Oliver — Test Requirements Response

Responses follow fosferon's original document section by section, with the answer set out directly beneath each question.

1. The baseline to beat (priority)

Confirmed — 25% is the figure to beat. Based on a dataset of several thousand claims, drawn not only from PCP but from the same reactivation pattern across other financial-services sectors; the drop-off-to-reactivation behaviour is consistent across markets. For the test, Oliver should achieve 25% or higher — 25% is the proven baseline; the objective is to match at minimum and materially exceed.

One clarification: WhatsApp is the primary communication and conversion channel throughout, not the final step. Email and SMS are used earlier purely to warm the claimant up for the WhatsApp interaction, where the substantive re-engagement and conversion happens. The landing page is a three-page form; because it requires credit-search information, a degree of drop-off is inherent — typically 50–65%. A screenshot of the form will be provided so Oliver is tested against the real journey.

Economics are tied to each lead-generator / insurance partner's cost-per-acquisition (CPA). 50% of CPA is the target per case (e.g. £40 CPA → £20; £60 → £30) — the rationale being that the marketing spend is already committed and intent already expressed; we charge to recover an already-funded opportunity. 50% is the target; in some scenarios the achievable share is closer to 25%. Illustration: a client generating 1,000 claims/day with 25% reactivation on drop-offs would see ~1,250/day.

2. Domain knowledge to ingest

Existing material covering eligibility, process and relevant dates (including the redress timeline) is ready to share — part of the first tranche.

Compliance is structurally built in. We only work with FCA- or SRA-regulated firms, and the claimant gives an explicit opt-in on page one of the original form — that opt-in is the legitimate basis for contact. Content restrictions mirror standard FCA financial-promotion rules: no quoting/implying a specific payout, no certainty of outcome or timeline, nothing construable as legal advice. Happy to confirm formally in writing so Oliver's knowledge base is built against an agreed wording standard.

Full claim flow and FAQ content already exists and is published across the live websites — can be provided or referenced for ingestion. Little genuine objection-handling is required at this stage: the claimant has already opted in and begun the form, so it's recovery of an in-motion opportunity, not a cold sell. The only real "objection" is non-completion itself.

No issue — materials can be supplied in any accepted format.

3. Behavioural rules for this vertical

Confirmed — the rule-set is right for PCP-via-insurance, with one emphasis on tone: the priority is a conversation natural enough that the question of whether Oliver is AI rarely arises organically. This is consistent with the disclosure rule — if a claimant directly asks, Oliver discloses honestly, every time, without exception. As long as Oliver operates within FCA guidelines, every interaction remains compliant by design.

No legal advice under any circumstance; scope strictly to the PCP claim (or the relevant sector if extended). Off-limits topics are largely self-limiting if the knowledge base is scoped correctly. Preferred unknown-answer wording: defer to the build team's judgement — a sensible, natural response over a rigid script. Oliver may close unproductive conversations guided by conversational intelligence ("No problem — I'll check back in a few weeks…"); as a general principle, step back after ~1 month of no engagement.

4. A sample of the target data

Not a constraint — either a sampled/obfuscated slice of real records or synthetic data matching the real demographic shape can be provided. Ownership note: the underlying sample data sits with the insurance-company partner, not Costello directly. Costello will coordinate to obtain and provide the sample in the agreed format, and confirm a timeline once arranged. (For this tranche the data is PCP-specific; the same approach applies whole-of-market.)

5. Personality & resonance

UK-wide, drawn from insured customers generally — not skewed to a particular age band, region or profile. A more precise breakdown to follow ahead of the test if available.

Natural and flowing above all — genuinely human, not transactional: no acronyms, a grammatically intelligent register, warm rather than clinical. The aim is that the claimant trusts Oliver and feels able to return to him — a relationship, not a one-off — which is also a foundation for future referral conversations (treated as a separate later-phase scope).

6. Routing & outcomes

The claimant is routed back into their original resume link, continuing the exact journey they abandoned, through to signing. Once signed, the completed claim flows automatically into the existing law-firm / insurer / lender database via current infrastructure — no new routing required.

All conversations must be retained in full — a firm requirement, so a complete, defensible record exists for any future FCA review or investigation.

An intelligence layer should recognise when a conversation needs a higher tier of human support, with Oliver handling the substantial majority — around 80% — independently. Typical human-escalation cases: claimants wanting to review/query their claim in detail, and claimants with a poor prior claims-firm experience (who tend to want a human only once already engaged).

Next steps (per Costello)

Assemble the Section 1 baseline materials, source documentation and form screenshot as the first tranche, with the sample data to follow once coordinated with the insurance partner. Aligned on the 1 July target.