Retail & distribution · Case study
Retail B2B Agent Foundation
One shared foundation, connectors, entity resolution, identity and authority, an agent runtime, so decision agents deploy across support, sales, and pre-sales without starting over each time.
One foundation · Three agents
Three use cases, from production to in-build, on a single substrate.
The first laid the foundation; each one after went faster.
Support, sales and pre-sales agents for one client, all on the same foundation. Each new agent reuses the connectors, resolved entities, and identity and authority rules the last one built, so it starts from working parts, not from scratch. Below, the pre-sales quote builder.
Case study · Quote builder · Pre-sales
Implemented
From competitor invoice to defensible quote.
Australian B2B office-products distributor · details anonymised
09:14, Tuesday. A prospect emails a four-page competitor invoice: “can you beat this?”
The problem
Answering means a person reading the invoice, matching every line against a 50,000-SKU catalogue, choosing margins, clearing pricing exceptions up the chain, and assembling a quote: 24 to 72 hours of heads, spreadsheets and email threads. Instant online pricing is taking the low end; the distributor’s edge is a defensible, relationship-priced quote, but the process is too slow to defend it. And when a discount is granted, there is no clean record of who authorised it, or against what margin.
What we built
A single reasoning agent embedded in the four-stage quoting workflow the client’s leadership validated: intake, match, price, assemble. It reads whatever the customer sent into a structured demand profile, maps every line across five confidence tiers with its reasoning shown, builds three margin scenarios, and routes pricing exceptions through an explicit authority chain, from account manager to sales lead to owner. Pronto Xi ERP stays the system of record; on close, a sales order is written back to it.
The account manager is the accountable human at four defined gates, not an orchestrator of agents.
Every pricing exception carries who authorised it, against which margin, and why.
Every override is captured as labelled signal. The pricing model compounds with each deployment.
Business outcomes
Design target
Hours, not days
Quote back before lunch, against a 24–72 hr baseline.
Implemented
50,000+
SKUs matched with visible reasoning across five tiers.
Structural
Auditable
Every exception carries authoriser, margin and rationale.
Structural
4 gates
Confirm, override, escalate: accountability explicit.
Moat
Compounds
Every override becomes proprietary training signal.
Why it holds
The reasoning agent sits inside a deterministic workflow, deliberately not a swarm, and the override corpus is proprietary to the client’s pricing context. The speed beats instant online pricing; the auditability is what instant pricing can’t offer.
Stack
GCP · Vertex AI · Gemini · Pronto Connect API · Decision Core