← Back to the fleet FTE-07 · AI Card-Deals Concierge

PerkPilot

Every card-linked food deal, cracked in a tap. Enter a customer's card and instantly see the exact bank-card discount that applies today — a deployed lookup app now, being rebuilt around an autonomous agent that keeps every co-branded campaign live, validated and instant.

MVP · deployed Fintech · card deals Next.js · serverless Postgres AI-native 0.5/5 FTE-fit 3/5

Card → deal

<1sBIN to live discount

Data model

3-tierdeal → entry → BIN

Access

OAuthadmin whitelist

Deploy

Liveserverless Postgres

What runs today vs. the roadmap

Running today

A deployed lookup app — deterministic, no AI yet.

✦Type a card, see the deal — staff enter a card BIN (4–8 digits) and instantly get the eligible discount by bank × network × tier, on a date selector that defaults to today.
✦Admin console — full create/edit for deals and the reference data (banks, card networks, tiers), with weekday windows and discount caps.
✦Customer offers page — a public, QR-ready one-pager showing the day's active deals, cached for the counter.
✦Real accounts & analytics — Google sign-in with an admin whitelist, plus per-page view analytics.
✦Deployed with live campaign data — running on serverless Postgres, loaded each seasonal and quarterly window from the bank's alliance files.

On the roadmap

The autonomous Deal-Ops FTE we're building.

◆Self-serve ingestion — an agent watches an inbox/drive for the bank-alliance PDF/XLSX and extracts every deal, window, BIN and discount — retiring the hand-coded import each campaign.
◆Validate every BIN — resolve inconsistent bank/card naming and check each BIN against a real IIN reference, flagging mismatches before they cause a wrong discount at the till.
◆Reconcile & publish — diff against live deals, propose adds/edits/expiries for one-click human approval, then publish and refresh the QR page.
◆Customer concierge — answer "what's the best card deal this Friday?" and surface the single best eligible deal in the moment.

Honesty note: there is no AI in the product today — it's a deterministic BIN→discount lookup. Everything in this column is the roadmap.

The role it replaces

A payments-ops analyst wrangling deal files.

Each campaign, someone reads bank-alliance PDFs and Excel, normalizes inconsistent naming, extracts hundreds of BINs, and a developer hand-codes an import — the whole ingest-and-normalize loop an agent would own.

~0.15FTE + recurring dev time · target

The 60-second showcase

veloce://fte-07 — card-to-deal

A diner's card BIN is entered at the counter (or scanned from their card).

In under a second, staff see the exact live discount for that card — bank, network, tier and % — so it's charged correctly.

The roadmap agent keeps that list correct: it ingests the next bank file, validates every BIN, and publishes after one human approval.

PerkPilot is the honest inverse of the fleet's AI-first units: the product — a fast, deployed card-to-discount lookup with a real admin console and customer offers page — already works and has real users, but it carries no model. The FTE is the layer above it: an autonomous Deal-Ops analyst that ends the quarterly PDF-and-Excel scramble, validates every BIN, and keeps the right deal in front of every diner. That upgrade path is the build.