STEP 1
Connect once1
Purchase a plan, then follow the connect guide for your assistant. Supported at launch: Claude hosted, Claude Code, Codex CLI.
HOW IT WORKS
Decision support for Walmart suppliers, Sam’s Club suppliers, and Walmart Marketplace sellers.
You connect Retail Reason to a supported AI assistant. It helps diagnose an operating problem, answer an operational question using maintained, dated material, and check a plan before anyone acts on it. It does not connect to retailer portals or pull data from customer systems; it processes only the question, plan, draft, or context you choose to submit.
THREE STEPS
Connect once. Ask in context. Act with the limits visible.
STEP 2
Ask in plain English2
Mid-task, in the tool you already work in. A question, a problem you’re diagnosing, or a plan you want checked before it goes out.
STEP 3
Act on it
Every answer arrives dated and classed. No person sits between your question and the answer.
WHAT IT IS, AND ISN’T
Three things this is not.
Not a general assistant
Keep the assistant you have; Retail Reason is what it connects to. The difference is maintained, retailer-specific judgment across the supported operating areas and explicit uncertainty: reference material a practicing advisor re-verifies and dates, answers that carry a confidence class, and a service designed to surface uncertainty or refuse when the available evidence is insufficient.
Not a consultant
It carries the recurring reference and diagnostic layer of the work: what changed, which system owns a task, what a program actually measures. Judgment, negotiation, and leading the client through it stay human, whether that human is you or the advisor you retain.
Not another dashboard
It complements the portals and tools you already run. It does not connect to or continuously ingest data from retailer or customer systems, monitor those systems, or automate actions in them; it processes only the questions, drafts, and context you choose to submit so it can answer questions and check plans.
You bring the situation. It brings what a practitioner knows about the system you’re up against, dated.
ONE ANSWER, IN FULL
A worked example.
PASTED PLAN
“Pull category share by subcategory in Scintilla and build our line-review deck off it. We’re a 1P supplier in Dept 92 on the Basic plan.”
PLAN CHECK · RANKED FINDINGS
FINDING 1
Category share is a Charter capability. On Basic the module isn’t there to open, so the deck’s share section needs Charter or outside data; decide which before the build starts.
FINDING 2
Line-review timing follows the category’s modular reset cycle; Dept 92’s window sets when this deck is due, and that date is Walmart’s calendar, not the plan’s.
FINDING 3
Scintilla carries two store-count bases (POS-store and valid-store), and per-store numbers move with the choice; the slide should name which basis it uses.
Caveats: the store-count basis in finding 3 is time-sensitive; re-check it in the portal before quoting per-store numbers.
CONFIDENCE NOTE
The confidence class applies to the whole answer: when any part touches time-sensitive material, the answer carries the cautious class and a Caveats note names the part. Classes are explained below.
EXAMPLE STATUS
A hand-authored example in the delivered answer format; not a customer transcript.
CONFIDENCE & METHODOLOGY
Confidence classes, defined.
Every answer carries exactly one confidence class and an “as of” date. The class is set by the state of the reference material behind the answer, not by how confident the model sounds.
Verified this cycle
The reference material behind the answer was re-checked in the most recent verification cycle, and the as-of date is attached.
Dated: re-check before quoting
The answer touches time-sensitive material (rates, schedules, thresholds). Accurate as of its date; confirm in the portal before you quote it.
Inferred
Reasoning from how the pieces fit, not a confirmed fact; flagged so you treat it accordingly.
What “as of” means
The date on an answer is when the reference material behind it was last verified, not when the answer was generated. Matt monitors relevant official documentation and release notes, re-checks higher-volatility facts before relying on them, and updates a content record’s date when he completes its re-verification. Stale or time-sensitive details are called out. This is not continuous monitoring and does not promise that every retailer change is caught as it happens.
The current verification index shows the latest recorded verification date for each public article. It is a current index, not an append-only release history.
When information conflicts, and when it refuses
Where the honest answer is unsettled, or reliable readings disagree, the service is designed to say so instead of picking a side quietly. When the available evidence is insufficient, it is designed to refuse rather than present a guess as a supported answer.
Corrections
If an answer was wrong, tell me: [email protected]. Matt verifies the issue against the applicable source, corrects maintained material when warranted, and updates the affected public record’s verification date when that article is re-verified. Known gaps stay listed on the limitations page.
What the labels do not guarantee
A confidence class tells you how the material behind an answer was maintained; it is not a warranty that the applicable retailer has not changed something since the date shown, and it does not replace your own read of your agreement or your merchant relationship. Verify anything you rely on against the applicable retailer’s official systems before acting.
READY WHEN YOU ARE
Choose the plan that fits the way you work.
Plans and purchase are on the pricing page; the complete coverage map is on the coverage page.