The slides went well. Forty minutes on wave three: three processes redesigned, the exception backlog finally moving, two new roles posted internally, a chart heading the right way. Two hundred and forty people from operations sat through it politely in a canteen with the tables pushed back. Then the floor opens, the third question comes from a shift supervisor with nineteen years in the building, and it is not about the chart. "Just tell us straight," she says. "How many of us are still here in three years?" The transformation director has three answers ready. The warm one: nobody's job is at risk. The smooth one: roles will evolve, they always have. The deferral: that is a longer conversation, let's take it offline. Every one of them gets her out of the room, and every one costs more than it saves. What she says instead is the subject of this lesson, and it is the hardest sentence-level work a transformation leader ever does.

Three Temptations, and Why All Three Are Traps

You will stand in front of people whose work is changing because of a program you lead. Not once: repeatedly, for years, in canteens and team meetings and skip-level one-to-ones. What you say will be repeated, screenshotted, and remembered with a fidelity that will astonish you. And the honest answer is frequently uncertain, partly outside your control, and different for every role in the room. Under that pressure, three answers present themselves. They feel different. They fail in the same place.

Reassurance beyond the evidence. "No one's job is at risk here." Warm, sincere, and almost always a promise about the future made by someone who does not control the future: headcount at enterprise scale is shaped by demand, margin, acquisitions, a new chief executive, a bad half-year in a market you do not run. Reassurance buys quiet now by writing a cheque on an account you do not own, and credibility is not a balance you rebuild with later good behaviour: it is a switch, flipped permanently by the first contradiction.

Vagueness. "Roles will evolve. They always have." Peculiar, because it is often the most factually accurate thing available. The problem is not content, it is reception. Delivered to someone who has just asked a specific question about their own livelihood, an abstraction reads as evasion, and an audience that detects evasion concludes "she knows something and will not say it," then fills the silence itself, never with the median scenario but with the worst plausible story.

Deferral. "We'll cross that bridge when we come to it." The most quietly corrosive of the three, because of what it does with the asymmetry in the room. You have the information; they have the exposure. Deferral hands the anxiety back to the exposed while the information stays with you. The leader sleeps fine, and two hundred and forty people update their assumptions about how much they will be told, and when.

The honest alternative is harder to deliver and better in every direction. It rests on a distinction most leaders never make out loud: the difference between what is decided, what is undecided, and what is unknowable. A decided thing gets stated as a commitment. An undecided thing gets named, with the date it will be decided and the date you will communicate it. An unknowable thing gets called unknowable, plainly, alongside what you will do regardless of how it resolves.

You can only promise what you control. Everything else you can describe, date, or name as unknown, and doing all three plainly buys more trust than any reassurance you are not entitled to give.

The Artifact: The Workforce Commitment Statement

The artifact is one document, written once at leadership level, delivered by every manager, and updated on a published cadence. Call it the Workforce Commitment Statement. It is short: one to two pages, and anything longer is a legal document pretending to be a conversation. It has four parts, and the order matters, because it moves from the most verifiable ground to the most valuable.

Part one: What we know

The factual position, and nothing else: the processes actually changed, what happened to the composition of the work in the roles that touch them, and where the freed capacity went. Level 4's role redesign work produced task composition data for each affected role, from queue instrumentation rather than a survey. In most operations deployments it shows a shift rather than a subtraction: routine reading, categorizing and keying falls, while verification, exception judgment and escalation handling rise. Say exactly that, with your own numbers, for your own roles.

The discipline that makes part one honest is a rule about what you leave out. Describe your organization's evidence, not the industry's forecasts. Predictions about what AI will do to employment are plentiful, they disagree violently, and the moment you cite one you have adopted it: you will be judged by it for years by an audience who remembers the number and not the caveat. Your own data is more credible, because people can check it against their own week, and more relevant, because it describes the thing actually happening to them. The test: every sentence in part one should survive "where does that number come from?"

Part two: What we have decided

Four questions belong here, and they are the four your workforce is actually asking behind whatever they say out loud:

• Which processes are in scope this year. Named, with their functions. Vagueness here reads as a list you are hiding.
• What happens to the hours saved. Hours redeployed to a named destination (a backlog, a service level, a quality function) is a completely different fact from hours banked as savings, and people know the difference.
• Whether there are headcount targets attached, and if so where. If there are none, that is a powerful fact and you should state it. If there are, this is where they get named.
• The redeployment and reskilling commitments, with their funding. An unfunded training promise is an aspiration and everyone recognises one on sight.

The discipline for part two is what makes the section worth reading: only things actually decided and actually funded go in. A decision stated here becomes a promise, whatever qualifier you attach. The test is temporal: if this sentence were read back to you in twelve months in front of the same audience, would it still be true? If the answer is "probably, if things go well," it belongs in part three.

Part three: What we have not decided, and when we will

The section almost every organization omits, and the one that buys the most trust per line. It is omitted because it feels like advertising your own uncertainty. What it actually advertises is that you distinguish not knowing from not telling, which is the distinction your audience most wants evidence of.

The format is an open question with a decision date and a communication date attached. Not "we are still working through the wave three implications," but: "Wave three covers customer operations. We have not decided its staffing model. That decision goes to the executive committee on the eleventh of November, and we will communicate the outcome within one week, to affected teams first."

That does not reduce the uncertainty by one gram, but it gives it edges: the supervisor knows what is open, who decides it, when, and when she will hear. People can plan around a date. They cannot plan around silence, so they plan around the worst case instead, which is what makes the good people update their resumes first.

The obligation attached to a dated unknown is absolute. You communicate on that date, even if the answer is another delay. "The committee did not decide on the eleventh; it is now the ninth of December and I will come back to you that week" is a kept commitment, mildly annoying and fully trusted. Silence on the eleventh is a broken one, and it converts every other date in the document into decoration.

Part four: What we commit to regardless

The standing commitments that hold whatever the decisions turn out to be. This is the most valuable content in the statement, for a structural reason: everything in parts one to three depends on facts that can move, and part four depends only on whether you keep your word.

• Notice. If a role is reduced, the notice period and consultation process that will apply, stated in advance rather than discovered at the moment of impact.
• Reskilling access. Who is entitled to what training, funded, during working hours. Training people are expected to complete on their own evenings is a message about how much you meant it.
• Internal-first hiring. New roles the program creates (practitioner, verification, exception specialist, model quality) posted internally first, with a stated window and support through the application.
• Transition support. What applies if a role does shrink: redeployment search, outplacement, retraining allowance, the named person who owns each case.
• The manager-first rule. Anyone materially affected hears it from their own manager, in person, before any public announcement, all-hands, or intranet post. A process commitment rather than an outcome commitment, and the one people cite back to you years later.

Two rules govern part four. First, every line is checked with human resources (HR) and, where relevant, legal and employee representatives, before you say it out loud. A leader who invents a notice commitment in a canteen has created an obligation the organization may not honour: reassurance in better clothes. Second, a single breach ends the statement's usefulness. Not damages: ends. An absolute with one documented exception is a preference. So the discipline happens before publication: put in only what you will honour on the worst day of the program, when it is expensive and nobody would notice if you quietly did not.

Delivering It Without Losing It

A statement that never survives contact with two hundred managers is a governance artifact, not a communication. Four mechanics decide the difference.

Cascade with fidelity

The statement is written once, centrally, and delivered by line managers, because people believe their own manager and treat broadcast messages as corporate weather. That design creates the program's largest source of message variance, and the variance comes from good managers, not bad ones.

A caring manager, watching a team member's face fall, softens: "Honestly, I don't think any of this will really affect us here." An anxious one, wanting to be trusted with the inside story, embellishes: "Between us, I hear wave four is going to be brutal." Both are trying to help, and both have created a version of the truth that collides with the official one within a fortnight, at which point the person concludes somebody lied. They cannot tell which, so they discount both.

The fix is not a script; scripts produce managers reading aloud in a voice their team has never heard, which is its own credibility event. The fix is preparation: brief managers before the cascade, in a room, with the reasoning behind each section rather than just its wording; give them written answers to the three hardest questions, agreed word-for-word so two managers cannot give different answers in the same week; and give them explicit permission to say "I don't know, and I will find out by Friday." Most improvisation happens because a manager believes not having an answer is a personal failure. Remove that belief, then make the permission real with a named escalation route and a two-day response commitment, because a manager who promises Friday and gets nothing never risks it again.

Cadence over events

A communication that happens once is an announcement. Trust is built by returning on a schedule, especially when there is nothing dramatic to say. Publish the cadence inside the statement: quarterly minimum, monthly for functions in an active wave. Then hold it, including the part that takes discipline: report the quarters where nothing changed. "Since October: wave two stable, no scope changes, the November decision still scheduled, no changes to any part four commitment" is a boring update and an enormously valuable one.

The failure mode is common: programs go quiet in exactly the periods when things are uncertain, because leaders wait for something solid to say. From the floor that pattern is legible within two cycles, and after it the absence of an update becomes the update.

The individual conversation

All of the above is machinery, and it exists to make one thing possible: a manager and one person, in a room, talking about that person's actual job. Level 4 taught those mechanics. The Role Change Brief carries the task composition, the responsibilities added and removed with their standards, and what is decided and undecided about level and band. The conversation names the loss where there is one, and it ends with the person's own question rather than the manager's summary.

The enterprise addition is a governance point, not a coaching one. Your job at this altitude is to ensure those conversations happen, which means checking rather than assuming. A cascade reaching ninety percent of managers and eighty percent of teams looks complete on a status slide and leaves several hundred people who got nothing: disproportionately those on shift patterns, at satellite sites, on parental leave, or reporting to the manager who was mid-restructure, which is to say exactly where trust is thinnest. So check it as you would any control: sample fifteen people at random and ask three questions (did you have a conversation about how your role changes, do you know what has not been decided, do you know when you will hear next). Ask employees, not managers, because managers report intent and employees report receipt. Then report coverage to the steering committee, since a commitment nobody measures is a preference with a nice font.

The leader's own position

A short section, and a serious one, because some readers are in it right now. Sometimes leadership's actual plan does include reductions. The honest path is to say so, within the constraints that genuinely bind: a listed company cannot pre-announce material information, a statutory consultation has a legally defined sequence, an unsigned transaction cannot be discussed. When a real constraint applies, name its existence without its content: "There is a decision in progress that I am not able to discuss yet, for reasons about regulation rather than about you, and here is the date by which I can say more." Uncomfortable, and far better than the alternatives, because it tells the truth about the shape of the situation while withholding only what must be withheld.

What you must never do, under any pressure, is allow a program that is about headcount to be described as being about augmentation. That substitution is the origin of most workforce cynicism you will meet, and it is unrecoverable: when the reductions arrive, everyone who was told the program was about freeing them from routine work learns at once that they were handled.

And the case this program cannot resolve for you: a leader asked to deliver a message they believe to be false is facing a professional and ethical decision, not a communications problem, not a framing exercise, and not something a better artifact fixes. The options are the ones you already know: refuse, escalate, insist on language you can stand behind, or decline to be the face of it. Each has a cost, and some are borne by you and your family. Nobody can make that decision for you, and you deserve to see it named plainly rather than dressed as a message-discipline challenge.

The Statement in Full: Wave Three

Here is the artifact assembled, for the enterprise program you have followed through this level, at the start of wave three. Every number is hypothetical; the shape, order and specificity are what you copy.

Workforce Commitment Statement, October, Operations and Customer Functions

1. What we know. Three processes have been redesigned with AI in the workflow: invoice exceptions, vendor inquiry handling, and claims triage. For invoice exceptions clerks, routine reading, categorizing and keying fell from 75 percent of a normal week to under 20 percent, while verification, exception judgment and escalation handling rose from 25 percent to over 80 percent. For vendor inquiry handlers the direction is the same and smaller. Across the two functions roughly 3,100 hours a year of capacity has been freed, and all 3,100 went to the exception backlog and the new verification standard, both previously carried as an accepted service failure. Nobody in these functions has left as a result of this program. That is what has happened, not a forecast of what will.

2. What we have decided. Waves one and two carried no headcount reduction target and none was applied: a fact about what happened, not a promise about the future. Wave three covers customer operations and begins in January with two processes named in the appendix. Hours freed go first to the customer-operations backlog and to the verification roles the redesign creates, as in waves one and two. The reskilling budget is funded at 340,000 for twelve months in the operating plan, covering verification certification and the practitioner track. Two new role types, AI practitioner and verification specialist, are established with published profiles and pay bands.

3. What we have not decided, and when we will. We have not decided the staffing model for customer operations under wave three: whether the redesigned process runs with the current team structure or a different one. That decision goes to the executive committee on the eleventh of November, and we will communicate the outcome within one week, to affected teams first, through their own managers. We have also not decided whether wave four extends to the two remaining shared-service processes; that sits in the March cycle. If either date moves, we will tell you on the original date.

4. What we commit to regardless. Internal-first posting, with a fourteen-day window, for every practitioner and verification role this program creates. Funded reskilling access during working hours for anyone in scope of any wave. If any role is reduced for any reason connected to this program, the notice and consultation process in the appendix applies, with a named HR case owner and funded outplacement. And the manager-first rule: anyone materially affected hears it from their own manager, in person, before any public announcement. These four are reviewed by HR and legal and do not change with the November decision. Next update: January, and quarterly thereafter, whether or not anything has changed.

The manager briefing pack

Six items, given to every manager in scope at least three working days before the cascade, in a live briefing rather than an email:

1. The statement verbatim, with a note on the reasoning behind each section so the manager can answer "why are you telling us this now?"
2. The three hardest questions with answers agreed word-for-word at the level of substance.
3. The permission line: "If you do not know, say so and name the day you will find out. That is a correct answer and it beats a guess."
4. A named escalation contact and a two-working-day response commitment, so the permission line is backed by something real.
5. Who in their own team is in scope of which wave, so the manager is not learning this in front of the team.
6. The date of the next update, which is also the manager's own commitment to their team.

The three hardest questions, answered honestly

"Is my job going away?" "Your role is in scope of wave three, and I am not going to tell you nothing will change, because the work is already changing and you can see it. What is decided: no headcount reduction target is attached to this program, waves one and two carried none, and hours freed went to backlog and new roles. What is not decided is the wave three staffing model, which goes to the executive committee on the eleventh of November. I will tell you what it says within a week, and you will hear it from me first."

"If the tool does seventy percent of what I used to do, why do you need all of us?" "Because the seventy percent it does was never the constraint. Our backlog and our error rate were not caused by people being slow at keying; they were caused by not having enough hours for judgment, checking and difficult cases. That work is now what the role mostly is, and we have more of it than we can staff. If that stops being true, I will tell you it has stopped being true."

"Why should we believe any of this?" The one managers most need help with: "Because of what is in the document and what is not. I have not told you nobody's job is at risk, because that is not mine to promise and you would be right not to believe it. I have told you what has happened here, what we have decided and funded, what we have not decided and the date we will, and four commitments that do not depend on any of it. If we break one of those four, stop believing the rest."

One moment, narrated

At the fourth of eleven briefings, a supervisor with a long memory asks the question the cascade lives or dies on. "Is the November decision genuinely still open, or has it already been made and you are waiting for the right time?" Two wrong answers feel reasonable: the indignant one ("of course it's open") sounds like what someone would say either way, and the soothing one ("I promise you it's open") stakes credibility on what she cannot verify.

What the transformation director says: "It is open. Here is who decides: the executive committee, six named people, and the paper goes in on the twenty-eighth of October. Here is what it will be decided on: the wave one and two throughput data, the backlog position in customer operations, and the redeployment options HR has costed. And here is the part to hold me to. If it had already been decided and I could not tell you, I would tell you it was decided and that I could not yet share it. I would not tell you it was open. That is the difference between me not knowing and me not telling."

That answer does not resolve her uncertainty. It gives her something better: a specific, checkable claim about how she will be treated, made in front of forty witnesses. Nine weeks later, when the November communication arrives on the eleventh with an outcome she does not love, the claim holds. Trust at enterprise scale is not built by good news; it is built by small verifiable claims that survive contact with bad news.

The Frame You Can Stand Behind, and the Promise That Could Not Be Kept

The pattern so far, stated accurately

Leaders reach for external evidence when the room gets hard, and there is evidence worth reaching for, provided you use it for the claim it supports. The measured enterprise record so far is not one of wholesale role elimination. McKinsey's State of AI work finds 88 percent of organizations using AI regularly while only about 39 percent report any earnings impact (EBIT, earnings before interest and taxes) at all, mostly under 5 percent. MIT's GenAI Divide study found roughly 95 percent of enterprise generative AI pilots delivering no measurable profit-and-loss return, naming adoption without transformation as the dominant failure. S&P Global found 42 percent of companies scrapping most of their AI initiatives in 2025. BCG's 10-20-70 rule (10 percent algorithms, 20 percent technology and data, 70 percent people and process) says the same from the cost side.

Read together, those findings describe a period in which the dominant measured effect has been task change and workflow redesign rather than mass role elimination, and in which most organizations have not extracted enough value to have changed much of anything. That is an observation about a record, not a forecast: it says nothing about 2030, nothing about your sector, and nothing about the role of the person asking. So the frame comes with a caveat you say out loud, every time: "That is the pattern so far, in this organization, for these processes, with the capability we have today, and capability is changing." A leader who says that has shown the room they distinguish evidence from prediction, which is the only reason anyone should believe them later, including when the news is good. Be the most accurate person in the room, not the most reassuring one. Accuracy is a durable asset. Reassurance is a loan.

The promise that could not be kept

A mid-sized manufacturer, three plants, an AI program eighteen months in and going reasonably well: quality inspection support on two lines, a planning assistant in scheduling, some document work in procurement. The transformation lead is well liked and genuinely decent. At a town hall, pressed by a group that has been reading headlines all year, he says the sentence that ends his usefulness: "I want to be absolutely clear. This program will not cost anyone their job." He means it. It is a sincere man trying to look after people in a room that wanted comfort. And it is not his to promise: he does not set headcount, does not control demand, and does not sit on the committee that will make the decision that eventually matters.

Fourteen months later, a restructuring driven by a lost contract and a soft market reduces headcount in operations, including a function the AI program had touched. The reduction is not caused by the AI program. The analysis is clear about that and completely irrelevant, because the sentence is still on the intranet with a date on it. Someone reads it aloud in a team meeting; it circulates by screenshot to plants that were not affected at all. Nobody argues with his analysis. They simply stop treating anything he says as information.

What follows is measurable. Adoption in the affected function declines over two quarters, not through sabotage but through quiet disengagement: fewer suggestions, less participation in improvement cycles, defects reported later. The next engagement survey carries one comment more than a dozen times: we no longer believe what we are told. Every later communication, including two carrying good news about new roles, is heard as sales.

The lesson is not "never say anything comforting." It is narrower. The most damaging sentence in workforce communication is a sincere promise made outside the speaker's authority. Cynical spin is survivable, because people discount it on arrival; a sincere promise gets believed, which is why breaking it costs. Run the ownership test on every sentence: do I control the thing I am about to guarantee? If not, it moves into part three or part four, or out of your mouth entirely.

Closing the chapter

This chapter has been about the seventy percent in BCG's arithmetic at enterprise altitude: the AI-native process organization, literacy beyond tool training, culture that survives fear, the flywheel that makes readiness continuous, and now the conversation underneath all of it, because a workforce that does not believe its leadership will comply with a transformation and will not carry one. Chapter 5.5 turns to the horizon: agentic operations as the next readiness bar, the 2027 to 2030 outlook, thought leadership, the maturity model, and your own 90-day plan.

What to Do Monday Morning

1. Draft the four parts on one page, in order. Write part one first and only from your own program's measurements: task composition changes, processes touched, where the freed hours went.
2. Audit part two for funded decisions only. Ask of each sentence whether it is decided and funded today. Anything "likely" or "planned" moves to part three with a date, or comes out. Expect the section to halve and to gain in value.
3. Put a date on every undecided item: a decision date and a communication date, both specific. If you cannot get one from the decision owner, that is the first thing to fix this week.
4. Verify every standing commitment with HR and legal before you say it: notice, consultation, reskilling entitlement, internal-first windows, transition support, each confirmed in writing by its owner. Anything unconfirmed stays out of part four.
5. Brief your managers with the three hardest questions and explicit permission not to know. Live, not by email, at least three days before the cascade, with a named escalation contact behind the permission.
6. Schedule the next update before you deliver this one, and put the date in the document. Then hold it, including the quarters when nothing changed, because the update that says "no change" teaches people that your silence would mean something.

Key Takeaways

• Reject all three temptations: reassurance beyond the evidence destroys credibility at the first contradiction, vagueness reads as concealment, and deferral hands the anxiety to the people with the least information and the most exposure.
• Build the Workforce Commitment Statement in four parts (what we know, what we have decided, what we have not decided and when we will, what we commit to regardless), separating decided from undecided from unknowable, written once at leadership level, delivered by every manager, updated on a published cadence.
• Ground part one in your own organization's measurements rather than industry employment forecasts, because your own task composition data is more credible and more relevant, and because whichever forecast you cite you have adopted.
• Date every unknown: an open question with a decision date and a communication date is far less corrosive than an undated one, and the obligation holds even when the only news is another delay.
• Treat part four as absolute: notice, funded reskilling, internal-first hiring, transition support and the manager-first rule are what you fully control, must be checked with HR and legal, and lose all value at the first breach.
• Cascade through prepared managers rather than scripts, give them agreed answers to the three hardest questions plus permission to say "I will find out by Friday," and check coverage by sampling employees rather than managers.
• Use the measured record (McKinsey's 88 percent using AI against about 39 percent seeing any EBIT impact, MIT's 95 percent, S&P Global's 42 percent) as evidence about what has happened so far rather than as a forecast, and never let a headcount program be described as augmentation.