Competitor roster

This is a competitor roster for a Zoom seller who needs to know, before a call, who else is in the deal and why. The main screen ranks 20 companies against Zoom — rank 1 first, the five pinned names held at the top, everything below them ordered by alignment. Each row carries Trayo’s verdict (direct, adjacent, not a competitor), a product-alignment and an ICP-alignment score out of 100, and the sentence explaining the placement. The filter splits the list by where a row came from: pinned by the rep, surfaced from the similarity pool, or named by the model during generation. Clicking a row opens the full record — provenance, both scores, the similarity figure and the seeds it matched, and the status of Trayo’s scan of that competitor’s own customer page. A second route, /method, lists every API call behind the screen.

Trayo supplied all of it. trayo_get_workspace returned Zoom’s onboarded positioning — 14 solution cards and 5 buyer personas — which was condensed into the positioning string sent with the roster call. The roster itself came from one trayo_find_competitors call against companyId 1123141 with five pins (Microsoft, Google, Webex, RingCentral, 8x8, taken in that order from the workspace’s own buyerProfile.competitorTerms) and cap 20: it took 15.7 seconds, drew a 150-company similarity pool, judged 137 candidates in 6 batches, and returned 20 entries — 5 from pins, 14 from the pool and 1 (GoTo) nominated during generation, with 0 unresolved nominations. All 20 were judged direct competitors, with product alignment between 90 and 98. A repeat call with the same pins came back instantly with reused: true and identical rows, as the recipe describes. Company logos are resolved by the kit from each domain against Trayo’s brand-image endpoint.

Two things went wrong and are worth knowing. The recipe’s first step, {“company”:{“self”:true}}, resolves this workspace’s website zoom.us — and Trayo files zoom.us under Unity Bank Plc, a Nigerian bank. That first roster ranked Jaiz Bank Plc as a direct competitor and judged all five pins not_competitor on the grounds that none of them is a bank. It was discarded; Zoom’s real record was found with trayo_find_companies (zoom.com, companyId 1123141, 14,062 employees) and the roster regenerated against that. Second, Trayo starts a background scan of each competitor’s own site for named customers; after twenty minutes of re-polling, 19 of the 20 were still pending or running, so the detail panel reports the scan status rather than inventing logos. Microsoft’s scan had completed and found no named customers. Push to Salesforce, Share to #compete, Export deck and Rebuild roster are stubs — no CRM, Slack or document service is connected — and each confirms with the kit’s preview toast naming exactly what would have been sent. On brand: brandFromWorkspace read #4A8CFF as the primary and #FFFFFF as the page, but gave #00031F a chart-accent slot rather than the ink slot, because it carries enough hue not to count as a neutral. With no ink supplied, applyBrand derived the top bar from the primary (#003995), tinted the white canvas to #F7FAFF, and darkened the primary to #1A5BCA for accent text and focus rings, which #4A8CFF does not pass contrast for on a light page.