Stakeholder Communication and Analytical Project Constraints

“Can you tell us by Friday why new hires are not finishing training?” sounds like one request. It can mean a completion count, a diagnosis of the course, or evidence for replacing it. Those deliverables require different data and support different decisions. Before agreeing to Friday, find out what someone intends to do with the answer.

This worked case uses a fictional training team, invented counts, and an illustrative schedule. It follows the request from initial agreement through a missing-data update, a scope change, and a difficult question. The aim is to make a defensible delivery promise and keep that promise aligned with the evidence.

Find the decision and the people affected

Ask the requester: “What decision is due on Friday? Would a count be enough to make it? What would you do differently if the result were higher or lower?” Here the training director needs to choose whether to keep the current course for the next intake or investigate revisions. A count can describe the current cohort; it cannot by itself establish why people stopped or whether a replacement would help.

A stakeholder is a person or group that can affect the work or be affected by its outcome. This includes employees whose records are being interpreted, even if they did not commission the analysis. Do not treat influence over the budget as the only reason to listen.

Role in this case	What to establish
Training director	Owns the Friday decision and approves a changed deliverable.
Course administrator	Explains enrollment, exemptions, and what counts as completion.
Learning-system owner	Provides the permitted extracts, their coverage, and a delivery estimate.
Analyst and reviewer	Define the measure, check the evidence, and challenge unsupported claims.
Employee representative	Explains access barriers and concerns about how results may be used.

These are roles, not five mandatory job titles. In a small team one person may cover several. Name who can resolve an access problem, who can revise the deadline, and who reviews the analysis. Access must still follow the organization’s rules; being the sponsor does not automatically authorize every use of employee data.

Turn the request into a checkable agreement

Scope specifies the questions, population, period, and outputs included in the work, together with explicit exclusions. A constraint limits feasible delivery, such as a fixed decision date or permitted data access. An assumption is something the plan currently treats as true; a dependency is something that must be supplied or completed for another task to proceed. Write these separately so an assumption does not quietly become a promise.

Agreement item	Training case
Decision and delivery	Friday 10:00: a one-page completion summary and a recommendation about the next investigation.
Population and period	The agreed roster of 200 eligible new hires; completion by the end of Monday.
Measure	Distinct people meeting the administrator’s completion rule by the cutoff, divided by 200. Document exemptions before fixing the roster.
Included / excluded	Cohort status and data gaps included; causal diagnosis, manager rankings, and a replacement-course trial excluded.
Dependency	System owner to supply the mobile-learning extract by Wednesday noon; analyst to check it before promising final coverage.
Acceptance	Roster and statuses reconcile, definitions and source versions are retained, unresolved cases are visible, and a peer checks the calculations and claims.

The acceptance criteria concern whether the agreed analysis was delivered, not whether the result pleased the requester. A report showing an unresolved gap can meet the criteria for an explicitly provisional delivery. It cannot silently replace a promised final result. Record who accepts a revised scope and which decisions that provisional report can support.

Agree on update times and the channel for urgent blockers. Ask how unfamiliar acronyms are used, whether a response needs acknowledgment or a complete answer, and who covers absences. A team may need a same-day escalation or a later considered reply; a universal “reply within 24–48 hours” rule would ignore that context. Make written requests specific enough to survive forwarding: requested action, owner, due time, and purpose.

A fast answer with a visible boundary

On Tuesday, the agreed 200-person roster contains 120 confirmed completions, 40 confirmed non-completions, and 40 unresolved statuses awaiting the mobile extract. These categories are mutually exclusive and cover the fixed roster. Do not drop the unresolved people from the denominator or count them as proven non-completers.

Confirmed completions are 120 / 200 = 60% of the cohort. If only those 40 unresolved statuses can change, the final completion proportion could range from 120 / 200 = 60% to 160 / 200 = 80%. This is a bound from unresolved classifications, not a confidence interval or a probability that the rate lies in that range. If the roster, cutoff, or already confirmed classifications are wrong, the bound no longer describes the full uncertainty.

The fast answer is therefore: “120 of 200 eligible people are confirmed complete; 40 are confirmed incomplete and 40 remain unresolved. We cannot yet report one final completion rate.” More time is useful if it resolves a specific gap or permits a needed check. It does not automatically turn observational patterns into causes. A descriptive count may be exactly what the requester needs; do not expand every count request into a research project.

Tell each audience what they need to act

Keep the underlying facts consistent while changing the detail. The system owner needs the missing fields, cohort identifiers through a permitted channel, cutoff, extraction version, and checks already attempted. The coordinator needs the blocked task, the dependency owner, and schedule impact. The director needs the consequence for Friday’s decision and a choice, not a raw error log.

Example update: “Friday training decision: 40 of 200 statuses remain unresolved because the mobile extract is not available. We can deliver a labeled cohort summary on Friday, but cannot yet support a final completion rate or manager comparison. The system owner is checking delivery by Wednesday noon. Please confirm by Tuesday 16:00 whether Friday should use the provisional summary or whether you want a revised delivery plan. I will update you at Wednesday noon even if the extract has not arrived.”

Use a short discussion when the options or responsibilities are disputed, then record the decision in writing. Do not infer approval from silence. Keep access credentials and unnecessary personal records out of forwarded messages. Communicating uncertainty early and near the claim follows the Government Analysis Function’s guidance; an appendix can carry detail but should not hide a limitation that changes the decision.

Make a scope change a decision

On Wednesday the director asks, “Can we also rank managers and explain who is responsible?” That is a scope change: a new output and intended use, not just one extra chart. It requires a suitable comparison population, reliable manager assignments, review of permitted use, and an assessment of alternative explanations. Completion records alone do not identify a manager’s causal effect.

Offer choices with consequences: retain Friday’s cohort summary; replace it with a clearly bounded exploratory breakdown after data and permission checks; or commission a separate causal investigation with a new plan. State the added work and dependencies before estimating a date. If the requester accepts a change, update the question, exclusions, deadline, owner, and acceptance criteria together. Preserve what was superseded and why. A small project can do this in a short decision note; it need not create a committee for every edit.

Prepare answers that can survive scrutiny

Before the review, ask a colleague unfamiliar with the work to explain back the population, numerator, cutoff, and proposed action. Confusion here calls for a clearer main report. Keep the source version, definitions, reconciliation, cleaning decisions, and calculation trail available for deeper questions. Decide which teammate answers data-access or operational questions; do not ask someone to improvise authority they do not have.

Listen to the full question. Restate it if its meaning or audience audibility is uncertain, then give the answer and the evidence boundary together. A short answer can still be misleading if its qualification is withheld until someone asks. “We tested alternatives and got consistent results” is an appropriate response only when that testing actually happened.

Suppose 30 of the 200 employees answered an optional survey and 22 described the course as confusing. Say “22 of 30 respondents reported confusion,” with the 15% response rate. Do not say 73% of all employees were confused or that confusion caused non-completion. Selection and nonresponse bias may separate respondents from the full cohort. The reports identify a question worth investigating; they do not establish its prevalence or causal contribution across all 200 people.

A valid objection can change the report. If “completion” means passing the assessment in one source but merely opening the course in another, pause the combined rate and reconcile the definitions. If asked whether a pattern is seasonal and you have not checked, say so, name the comparison needed, assign an owner and a follow-up time, and state whether the current recommendation must wait. Being composed is useful; it is not evidence that the analysis is correct.

Practice the next conversation

The director writes: “Please put 60% completion on Friday’s slide. Keep caveats in the appendix.” Draft a two-sentence reply.

One possible response

“We can report that 120 of 200 eligible people are confirmed complete, with 40 confirmed incomplete and 40 unresolved shown beside that figure. Calling 60% the final rate would hide an unresolved range of 60–80% under the fixed-roster assumptions, so I propose a clearly labeled provisional status summary.” The important change is the claim, not a smaller footnote.

The mobile extract will arrive after Friday, and an extra analyst is available today. Does that solve the schedule? Write the next decision to request.

One possible response

Extra capacity cannot supply missing source records. Ask the decision owner to choose an explicitly provisional Friday delivery or a revised date for the final result, while the system owner checks for a valid alternative source. Do not promise that the second analyst makes the original deadline feasible. Record who owns each action and when the next update is due.

A reviewer says: “The 22 survey comments prove the course design caused the low completion rate.” Give a response and one useful follow-up.

One possible response

The comments report respondents’ experiences, with 30 respondents out of 200; the final completion rate is also unresolved. They do not identify a causal effect. A useful follow-up is a permitted usability check of the reported confusing steps, recording observed barriers and defining how a subsequent course change would be evaluated. Do not claim that interviews alone establish the population-wide cause or the effect of a redesign.