Business logic — reasoning about effects, constraints and consequences to make recommendations
Turn analysis into decisions — reason about campaign, cashback, bonus, discount, partnership and commission effects; respect constraints; quantify consequences; recommend; and measure impact afterwards.
This lesson builds on: Correlation vs causation, causal inference and time-series concepts, Financial models, scenarios, sensitivity — and bonuses, commissions, cashback and discounts.
- Anticipate second-order effects of commercial levers (pull-forward, cannibalisation, gaming)
- Incorporate business constraints and operational limits into recommendations
- Quantify financial consequences with ranges
- Write recommendations that are specific, prioritised and measurable
- Design impact measurement before the decision is implemented
Every lever has side effects
| Lever | Intended effect | Typical side effects to check |
|---|---|---|
| Campaigns | More customers / sales now | Pull-forward (sales borrowed from next weeks); discount given to customers who would have bought anyway |
| Cashback | Repeat purchases | Selection bias in evaluation; liability for unredeemed balances; margin erosion |
| Bonuses | Motivation toward targets | Gaming (pushing sales into the measured month), sandbagging plans, neglect of unmeasured work |
| Discounts | Volume | Margin loss; reference-price damage; stockpiling |
| Partnerships | Access to new customers | Cannibalisation of own channels; partner margin; data sharing |
| Agent commissions | Partner sales effort | Low-quality sales (high churn) if paid on activation only — add clawbacks or tenure-based payouts |
Customer behaviour adapts: once customers learn there is a Black Friday discount, some wait for it.
Worked example: should Orion repeat a 25% Black Friday discount?
Facts (Module 8): orders/day rose from ~41 to ~57 during Black Friday 2025 at an average discount of ~22% vs ~0.5% baseline.
Reasoning chain:
- Incremental volume: 57 − 41 = 16 extra orders/day × 11 days ≈ 176 orders — but subtract pull-forward measured in the following 2 weeks.
- Discount cost on baseline demand: ~41 orders/day × 11 days × average order ~90 × ~22% ≈ 8.9k of revenue given away on orders that would have happened anyway.
- Incremental margin (approximation: cost ≈ 60% of full price, so margin after a 22% discount ≈ 18% of full price): 176 × 90 × 0.18 ≈ 2.9k.
- Budget: 80k campaign spend.
- Conclusion: 2.9k incremental margin − 8.9k discount on baseline orders − 80k budget ≈ −86k. On these assumptions the campaign destroyed value — unless it brought new customers with lifetime value large enough (check: how many Black Friday buyers were first-time, and how did they retain?).
The recommendation might be a smaller, targeted discount for new customers only, tested against a holdout — and a measurement plan agreed in advance.
Constraints and operational limitations
A recommendation that cannot be executed is not a recommendation. Check:
- Capacity: warehouse throughput, call-centre agents, delivery slots.
- Stock: will the promoted items be available? (stock-outs turn demand into frustration)
- Budget & cash: timing of spend vs revenue.
- Regulation & contracts: pricing rules, data protection, partner agreements.
- Systems: can the CRM actually target the segment?
Financial consequences with ranges
Present outcomes as ranges with stated assumptions: "Expected incremental margin 15–40k (base 25k), mainly sensitive to the conversion uplift (± 2 points) and redemption rate." Show the downside: what happens if the key assumption is wrong?
Writing recommendations
Structure (one page):
- Answer first: "Recommend replacing the blanket 25% discount with a 15% new-customer offer and a holdout test."
- Why (3 evidence points with numbers).
- Expected impact (range) and cost.
- Risks and mitigations.
- Measurement plan: metric, control group, duration, decision rule.
- Next step and owner.
Data storytelling principles: lead with the conclusion; one message per chart; annotate the chart with the insight ("Pull-forward: −18% orders in the 2 weeks after"); use the audience's language (margin, cash, customers) rather than method names.
Business impact measurement
Decide how impact will be measured before implementation:
| Design | Strength |
|---|---|
| Randomised holdout (A/B) | Strongest causal evidence |
| Staggered rollout (regions start at different times) | Natural comparison groups (DiD) |
| Matched control stores | Good when randomisation impossible |
| Before/after only | Weak — confounded by trend and seasonality |
Then close the loop: report realised vs expected impact, and update the assumptions used in the next decision.
Practice
Sales jumped 40% during a 1-week promotion and fell 15% below normal for the next two weeks. What is this?
Reps are paid a quarterly bonus on bookings. Bookings spike in the last week of each quarter and cancellations rise the following month. Likely explanation?
A new loyalty feature will launch to all customers next month. Which measurement design gives the strongest evidence of impact?
Write a three-sentence recommendation (answer first, evidence, measurement) for replacing a blanket 25% Black Friday discount that produced ~2.9k incremental margin on an 80k budget.
Summary
- Anticipate side effects: pull-forward, cannibalisation, gaming, selection bias, margin erosion.
- Check constraints before recommending; quantify consequences as ranges with assumptions.
- Recommendations: answer first, evidence, impact, risks, measurement plan, owner.
- Design impact measurement up front; holdouts beat before/after comparisons.