizations that need someone who has done it. Those paths run on reputation, because there is no credential for them. Optionality has value even unexercised: knowing that three organizations would talk to you tomorrow changes the posture you bring to an internal conversation about scope or budget.

The third property requires honesty, because it sells badly. This asset compounds slowly. A realistic timeline: a functioning peer circle within a few months of deciding to build one, since that is mostly a matter of asking; a first genuinely useful public contribution within about a year; and standing authority over three to five years, built from a few contributions actually worth reading. There is no version that runs faster, and anyone selling a faster path is selling something. Volume tactics generate attention, and that attention does none of the five things this lesson opened with.

Which is why starting during a transformation rather than after it matters. The evidence is being generated right now, in documents you already have to write, and the details fade fast: within eighteen months you will remember that the pilot failed and will have lost the reason, the sequence, and the number. Stage one is a preservation mechanism as much as a communication one. And the last caution follows: authority is downstream of results, and the order does not reverse. The practitioner who becomes known for talking while their own program stalls has traded a real asset for a counterfeit one, in a market small enough that everyone eventually finds out. Authority rests on a clear and defensible view of where organizations actually are, which is what the next lesson supplies: the readiness maturity model, and the places where organizations reliably get stuck.

What to Do Monday Morning

Five moves, in order.

1. Write your next internal document as though a smart outsider will read it. Take whatever is already due (a quarterly, a decision memo, a postmortem) and apply the four habits: define your jargon, state every baseline and measurement window, add the paragraph saying what would have changed your conclusion, and name the confounder you cannot rule out. Then hand it to a colleague two functions away and write down every question they ask.
2. Approach three practitioners in comparable roles at non-competing organizations this month. Reference their specific problem, offer your specific artifact, propose a bounded forty-five minutes. Three approaches typically produce one or two relationships, which is how a circle of six gets built.
3. Negotiate the standing clearance agreement with communications while nothing is pending. Ask for the four elements: pre-cleared topics, always-reviewed topics, the named reviewer, and a turnaround commitment. Do it in a week when you have nothing to publish, which makes it a conversation about principles.
4. Draft one method piece that hands over an artifact rather than a result. Your scorecard structure, intake rubric, or kill-criteria template: publish the thing a reader can use, with zero company figures, applying abstraction and indexing to any example.
5. Set your time budget before the invitations start arriving. Write down the days per quarter, the ceiling on speaking and writing commitments, and the rule that this comes out of your calendar, not the program's. A budget set in advance is a policy; set later, it is a negotiation you will lose with yourself.

Key Takeaways

• Recognize the field's evidence problem as your opening: vendors describe products, consultancies describe frameworks, researchers describe aggregates (MIT's 95 percent and roughly 5 percent, S&P Global's 42 percent, McKinsey's 88 versus 39, Gartner's 40 percent-plus agentic cancellations), and the practitioners who scaled something say nothing.
• Treat external authority as a program asset, not an indulgence: it recruits scarce practitioners, changes vendor behavior, unlocks peer benchmark data, gives the program air cover, and creates career optionality at once.
• Walk the Evidence-to-Authority Path in order (internal writing, peer circle, selective public contribution, standing authority), because starting at stage three turns contributions into opinion.
• Write every internal document to an outsider's standard: jargon defined, baselines and windows stated, what would have changed your conclusion said, the confounder you cannot rule out named.
• Build a peer circle of five to eight practitioners at non-competing organizations under a no-attribution rule and enforced reciprocity, the only forum where failure stories carry enough detail to change a decision.
• Publish the method and the failure modes rather than the results, using the indexed case study, the artifact-bearing method piece, or the myth-correction piece built on the public failure record, and never a claim you could not defend to your own CFO's analyst.
• Apply the four confidentiality transformations (abstraction one level above the identifying detail, indexing to a baseline of 100, labeled compositing, clearance under a standing agreement negotiated early), and ask rather than guess about material non-public information.
• Accept the honest timeline: a peer circle in months, a first useful contribution within a year, standing authority over three to five years, held in place by a time budget set in advance and by the rule that authority is downstream of results, never the reverse.