Convey Data Understanding to Leadership
Key Takeaways
- Executive summaries lead with the recommendation and the decision required, then the evidence; methodology belongs in the appendix.
- Report data readiness as a named-gap status — each requirement met, partially met, or not met with its business consequence — never as a percentage complete.
- Translate every technical finding into who is affected, how much volume, and what it costs: '18% missingness in the SMB segment' becomes 'we cannot serve small-business customers at launch, 30% of target volume'.
- Agree the reporting cadence before Phase 2 starts so bad data news travels through an existing channel instead of needing a special meeting.
- When findings undermine the business case, present them at the next steering opportunity with costed options rather than continuing into modeling and hoping.
The Highest-Leverage Communication in an AI Project
Task 9 closes Domain III: convey data understanding to leadership. Everything Phase 2 discovered is worthless if it does not reach the people who can change scope, funding, or timeline. Leaders rarely reject a data finding they understand; they reject findings that arrive late, arrive vague, or arrive without options. This task is where a project manager either protects the organization from an unwinnable build or lets it walk into one.
Prepare Executive Summaries of Data Assessment Findings
Lead with the recommendation and the decision required, then the evidence. Executives read the first paragraph; if it opens with methodology, the finding is buried. A workable structure fits on one page:
- Recommendation — proceed, proceed with a narrowed scope, pause to remediate, or stop.
- Decision required, and by when — name the decision and the date it becomes expensive.
- What we found — three to five findings, each stated in business terms with a number.
- Options — two or three, each with cost, duration, and consequence.
- Risk of delay — what changes if the decision slips a month.
Write for someone who will not read the appendix. Put the profiling detail behind the summary, not in front of it.
Create Visualizations and Reports That Communicate Data Insights
Choose visuals that make a data problem legible in five seconds. What works: a simple bar of coverage by business segment against the segment's share of target volume, so a gap is visible as a mismatch; a timeline showing the usable history against the business cycle it must span; a single before-and-after count showing how many records survive the joins. What misleads: percentage-complete bars that imply steady progress toward a guaranteed outcome; a pie chart of missingness that hides which segment is missing; truncated axes that exaggerate a difference; correlation heatmaps and null-rate tables shown to people who will read them as reassurance because they look rigorous. If a chart cannot be captioned with one sentence about a business consequence, it does not belong in front of leadership.
Present Data Readiness Status and Recommendations
Report readiness as a named-gap status, not a percentage complete. Ninety percent complete tells a sponsor nothing about whether the remaining ten percent is a formatting nuisance or the missing target variable. Instead, list each material data requirement as met, partially met, or not met, and attach the business consequence and the decision it forces. That format survives being forwarded, and it lets a leader who reads only the not-met rows still make the right call. Recommend explicitly; a status without a recommendation pushes the judgment onto people with less information than you have.
Translate Technical Data Concepts into Business-Relevant Language
Translation is the skill this task actually tests. The rule is to convert every technical statement into who is affected, how much volume, and what it costs or prevents.
| Technical finding | Business translation | Decision it forces |
|---|---|---|
| 18% missingness in the income field, concentrated in the SMB segment | We cannot reliably score small-business customers at launch — 30% of the target volume | Launch to enterprise only, or fund SMB data collection |
| Labels exist only for escalated claims, about 12% of volume | The model will be strong on cases we already catch and weak on the routine claims where the savings were forecast | Fund an SME labeling effort, or re-forecast the benefit case |
| Sensor telemetry lands 36 hours after the event | We can explain failures after they happen but cannot warn the plant in time | Fund a real-time feed, or change the use case to root-cause analysis |
| Nine months of history, no peak season included | The model has never seen the peak period it will be judged on | Delay launch past one peak, or launch with human review during peak |
| 82% of training rows come from two metro regions | A national launch would serve the whole country with a model that has seen two cities | Narrow the launch geography, or acquire regional data |
| Positive class prevalence is 0.3% | Real cases are rare in the data too; expect several false alarms per genuine catch | Size and staff the review queue now, not after go-live |
Keep the numbers; drop the vocabulary. Missing-not-at-random becomes the data we lack is concentrated in exactly the customers we planned to serve.
Provide Regular Updates on Progress and Challenges
Set a cadence before Phase 2 begins — a short weekly written update to the sponsor and a standing item at the steering committee — so that bad news travels through an existing channel instead of requiring a special meeting nobody wants to call. Report challenges while they are still questions, not after they harden into blockers. The discipline that matters is surfacing bad data news early: a gap raised in week two is a scope conversation, and the same gap raised in month four is a credibility event.
Expect an expectation gap. By the time Phase 2 finishes, the AI capability has often been announced internally, promised to a customer, or mentioned to a board. Address it directly: acknowledge what was promised, state what the data supports, and offer a path that preserves as much of the commitment as the evidence allows — a narrowed first release, a phased rollout, a human-assisted launch. Do not let a public promise become the reason a finding goes unreported.
The Exam Behavior: Report Promptly, with Options
When the data assessment undermines the original business case, the credited action is to bring it to the sponsor or steering committee at the next opportunity, with a recommendation and costed options — never to continue into Phase 4 hoping modeling rescues it, never to soften a material finding into minor data gaps, and never to hunt privately for replacement data first and report only if the hunt fails. Leadership can accept a narrowed scope, fund remediation, or stop. It cannot make any of those choices from information it has not been given.
Phase 2 findings show the promised capability can serve only about 70% of the target transaction volume at launch. The steering committee meets in two days, and the sponsor announced the AI capability at an all-hands meeting last month. What should the project manager do?
A project manager must report data readiness to an executive steering committee at the end of CPMAI Phase 2. Which reporting format is most useful to that audience?
The data analyst reports: “Outcome labels exist only for claims that were escalated for review, roughly 12% of total claim volume.” Which wording best belongs in the executive summary?