RetrofitMarket sits between two anxious audiences: homeowners researching retrofit and installers who need predictable, qualified work. Before this engagement, matching lived in spreadsheets, ad hoc calls, and manual follow-ups. That capped throughput and made it impossible to market the service nationally with a straight face.
Context and constraints
The product had to feel credible on day one: accreditation cues, clear geography, transparent next steps, and a path from enquiry to installer contact that did not feel like a black box. Performance mattered because traffic would arrive from search and partner referrals, often on mid-tier phones on patchy connections.
We aligned early on three non-negotiables: trust, throughput, and operator sanity. Everything else was negotiable.
IA and the marketplace spine
We structured discovery around jobs homeowners actually articulate (property type, urgency, geography, rough scope) rather than internal installer taxonomy. Installer-facing profiles mirrored what homeowners need to compare: coverage, certifications, response norms, and proof of past work categories.
Search and filters were kept deliberately narrow at launch. Extra facets ship better once baseline conversion is understood.
AI-assisted matching
Matching was never positioned as a gimmick. The goal was to route enquiries faster while preserving human oversight where installers wanted it. We implemented quote-request flows that collected structured signals, scored compatibility against installer rules, and surfaced explainable rationale to admins.
Operators could audit decisions, override, and tune thresholds without engineering deploys for every tweak.
Installer onboarding and verification
Onboarding was rebuilt as a guided pipeline: document capture, checklist completion, and staged activation by region. Automated reminders and status dashboards replaced slack messages chasing PDFs.
That reduced time-to-active listings and made expansion into new postcodes a repeatable playbook rather than a project each time.
Measurement
We report directional uplift using indexed metrics and operational indicators agreed with the client — not revenue figures on the public page. Lead throughput and response-time indexes moved sharply once routing and onboarding stopped competing with manual work.
What we would do earlier next time
Earlier instrumentation on drop-off between matched hand-offs and installer acceptance would have shortened one optimisation cycle. We added it in phase two.