gabrielwang.ai ← back to the workshop
Edition 002 · interactive demo

The accounts that went quiet — ranked by what's at risk.

Marlin Base Metals is a (fictional) Singapore trading house moving aluminium billets, copper cathode and zinc ingots to ~60 industrial counterparties across Southeast Asia. Every counterparty buys on its own rhythm — some weekly, some quarterly. This desk sheet flags the ones who have gone quiet relative to that rhythm, ranks them by revenue at risk, and drafts the outreach. It re-runs entirely in your browser on synthetic data — no upload, no server, no live market feed.

Honest-AI note. The detection and ranking here are deterministic analytics, not an LLM. The AI layer is the outreach note on each account — those were drafted by Claude at build time from each counterparty's own trade history and shipped as static text. In production this runs per-counterparty on live data. Nothing on this page makes a live model call, so there is no fake "thinking…" spinner.

Plain-language key (counterparty, T12M, indicative, MT)
Counterparty
A firm the desk sells to — here, industrial buyers of metal.
Cadence
How often an account normally buys (its median gap between orders).
T12M revenue
Trailing-twelve-month revenue — what the desk billed this account over the last year.
Indicative / re-quote
The current ballpark price the desk would quote, adjusted for this account's usual premium or discount.
MT
Metric tonne — the unit metal is traded in.

Revenue at risk right now

Dormant counterparties
flagged at current sensitivity
Revenue at risk (T12M)
trailing-12-month revenue of the flagged set
Largest single account
biggest one at risk

Why not just "no order in 90 days"?

flagged by a fixed 90-day rule
of those are false alarms — seasonal buyers still inside their normal gap
missed by the 90-day rule — fast-cadence accounts that genuinely went quiet

A fixed calendar rule is wrong in both directions. Baselining each account against its own cadence flags the right set.

Contact these 15

Ranked by trailing-12-month revenue, discounted by how long they've been silent.
See the rest of the series →

Show your work

Why the other counterparties aren't on the list

The credibility of a "call these" list is in who it correctly leaves off. At the current sensitivity: