AI that's grounded in your lots, your orders, your customers.
Forecasting from your demand history, AI-suggested POs from your reorder points, and an operator chat assistant that reads your data — never a hallucination, always a citation.
Generic AI knows nothing about your warehouse.
A chatbot bolted onto an ERP can write a poem about beef striploin but can't tell you how many cases are expiring this week, which customer is past due, or what to order from your produce supplier on Tuesday.
APFoods AI runs on your data: lot inventory, demand history, AR aging, customer behavior, supplier lead times. Every answer cites a source row. Every recommendation has a number behind it.
What's in the AI assistant.
Four capabilities, all grounded in the operating data you already have.
Demand forecasting
Category, customer, and SKU-level forecasts updated daily. Seasonality, trend, and customer-cycle aware.
AI-suggested POs
Reorder points + forecast + lead time + on-order = a draft PO every morning. Operator reviews and sends.
Operator chat assistant
"What's expiring this week?" "Why is Trader Joe's order on hold?" Grounded answers with citations to source rows.
AI forecast insights
Daily plain-English summary of revenue trends, expiry alerts, top-performing SKUs, and anomalies.
No hallucinations
Every answer is grounded in retrieved rows. The assistant says "I don't know" when the data isn't there — never invents.
Tenant-isolated
Your data never leaves your tenant. No cross-tenant retrieval, no shared embeddings, no training on your customer list.
What to buy, how much to order, when to act.
The forecasting screen is where the AI tells you what's coming. Category demand for the next two weeks, customer churn signals, expiring SKUs that need a push, top-performers worth promoting. Numbers, sources, and a recommended next action.
- Demand forecast at category, customer, and SKU level
- Expiry-aware true demand — not just velocity, but velocity that accounts for what's about to write off
- Customer churn signals based on order cadence drift
- Margin alerts when a SKU's GP slips below tolerance
Tomorrow's purchase orders, drafted by the AI overnight.
Every morning, the AI scans reorder points, forecast demand, on-order quantities, and supplier lead times. The result is a queue of draft POs grouped by supplier — quantities, prices, and rationale all visible. Approve as-is, edit, or discard.
- One draft PO per supplier with line-item rationale
- Reorder-point breach signals visible per line
- Lead-time aware — supplier with 7-day lead time gets ordered 7 days before stockout
- Operator approval required — AI never sends a PO by itself
AI Assistant FAQ
Does the AI ever auto-act on my behalf?
Not by default. The AI drafts, suggests, and recommends. A human approves every PO, every customer email, every order release. The "auto-act" flag exists for some workflows (e.g. auto-resolve catch-weight variance within tolerance) and is always tenant-configurable.
What model is behind the AI?
Best-in-class commercial models (OpenAI / Anthropic) for reasoning, plus dedicated forecasting models for demand. We route different tasks to different model strengths.
Is my data used to train the AI?
No. Tenant data is never used for model training. We use private retrieval (RAG) over your tenant data with strict isolation.
How accurate is the demand forecast?
MAPE (mean absolute percentage error) varies by category and history depth. For SKUs with 12+ months of history we typically see 15-25% MAPE at category level, 25-40% at SKU level. The forecasting screen shows confidence intervals.
What if the AI is wrong?
Every recommendation cites the data rows that fed it. The operator can drill in, see what the AI saw, and override. We track override rates per recommendation type so the AI improves over time.
Can we turn AI off?
Yes. Per-tenant toggles disable forecasting, the chat assistant, or AI-suggested POs independently. Some customers run AI-suggested only after parallel-running it manually for a quarter.
AI pairs naturally with…
See the AI on your data — not a generic demo.
Drop us a sample of your demand history. We'll load it into a sandbox, generate a forecast, and run a week of AI-suggested POs against it. Then walk you through what the AI saw.
Book an AI demo