Personalization Ambitions and a Single Data Scientist
Lumen & Loom is an online home-goods retailer with 2.1 million monthly visitors and an 11-person digital team. The CEO has set a 12-month objective to lift conversion from 2.4% to 3.0% through personalization. Marketing wants to launch eight campaign landing pages by next quarter; the product manager wants a machine-learning recommendation engine; legal has flagged that the current cookie consent banner does not meet new privacy rules; and the data scientist is one person with 60% of her time already committed to fraud analytics. The team has six-week release cycles. A discovery workshop produced 47 ideas, and the head of sales has promised three enterprise customers 'personalized portals' in writing. Nobody has defined what 'personalized' means or how success will be measured beyond conversion.
1. How should the team frame the analysis across the Strategy, Initiative and Delivery horizons before committing to the 47 ideas?
Read the worked answer
The 47 ideas are solutions. Frame the work by horizon so each idea can be connected to an outcome before it consumes capacity.
Strategy horizon: clarify the CEO objective as an outcome (conversion 2.4% to 3.0%), define customer segments and the value proposition, and agree leading indicators plus guardrails such as consent compliance, margin, and page speed. Use the customer journey and personas to locate the biggest friction.
Initiative horizon: turn ideas into a small portfolio of hypotheses and minimum marketable features. Each hypothesis states the segment, the proposed change, the expected behavioural effect, the evidence needed, and the cost. Apply bang for the buck: learning per unit of cost. Legal's consent gap is a gating constraint, not a competing idea.
Delivery horizon: only for the hypotheses selected. Write stories with acceptance criteria and measurement instrumentation just in time.
Reasoning: this prevents the team from optimizing outputs such as landing pages or an ML engine, and keeps them focused on the conversion outcome. It also surfaces conflicts, such as marketing's campaign pages versus product's recommendation engine, as portfolio trade-offs rather than political fights.
2. What should be the first experiment, and what analysis supports that choice?
Read the worked answer
Choose the experiment with the highest learning-to-cost ratio against the riskiest assumption. The riskiest assumptions are: that personalization improves conversion, that consent compliance can be achieved quickly, and that the team can operationalize a data scientist who is 60% committed elsewhere.
A first experiment could be a rule-based personalization of the home page for one high-traffic segment, with a clear hypothesis, a control group, and a success threshold (for example, a 0.1 percentage point conversion lift at 95% confidence). Support it with a lean canvas, an assumption map, a sample-size calculation, and a consent review. The ML engine is a poor first experiment because it is expensive, depends on scarce data-science capacity, and does not test customer behaviour fastest.
Reasoning: agile analysis favours experiments that buy information cheaply. A thin slice also produces reusable instrumentation and a consent pattern. If the thin slice fails, the team learns before committing to the engine; if it succeeds, the result de-risks the larger investment.
3. The head of sales has made written promises to three enterprise customers. How does agile analysis help address this?
Read the worked answer
Treat the promises as stakeholder commitments that affect value and risk, not as an automatic scope mandate. First, analyse the three accounts: revenue at stake, churn risk, the specific personalized-portal outcome they expect, and whether a thin slice would satisfy them. Map them on an influence and interest grid and agree who owns the relationship.
Then make the trade-off transparent: show the roadmap, the current portfolio of hypotheses, and what would be delayed if the portal is pulled forward. Explore options: a customer-specific portal minimum marketable feature built on shared components; a manual concierge version for the first quarter; or a phased commitment with explicit assumptions. Bring sales into backlog refinement so promises are converted into hypotheses with acceptance criteria.
Reasoning: ignoring written commitments damages trust and revenue, but treating them as fixed scope bypasses discovery and can create waste. Agile analysis makes the cost of delay and the options visible, so the decision is made on value rather than escalation.