Experimenting with conversion

How do you turn a friction point in the customer journey into a testable hypothesis, then make a decision based on verified results?

1960 € per session, excl. VAT

In short

In one 7-hour day, excluding breaks, you structure an approach to marketing experimentation: identifying a friction point, prioritising hypotheses, preparing an A/B test and interpreting its results. Generative AI helps formulate variants and challenge the protocol, without replacing statistical checks or business judgement.

Why this training?

Changing a button or an argument without an explicit hypothesis makes results difficult to interpret. An apparent increase may stem from traffic, seasonality or a measurement flaw rather than the variant. Generative AI makes it easier to explore objections and draft variants. Its contribution becomes usable when each proposal is linked to an observation, an expected mechanism and a decision rule defined before the test. Running case study: Orbe Papeterie, a fictitious organisation selling customisable notebooks, observes customers dropping out between customisation and payment. Using a documented customer journey, fictitious customer comments and simulated data, participants examine a hypothesis: making the delivery timeframe visible earlier could help customers complete their orders. Transfer to the workplace: On your return, select a real friction point with the marketing team, have the protocol validated by the person responsible for data, then schedule a decision review at the planned deadline, without an automatic launch.

Objectives of this training

  • Distinguishing an observation of the customer journey from an assumed explanation
  • Formulating falsifiable conversion hypotheses
  • Prioritising hypotheses by evidence, expected impact and effort
  • Defining an A/B protocol and its decision rules
  • Interpreting simulated results without overstating conclusions

Who is it for and what are the prerequisites?

Roles and tasks concerned

  • Managers responsible for acquisition and conversion in e-commerce customer journeys
  • Product managers and marketing analysts involved in experiments

Prior skills

  • Regularly use a generative AI assistant with structured instructions.
  • Know the stages of your funnel and the associated marketing metrics.
  • Be able to calculate proportions in a spreadsheet.

What you will work on

  1. Locating the friction point

    Objective: Choose a measurable conversion stage. Funnel events, denominators, tracking quality and the distinction between correlation and causation.

  2. Formulating and challenging hypotheses

    Objective: Make each proposal refutable. Observation–change–mechanism–metric structure; confirmation bias and missing evidence.

  3. Prioritising and drafting the variant

    Objective: Select a change to which the effect can be attributed. Reasoned prioritisation; consistency of the promise; the risk of changing several factors simultaneously.

  4. Locking down the protocol

    Objective: Set the rules before observing the results. Randomisation by user, eligible population, primary metric, guardrails, minimum detectable effect, statistical power, volume and duration.

  5. Reading the results without rushing

    Objective: Separate actionable signals from uncertainty. Absolute and relative differences, confidence interval, allocation imbalance, multiple comparisons and early stopping.

  6. Deciding and capturing learning

    Objective: Justify an operational next step. Adoption, abandonment or an inconclusive result; business implications and archiving of lessons learnt.

What you take away

  • Prioritised register of hypotheses with supporting sources
  • Variant specification with business justification
  • A/B protocol and metrics dictionary
  • Decision paper on simulated results

Method and assessment

  • Short instructional sessions followed by immediate practice on the running case study.
  • Individual work followed by work in pairs, with peer review of outputs.
  • Guided group discussion: what works, what needs correcting and what remains a human decision.
  • Preparation: Choose a customer journey to examine and describe its target conversion precisely. Prepare anonymised material only, or use the fictitious dossier supplied; check which tools the organisation permits.

Observable criteria: Each hypothesis links an observation, a modification and a measurable effect. ; The variant isolates an interpretable change. ; The protocol specifies the population, allocation, duration, required volume and guardrails. ; Rates and differences are recalculated outside AI. ; The conclusion distinguishes the estimated effect, uncertainty and limitations..

Frequently asked questions

Is a real test launched during the day?

No. The protocol and analysis are based on a fictitious case study and simulated results.

Does AI calculate the reliability of the result?

AI helps with explanations; calculations are checked in the spreadsheet using a documented method.

What can you do with low traffic?

Assess feasibility before launch; prioritise qualitative investigation if the required volume is unattainable.

Can you use internal data?

Only anonymised data, and only in authorised tools; no raw personal data is required.

Last updated: