Writing Data-Driven Recommendations
Practice writing engineering recommendations backed by data: structure, confidence level, and call-to-action.
Data-driven recommendation structure
- Opening: "Based on the data, we recommend..." — specific action with impact scope
- Confidence: "With high confidence..." / "Early signals suggest..." — match your evidence base
- Call-to-action: actor + action + deadline + validation gate + proceed condition
- Risk framing: "If we do nothing, we project..." — quantified consequence with timeline
- State vocabulary: recommendation (pending) vs. decision (approved, with name + date)
Question 0 of 5
Which sentence best opens a data-driven recommendation?
"Based on the data, we recommend..." is the canonical opening for a data-driven recommendation.
- "Based on the data" — signals that what follows is evidence-grounded, not opinion. This phrase triggers a different kind of reading than "I think" or "we feel".
- "we recommend" — first person plural, present tense, active voice. Cleaner than "it is recommended that" (passive) or "we should consider" (hedged).
- Specific recommendation — the reader knows exactly what is being proposed: migrate session cache, from Redis, to local in-memory, for read-heavy endpoints.
- Scope of impact included — "94% of cache hits" — quantifies why this matters, making the recommendation legible to a stakeholder who must approve it.
Which sentence correctly uses a confidence qualifier in a data-driven recommendation?
Confidence qualifiers distinguish what the evidence firmly supports from what is still being validated. This recommendation uses two levels:
- "With high confidence" — six weeks of production data is a strong evidence base for the general direction.
- "Early signals suggest... though this requires full rollout to confirm" — the canary result is promising but not conclusive. Being explicit about this prevents stakeholders from treating a 5%-traffic result as a 100%-traffic guarantee.
- High confidence: "the data shows", "we are confident that", "the evidence strongly supports"
- Moderate confidence: "the data suggests", "early results indicate", "this is consistent with"
- Low confidence / hypothesis: "early signals suggest", "it is possible that", "we hypothesise that"
Which sentence best writes a call-to-action in a data-driven recommendation?
A call-to-action must name who does what, by when, and what confirms success before the next step.
- Actor named — "the platform team" — not "someone" or "the team".
- Specific action — "deploy the updated connection pool config to 25% of production traffic".
- Timeframe — "by end of sprint 14" — a concrete deadline, not "soon".
- Measurement step — "measure the p95 latency impact for 48 hours" — builds in a validation gate.
- Proceed condition — "if the improvement holds" — defines the decision criteria for the next phase. This prevents the rollout from becoming unconditional after a partial deployment.
Which sentence correctly uses "if we do nothing" risk framing in a recommendation?
"If we do nothing, we project..." connects inaction to a specific, quantified consequence with a timeline.
- "If we do nothing" — frames the alternative explicitly. Most recommendations only argue for the positive case; naming the cost of inaction makes the tradeoff concrete.
- "we project" — honest epistemic framing. This is a forecast, not a certainty. "Will happen" would overstate the confidence.
- Quantified trajectory — "4.2-second p99 checkout latency will exceed our 5-second SLA threshold" — names the metric, current value, and breach threshold.
- Timeline from data — "within 6–8 weeks... based on forecasted 15% per month traffic growth" — the forecast is grounded in a specific growth rate, which is checkable.
- Business consequence — "automatic SLA penalty payouts" — translates a technical metric into a financial risk that a non-technical stakeholder can act on.
Which passage correctly distinguishes between a recommendation and a decision in a design document?
In design documents, "recommendation" and "decision" are distinct states with different audiences and implications.
- Recommendation — the analysis team's proposal, subject to approval. Use "we recommend" + who must approve. This state invites challenge and input.
- Decision — the approved direction, which is now binding. Use "we will" (imperative, present tense). Include the decision-maker's name and date — this creates an audit trail and prevents decisions from being relitigated.
- A recommendation without approval cannot drive engineering work.
- A decision without a name and date can be challenged as undecided.
- Conflating the two leads to teams starting work on unapproved directions, or approved decisions being re-opened without cause.