For decisions that cannot wait for another guess.
- A marketplace launch: decide how to frame and package the offer before spending on distribution.
- Competitive displacement: learn why an alternative is selected instead of assuming price is the problem.
- A pricing change: compare the buyer’s outcome and total job cost, not just the unit rate.
- Selection without success: find whether the promise, access or execution path is breaking.
Customer research shapes the variants.
Jobs-to-be-Done research asks what progress the customer wants, in what situation, with which constraints. We review available interviews, sales conversations, reviews and customer feedback to understand the job and its alternatives.
Quantitative research examines available usage, funnel, cohort, survey and delivery-cost evidence. Pain-point mining and sentiment analysis help organise market language and attitudes. These inputs generate hypotheses; they do not prove an agent will prefer an offer.
Sources, access and research depth are agreed in scope. New interview programmes, extensive data collection and additional instrumentation are separately scoped.
The offer is a system, not a headline.
- The job: a specific outcome and the conditions under which it counts as success.
- The promise: capabilities, expected outputs and constraints the product can actually honour.
- The commercial structure: pricing units, packaging, caps, failure exposure and total job cost.
- The evidence: documentation, output examples and credible proof that make comparison possible.
- The experience: the path from selection through access, successful execution and first value.
Is this the decision your team needs to make?
Share the product, current friction and target outcome. We’ll establish fit and agree a useful scope before paid work.
Offer Experiment Sprint
From $7,500 USD. Four weeks after required access and scope agreement. One priority job and channel, three offer variants, controlled evaluation and an implementation-ready recommendation.
Payment is 50% upfront and 50% at handoff. Production rollout, extensive instrumentation and additional usage costs are scoped separately.
- Week 1: establish the job, alternatives, baseline and economic constraints.
- Week 2: construct the variants and predeclare the decision criteria.
- Week 3: run bounded evaluation, robustness and execution checks.
- Week 4: analyse results, review limitations and hand over the recommendation.
What we test.
Market sentiment and customer language supply hypotheses. They do not establish how an agent will respond. That connection is tested.
- Job and outcome framing, offer eligibility and documentation clarity.
- Pricing units, expected total job cost, caps and failure billing.
- Evidence, guarantees and constraints that can actually be honoured.
- The buying path for humans and execution path for agents.
How we keep the result honest.
Use credible alternatives, varied presentation order, documented contexts and versioned runs. Keep losing offers, failed executions and excluded runs in the record.
Repeated model runs may be correlated. We report the actual unit of analysis and avoid treating every response as an independent customer. Uncertainty and statistical power depend on the design, not the confidence of the prose.
New research and implementation use Codex-native execution. Historical records can include earlier multi-model work; those records are described with their original limitations.
What the readout answers.
A sandbox preference result is not live marketplace share or realised revenue. Human conversion claims need measured customer behavior. Where feasible, the handoff includes the design for a live pilot or controlled customer test.
- Which offers were eligible, selected and successfully executed?
- What changed across jobs, budgets and contexts?
- What is the total cost and contribution margin for a completed outcome?
- What can we recommend now, and what needs live validation?