A bespoke system that takes a piece of IP and produces a defended, audited shortlist of brand partnership candidates. Internal-only. Designed around how Bridge actually works. Built to compound — every run sharpens the next.
Not more yeses. Quicker no's. Sharper rationale. More confident pursuit of the candidates worth pursuing.
One run = one piece of IP being profiled. Bad Fairies is a run. Dune Three is a separate run. Each run has its own folder, its own context, its own audit trail. Runs do not share state — but they share the underlying archive: case studies, accumulated brand intelligence, recurring precedents.
The unit of work is deliberately small. Starting a new run should feel cheap.
| # | Stage | Reviewer | Approx. time |
|---|---|---|---|
| 00 | Kickoff | — | — |
| 01 | IP Brief | Operator | ~5 min |
| 02 | IP Profile | Reviewer | ~5 min |
| 03 | Candidate Generation | — | — |
| 04 | Shortlist Audit | Reviewer | ~10–15 min |
| 05 | Handoff | — | — |
Setup. Create the run folder. Capture the basic identity of the IP: name, type (film / show / live event / other), client, expected release window, source of brief.
If a similar IP has been profiled before — same franchise, same genre, same tonal family — flag it. The earlier profile becomes context, not template.
The operator captures everything currently known about the IP through a structured intake.
Every field has an "I don't know" option. The system records confidence rather than forcing certainty. "Director attached: unknown — distributor mentioned 'name TBC' in the brief."
The system synthesises a working profile of the IP, even when the IP itself doesn't fully exist yet. Three independent sub-routines:
Comparable triangulation. Find 3–5 released properties genuinely close to this one. For each, surface what the marketing register was, who the brand partners ended up being, what worked and didn't.
Audience attitudinal mapping. Build an attitudinal signature — not just demographic ("women 30–45") but psychographic ("women 30–45 nostalgic for 90s feminism but cynical about its sequels"). Grounded in cultural signals, not invented adjectives.
Tonal sampling. Sample the comparables' actual marketing language — taglines, trailer voiceover, social copy. Build a register profile (formal/playful, sincere/ironic) the eventual partnership has to fit alongside.
Does this read as a fair synthesis? Does the confidence flagging match gut on what's solid vs speculative? Do the comparables surface useful proxies, or are they superficial?
The reviewer can: approve, redirect ("audience read is off, lean younger"), redefine ("scrap that comparable, it's misleading"), or branch the profile manually.
If the profile is wrong here, every candidate generated downstream will be wrong.
Two genuinely independent passes run in parallel.
Pass 1 — Category-led. Given the IP profile, what brand categories make sense? Who are the active players, who has budget, who is seasonally aligned, who has done partnership work in adjacent territory recently? Produces the safe matches — the candidates a competent human strategist would also reach.
Pass 2 — Attitude-led. Which brands share tonal DNA regardless of category? Greggs/Wicked is the canonical example: not category-adjacent, but the irreverent-Britishness register matched. Produces the surprising matches.
The two passes don't see each other's output. A third sub-pass enriches each candidate with current public context: recent campaigns, named marketing leadership, partnership history if findable. Confidence on enrichment is flagged separately from confidence on the match.
Collation assembles the shortlist: ~10–15 Safe Matches and ~5–8 Surprising Matches. Each candidate carries name, category, why-they-fit, confidence on the match, confidence on the enrichment, and any flags from Bridge's archive ("we worked with them on X in 2022") or exclusions ("currently in a category exclusivity with Y client").
No reviewer checkpoint here. Surfacing candidates mid-generation invites premature judgement. The reviewer sees the shortlist as a complete artefact at the next stage.
The most important review point in the system. The reviewer goes through candidates one at a time. Three verdicts per candidate: Approve (proceed to handoff), Reject (out, with optional reason — reasons compound into precedent), Override (adjust the rationale, the tier, or the framing before approving).
Single-keystroke verdicts. Reasons captured by voice or short text. Twenty candidates in 10–15 minutes. The reviewer can also add candidates the system missed, promote a Surprising Match to Safe (or vice versa), or flag for deep-dive before outreach.
The audited shortlist gets formatted for whatever happens downstream — internal outreach planning, a client deck, a CRM. Format depends on what the team actually needs.
The handoff also captures what would close the loop: when a candidate eventually responds (yes / no / no-reply), that response becomes evidence. Even if outreach itself happens outside the profiler, the system records outcomes against candidates. This is what makes the moat compound — a year of run data is a year of validated/invalidated rationale.
Three structural choices, taken together.
A single prompt asking "find good candidates" produces obvious answers. Running category-led and attitude-led separately, with neither seeing the other, is what makes surprise candidates surface alongside safe ones — without the surprises being parlour tricks.
"Greggs fits because Wicked-style irreverent Britishness aligns with the IP's tonal register, and Greggs has historically punched above its weight on cultural-moment partnerships." Auditable. Arguable. Capturable.
Every Bridge case study, past pitch, brand the team has formed an opinion on — that institutional knowledge feeds the system as context. "You worked with this brand on something adjacent in 2022." This is the moat. GWI doesn't have it. 2010 doesn't have it.
When the system isn't sure, it says so. Low-confidence candidates carry the flag. Thin enrichment data carries the flag. The reviewer always knows what they're working with.
Day one, the universe is the live web. The profiler uses web-grounded discovery to surface candidates. Works on day one, picks up brands that did something interesting last week, isn't constrained to a curated list. Can't claim full comprehensiveness — the team will sometimes catch candidates the system missed.
Over time, the universe becomes Bridge's. Every audited candidate adds to a structured record. By run twenty, the system knows things like "this brand has been considered for four IPs, approved on two, pitched on one, no response. Last considered six weeks ago." By run fifty, this is genuinely proprietary — comprehensive against Bridge's actual operational scope, augmented with live discovery for everything outside it.
Comprehensiveness is a property the system grows into, not one it has on day one.
Two artefacts accumulate across runs and shape future ones. The brand intelligence layer — every candidate that's been audited adds to a structured record on that brand. By run 20, the system has real institutional memory. The precedent layer — patterns recurring across multiple runs become rules the system applies by default. "Beer brands respond better when there's a clear social-occasion hook in the IP."
None of these get built before the foundation is proven.
Run the system in Claude Code on two or three real Bridge campaigns. The point isn't to give Rich a tool yet — it's to validate the matching. Iterate the process based on what we find.
Same environment, but now Bridge runs it. Tests something different: where does the process assume context only Charlie has? Where do the prompts need to be more guided?
Once the sequence is stable across operators and runs, port it to a portal the wider team can access. The portal becomes a thin presentation layer over a process that's already proven.