What You Need to Know
A stakeholder here is a colleague who reads your report on a Claude trial and decides what happens next. Reporting to them has one rule: say what is true, including the parts that are not impressive. The rule serves two audiences: the enthusiast and the sceptic.
This lesson uses one made-up example throughout: a travel agency whose office manager is reporting on a three-month Claude trial. An accurate report is honest, and it also protects the work. If she over-promises, she wins approval for a plan that she cannot deliver. Then stakeholders read the first visible error as proof that the whole trial was oversold. If she under-claims, the work does not get the support that it earned. A sceptic cannot disprove an accurate claim.
Stating Value in Four Parts
State value in four parts: the task, the before, the after, and how you know. "The monthly sales report takes two hours instead of five, measured across three months" is a claim the travel agency's office manager can defend. "Claude has transformed how we work" is not a claim: it names no task and no measurement, so nobody can check it. An audience that has heard that sentence before will discount everything that follows it.
Anthropic's enterprise guidance says the same. It advises measuring productivity gains, such as hours saved, through periodic user surveys, and reporting to stakeholders at regular intervals. The guidance pairs quantitative data with qualitative examples, in place of enthusiasm.
Every figure in the report appears because somebody measured it over a stated period, never because it sounds about right. A measured saving applies to its own task. The report never projects the sales report figure onto work that the trial did not include.
The Untested Category
State limitations plainly: what still needs a person and why, what went wrong during the pilot, and what has not been tested. People omit the untested category. This gap causes trouble later, because stakeholders take the lack of a statement as a claim. If a stakeholder is never told that some work went untrialled, the stakeholder will reasonably assume that the trial covered that work. The eventual failure is then blamed on the report that said nothing about it, as well as on the work. If the travel agency never tried Claude on refund requests, naming that gap costs one sentence. If the report leaves the gap unnamed, the report loses its credibility when the gap becomes visible.
A Caught Error Is Evidence
A caught error is evidence for the process. Suppose a booking confirmation used the wrong date and the check before sending caught the error. When you report the error, a sceptic learns that the checking step works. This is the very thing that the sceptic doubted. If the report hides the error and the error becomes visible later, readers re-read every other claim in the report as selected to look good. Reporting a caught error is the Diligence element of Anthropic's AI Fluency course, applied to a report. Diligence means that you stay accountable for what the AI produced, including in what you tell colleagues about it.
The Enthusiast and the Sceptic
Stakeholders have one of two attitudes, and each attitude needs something different from the same truthful account. The enthusiast, perhaps the travel agency's owner, wants Claude in use across the whole business by Friday. The enthusiast needs to hear the limitations and the human gate without losing the enthusiasm that will make the work progress. The human gate is the step where a person checks each output before it goes out. So the answer to "can it handle all of it now?" is not a plain no: the answer names what stays human. The sceptic, perhaps one of the sales staff, suspects that Claude is a useless novelty, a risk or a threat to somebody's job. The sceptic needs the measured result, the named human step, and an honest account of what went wrong and how it was caught.
Overcorrection fails with both audiences. Overcorrection means so many caveats that a working result sounds like a failure. Under-claiming is not honesty. It gives the sceptic a conclusion that the data does not support.
In the exam
A prepared candidate treats a caught error as evidence. Whether a use is permitted, and who must approve it, belongs to Domain 6. This objective covers the honest account of results to colleagues, never a governance sign-off.
Use only measured numbers
In this objective's questions, the only number that can support the right answer is one that the question itself says was measured. Any option that quotes a saving that nobody measured, or that projects a measured saving onto untested work, is wrong. The option is wrong however reasonable the figure sounds.
Exam Traps
Presenting enthusiasm as evidence
A transformation claim with no task, no before, no after and no measurement cannot be checked. So a sceptical audience discounts it. The audience also discounts the genuinely measured result underneath it. A defensible claim states the four parts and stops.
Projecting a measured saving onto untested work
A measured result belongs to the task it was measured on. Extending it to a category the trial never included quotes a number nobody measured, and the exam treats it as wrong regardless of how plausible the projection looks.
Leaving the pilot's caught error out of the report
When the error becomes visible later, every other claim gets re-read as selected to look good. Disclosed, the same error was evidence that the checking step works, which is exactly what the sceptical audience needed to hear.
Hiding a working result among many caveats
So many warnings make a working process sound like one that failed, handing the sceptic a conclusion the data does not support. State the measured result and its period plainly, keep the limits the trial found and name what it never tested.
Practice Scenario
The operations manager of a law firm has trialled Claude for three months. Client letters now reach a first draft in minutes rather than half an hour, measured across the trial, with every letter still read and approved by a lawyer before it goes out. Contract summaries saw no saving, because the partner review was always the bottleneck. One draft gave the wrong date for a client meeting and the lawyer caught it before sending. The partners have asked for what the trial showed, with nothing projected or hypothetical and nothing left out. What should the report to the partners say?
Build Exercise
The Three-Part Account
What you'll learn
- How to state a value claim in four defensible parts
- How to name the untested category explicitly
- How to present a caught error as evidence for the process
- How to answer the enthusiast and the sceptic from the same truthful account
- Take any Claude use you have actually run for a few weeks; the augmentation from the 4.4 exercise is ideal
Why: The account only teaches honesty when the material is real; invented pilots have no inconvenient findings.
You should see: One named piece of work with at least a few weeks of history behind it.
- Write part one: what changed, with the task, the before, the after, and how you measured it
Why: The four-part shape is what makes a claim defensible to a sceptical reader.
You should see: One or two sentences a sceptic could check rather than merely believe.
- Write part two: what did not change and why, and what you have not tested, named explicitly
Why: Saying nothing about the untested category is taken as a claim; naming it is what prevents the later surprise.
You should see: An honest paragraph including at least one thing the trial never included.
- Write part three: what stays human and why, including one example of something the human step caught, however small
Why: The caught error is the evidence the sceptic needs that the checking step works.
You should see: A named human gate and one concrete caught error, presented as the system working.
- Delete every number you did not measure, then rehearse two replies aloud: the enthusiast asking can it do all of it now, and the sceptic asking so it makes mistakes
Why: The two attitudes are the objective’s test of whether one truthful account can serve both audiences.
You should see: An enthusiast answer that names the human gate without a plain no, and a sceptic answer that treats the caught error as evidence.
Sources
- Claude Enterprise Administrator Guide — Anthropic
- AI Fluency: Framework and Foundations — Anthropic (the element names are the course's module titles; the one-clause glosses of each element are our paraphrase of the course content)
- Claude Certified Associate – Foundations Exam Guide — Anthropic. No Anthropic document covers stakeholder communication as a skill. This lesson draws on the exam guide's objective and standard professional practice; the Enterprise Administrator Guide and AI Fluency support specific points.