Inventory software and AI: the honest version

The key facts

• Where AI genuinely helps today: querying records in plain language, drafting communications, surfacing patterns in reports.
• Where scepticism pays: "AI forecasting" claims without published accuracy, and auto-reordering nobody is accountable for.
• BSimple's honest AI story: an MCP endpoint that lets AI assistants read your operations data — suppliers, locations, PAR, reorder, stocktakes, purchase orders, transfers — through the public REST API.
• What we do not claim: no AI demand-forecasting engine, no autonomous reordering.
• The full picture: how the inventory record works is the hub page for everything above.

What AI actually does in inventory software today

Strip the marketing and three uses are real. First, access: asking questions of your own data in plain language — "how many of this SKU across all locations, and when does it need reordering" — which AI assistants answer well when they can read the record. Second, drafting: order confirmation emails, supplier chase-ups, reorder lists — text-shaped work over record-shaped data. Third, pattern-surfacing in reporting: slow movers, seasonal swings, and margin outliers, where a model is a convenience on top of honest data.

The pattern across all three: the value depends on the record being true. AI cannot forecast reliably from quantities that are corrected monthly, and it cannot answer questions the record cannot answer. The unglamorous truth is that a clean, live stock record is the precondition for any AI benefit.

The claims worth scepticism

• "AI-powered forecasting" with no published accuracy and no way to see why a number was produced. Forecasting has value, but treat unexplained claims as unexplained.
• Autonomous reordering — software that buys without a human approving. Guardrails matter more when money moves automatically; ask who is accountable when it buys wrong.
• A chat box bolted onto a dashboard — answering questions the underlying reports already answered, with the same numbers and extra confidence.

None of this means skip AI. It means the software layer (record, guardrails, audit trail) and the AI layer should be evaluated separately — which is also why an open record beats a closed black box. The AI tools landscape surveys the wider category, and inventory using AI narrows it to stock.

BSimple's honest AI story

We build BSimple, so weigh that. We have not built an AI forecasting engine and we do not claim one. What we have done is open the record: BSimple exposes a public REST API with scoped API keys, and an MCP endpoint that lets AI assistants — Claude and other MCP-capable tools — read live operations data directly: suppliers, locations, PAR levels, reorder positions, stocktakes, purchase orders and transfers. In practice that means your AI assistant can answer operational questions from the real record instead of a CSV you exported last week, while writes stay in the product so the audit trail stays intact.

That is the honest boundary and the honest offer: the record, the guardrails and the accounting handoff (Xero and MYOB) live in BSimple from $180/month (AUD); the intelligence lives in the AI tools you choose, connected by design. The automation overview shows the wider integration picture, and the free trial lets you test how an assistant reads your own data.