Making Recommendations

Who this is for

This subskill is for Data Analysts who turn findings into clear, actionable recommendations that decision-makers can trust and implement.

Prerequisites

• Basic descriptive statistics (means, proportions, trends)
• Comfort building simple charts (line, bar, funnel)
• Ability to frame a business question and define a primary metric

Why this matters

In real roles, you will not be judged only by charts—you will be asked: What should we do and why? Strong recommendations help teams choose, act, and measure results. Typical tasks include:

• Proposing A/B tests or rollouts after identifying a conversion drop
• Prioritizing product or marketing actions using impact vs effort
• Summarizing trade-offs and risks for leadership decisions
• Defining success metrics, owners, and timelines

Concept explained simply

A recommendation is a concise decision proposal backed by evidence and designed for action. It should answer: What should we do, why, what impact do we expect, how will we do it, what could go wrong, and how will we know it worked?

Mental model: The WWH-HR-D loop

• What: The action or decision to take
• Why: Evidence and logic that justify it
• How: Steps, owner, timeline, resources
• Hypothesis & Impact: Expected change in a key metric
• Risks & Alternatives: What to watch, options if wrong
• Decision & Measurement: Go/no-go rule and success criteria

A simple structure for recommendations

1. Decision: Approve X (or run an experiment)
2. Objective metric: Target metric and baseline
3. Expected impact: Direction and magnitude (even if a range)
4. Rationale: Key facts, not the entire analysis
5. Plan: Steps, owner, timeline
6. Risks & mitigations: Biggest uncertainties and how to monitor
7. Success check: What result qualifies as success and next step

Prioritization methods

When you have multiple ideas, choose the best few to propose.

• Impact vs Effort: High-impact, low-effort first
• ICE: Impact × Confidence ÷ Effort (score each 1–10)
• RICE: Reach × Impact × Confidence ÷ Effort (consider audience size)

Confidence and assumptions

Always state your confidence and the assumptions your recommendation depends on. If confidence is low, recommend a test with a decision rule.

• Confidence bands: High (≥80%), Medium (50–79%), Low (<50%)
• Assumptions: e.g., seasonality stable, data complete, no major channel changes
• Decision rule: e.g., Ship if uplift ≥ +1.5pp at p < 0.05 for 2 weeks

Worked examples

1) E-commerce: Mobile add-to-cart drop

• Finding: Mobile add-to-cart rate fell from 8.2% to 7.1% (−1.1pp) after new image carousel; page weight +1.4MB; LCP +1.2s.
• Recommendation (draft): Run an A/B test replacing carousel with a static hero on mobile.
• Why: Performance correlation across 4 product pages; historical tests show +0.6–1.0pp when LCP −1s.
• Impact: Expect +0.6–0.9pp add-to-cart (Medium–High impact), Confidence: Medium.
• Plan: Eng team implements variant (2 days), QA (1 day), test for 2 weeks. Owner: Web PM.
• Risks: Creative consistency; mitigate with brand-approved static image.
• Decision rule: Ship if uplift ≥ +0.5pp at p < 0.05; otherwise iterate on image size.

2) SaaS: Reducing churn in first 30 days

• Finding: 35% of churned users never completed onboarding step 2; emails have 9% CTR.
• Recommendation: Add in-product checklist with progress bar + contextual tooltips; deprecate email nudges.
• Why: Session replays show confusion at data import; best practices and internal micro-test suggest +12–18% step completion.
• Impact: Predict churn reduction −2–3pp (Medium), Confidence: Medium.
• Plan: Design (3 days), build (5 days), release behind feature flag. Owner: Growth PM.
• Success: Step-2 completion +15% and churn −2pp vs control after 4 weeks.

3) Marketing: Reallocate spend

• Finding: Paid Social CPA $78 vs Paid Search CPA $49; marginal CPA on Social rising week over week.
• Recommendation: Shift 15% budget from Social to Search branded + high-intent non-brand.
• Why: Diminishing returns curve; Search impression share 72% with available inventory.
• Impact: Forecast blended CPA −6–9% (High), Confidence: High.
• Plan: Reallocate for 14 days; monitor daily CPA and impression share. Owner: Performance Lead.
• Risk: Lower top-funnel; mitigate with retargeting cap adjustment.
• Decision: Maintain shift if blended CPA ≤ −5% without MQL quality drop.

Exercises

Exercise 1: Turn findings into a recommendation

Scenario: Mobile add-to-cart rate dropped 12% relative after launching an image-heavy carousel. LCP worsened by 1.2s. Desktop unaffected. Your task: write a recommendation using the structure above.

• Includes What, Why, Impact, How, Risks, Decision rule
• Names owner and timeline
• States confidence and key assumptions

Exercise 2: Prioritize with ICE

Score each idea (Impact 1–10, Confidence 0.5–1.0, Effort 1–10; higher Effort = harder). Compute ICE = Impact × Confidence ÷ Effort and rank.

• A) Reduce image size (I=6, C=0.8, E=2)
• B) New referral program (I=7, C=0.6, E=6)
• C) Improve search relevance (I=9, C=0.7, E=8)
• D) Add onboarding checklist (I=5, C=0.9, E=2)
• E) Reallocate ad spend (I=6, C=0.9, E=3)

Quality checklist (self-review)

• I stated a single clear decision or experiment
• I mapped recommendation to a target metric and baseline
• I estimated impact and my confidence level
• I named owner, steps, and timeline
• I noted top risks and a decision rule

Common mistakes and how to self-check

• Vague actions: Fix by starting with a verb and a decision (Approve X / Test Y).
• No metric: Always name the primary metric and baseline.
• Overstuffed rationale: Keep 2–4 key facts; move details to appendix if needed.
• No owner/timeline: Assign one person and a realistic timeframe.
• Ignoring risk: State uncertainties and define a monitor/rollback.
• Overconfidence: Use confidence bands and ranges for impact.

Practical projects

1. Conversion rescue memo: Analyze a recent traffic or conversion change (real or sample data). Produce a one-page recommendation with decision rule and owner.
2. Prioritization board: List 10 growth ideas, score with ICE or RICE, and propose the top 3 with short rationale.
3. Experiment playbook: Write 3 test cards (hypothesis, metric, MDE, duration rule of thumb) and a go/no-go template.

Learning path

1. Confirm your business goal and primary metric
2. Summarize findings into 2–4 key facts
3. Draft a recommendation using the structure
4. Estimate impact and confidence; decide test vs ship
5. Prioritize with ICE/RICE
6. Get feedback from a peer and revise
7. Present and capture decisions and follow-ups

Mini challenge

You find that 22% of users abandon at checkout step 3 (payment). A new payment provider promises faster load times but a 0.3% fee increase. Draft a 6-bullet recommendation (What, Why, Impact, How, Risks, Decision rule). Keep it under 120 words.

Next steps

• Complete the quick test below to lock in the structure and prioritization concepts.
• Note: Anyone can take the test; if you log in, your progress will be saved.