Skip to main content
The UNFI AI agent is in alpha and only available to UNFI Insights suppliers. Reach out to your Crisp contact before sharing this page externally.
The UNFI AI agent analyzes UNFI distribution data through two channel-specific agents — Natural and Conventional — each backed by its own semantic model and scoped to your UNFI Insights entitlements.

​
What the agent supports

  • Distribution expansion and whitespace analysis
  • Inventory and weeks-of-supply assessment with stock-out and overstock risk
  • Daily and weekly transaction data
  • Store and DC-level detail
  • Purchase order and payables lookups by PO number — Group, filter, and report purchase order and open payables data by UNFI PO number, so you can trace a specific order end-to-end or reconcile a payables line back to its originating PO.
  • Retailer-level access controls and supplier-intersection filtering
  • Weather-impact analysis — Ask whether a weekly sales miss lines up with observed weather. Both Natural and Conventional agents join temperature, precipitation, snowfall, wind speed, humidity, sky coverage, UV index, heat index, wind chill, and wind direction to store-week sales, so you can group performance by week, store, state, or region alongside the weather that week. Keep in mind that wind chill is only reported when it’s cold enough and heat index only when it’s warm enough, so you’ll see gaps in those fields during spring and fall. Results are framed as correlational context (not causal), and the agents draw on observed history only — forecast weather isn’t included.
  • Background on UNFI as a company — scale, distribution network, programs, and capabilities
  • A built-in 6-lesson onboarding tutorial for users who are new to AI agents

​
Getting started with a guided tutorial

If you’re new to the UNFI agent or to AI agents in general, ask the agent to walk you through onboarding. The agent ships with a self-guided, 6-lesson program that you can run at your own pace — one lesson per session or all in one sitting. Trigger the tutorial with any of these prompts:
  • “Where do I start?”
  • “Can you teach me how to use AI agents?”
  • “Give me a tutorial.”
The agent first asks for a short profile — your role, what eats up your week, and the one thing you want the agent to help with — then personalizes the rest of the program around that use case. The lessons cover:
  1. AI fundamentals — what the agent is and how it works with your UNFI Insights data.
  2. Prompt writing — context, specificity, examples, and format instructions.
  3. Investigating AI outputs — asking for confidence checks and keeping a human in the loop.
  4. Build and refine your use case — iterating on a prompt for your real work.
  5. Edit and schedule — saving a prompt and putting it on a recurring schedule.
  6. Close and share — confirming your first automated report and choosing what to try next.

​
Asking about UNFI as a company

The agent can also answer questions about UNFI itself — useful background when you’re orienting a new team member or framing analysis for a stakeholder. Ask things like:
  • “Give me an overview of UNFI.”
  • “What programs does UNFI offer to suppliers?”
  • “How big is UNFI’s distribution network?”
Responses cover UNFI’s scale (annual sales, SKU count, customer locations, warehouse footprint), distribution banners (Natural, Conventional, Produce, Wellness, Canada, Nor-Cal, Tony’s, Woodstock Farms), supplier programs, and sustainability goals. For pricing, contract terms, or your specific account, the agent still routes you to your Supplier Relationship Manager.

​
Choosing between Natural and Conventional

Pick the agent that matches the channel you want to analyze:
  • Natural — UNFI Natural distribution data sourced through UNFI Insights.
  • Conventional — UNFI Conventional distribution data sourced through UNFI Insights.
Each agent uses a dedicated semantic model, so switch agents when your question shifts channels rather than asking one agent to reason across both.

​
How the agent handles common analysis inputs

The UNFI agents apply a few defaults so you don’t have to spell them out on every prompt.

​
Weekly output uses calendar dates

UNFI runs on a Sunday-start fiscal week. When you ask for weekly results — tables, charts, or CSV exports — the agent labels each row with the YYYY-MM-DD week start date (a Sunday) instead of the internal YYYYWW code. Ask for the “full date range” and it also includes the week end date (the Saturday). The YYYYWW code is still used behind the scenes for filtering and year-over-year comparisons, so results align with UNFI dashboards. If you specifically need the UNFI YYYYWW week code in the output (for example, to match an existing report), ask for it explicitly:

​
Product taxonomy defaults to your MDM

When you mention a category, subcategory, or product, the agent searches your MDM taxonomy first — the same product master your team uses in other Crisp tools. UNFI’s own category, class, and subclass fields are only used when you ask for them by name.
  • Default: Ask about sales in the "Snacks" category → matches on your MDM category.
  • Opt in to UNFI taxonomy: Show sales for the UNFI category "Snacks" (or “UNFI class”, “UNFI subclass”).
If your term could belong to either taxonomy, the agent asks a single clarifying question before running the query.

​
Weeks of supply defaults to the demand forecast

When you ask about weeks of supply (WOS) or your inventory position, the agent answers with Forecast WOS — the same measure shown on the Distribution Center Inventory dashboard. Forecast WOS walks your latest on-hand inventory forward week by week against UNFI’s weekly demand forecast for the next 26 weeks, so the answer reflects forward demand (including seasonality and promotions) rather than recent sales pace. If on-hand inventory outlasts the full horizon, the agent reports it as >26 weeks. Two related defaults:
  • Demand Forecast Quantity — when you ask for the forecast alongside inventory, the agent reports the 4-week average weekly demand forecast, matching the dashboard’s Demand Forecast Quantity column. Ask for a specific week (for example, “forecast for the week of October 4”) to get week-by-week forecast values instead.
  • Historical WOS — the previous sales-based calculation (on-hand inventory divided by average weekly sales) is still available. Ask for “historical,” “sales-based,” or “trailing” weeks of supply to use it.
A product and DC with no published demand forecast would also read as >26 weeks, so the agent cross-checks Demand Forecast Quantity. When that value is 0, the agent tells you there is no forecast demand to cover instead of reporting a 26-plus-week supply.

​
Store number disambiguation

Store numbers and source store numbers can repeat across UNFI DC regions, so a bare number like “store 46622” is ambiguous. If you provide only a store number — without a DC region, city, or state — the agent runs a quick preflight to see which regions carry that number:
  • If exactly one region matches, the agent applies that region filter and continues.
  • If multiple regions match, the agent stops and lists the choices before running the full analysis, so a shipment or inventory number never gets attributed to the wrong location.
To skip the disambiguation step, include a region, city, or state alongside the store number:

​
Schedule delivery

Schedules that run a UNFI agent deliver by email.

​
Governance and guardrails

The UNFI agents follow a governance policy that keeps responses scoped to your UNFI Insights data:
  • Confidential retailers — Whole Foods and other confidential retailers stay masked. The agent includes their volume in top-line and regional rollups but will not identify or unmask individual stores. Contact UNFIInsights@UNFI.com for unmasking options.
  • Cross-distributor and cross-brand questions — The agent only sees your brand’s UNFI data. It will not pull KeHE numbers, use UNFI data as a proxy for other retailers, or surface another supplier’s performance.
  • Entitlement and payables figures — Open payables, MCB, and UNFI-owed balances are surfaced with a “verify through UNFI official channels” disclaimer. The agent informs but does not confirm definitive owed amounts — check with your UNFI rep before acting on a figure.
  • Pricing, contract terms, and UNFI policy lookups — Routed to your Supplier Relationship Manager (SRM) instead of being answered or web-searched.
  • Off-topic and advisory requests — Investment advice, general-purpose tasks (writing emails, building apps), and other non-UNFI work are declined so your monthly token allowance stays focused on analysis.