How-to
Setting up demand and lead time
Every simulation needs three things: a demand estimate, a lead-time estimate, and either a target service level or a pair of cost inputs. This section covers where those estimates can come from and how to set them up. For how the tool turns these inputs into a recommendation, see Core concepts.
Demand
| Source | How it works |
|---|---|
| Fit from data | Pick a Data Hub file and column (via the AI or the Detail tab's "estimate this from your data" flow). The tool fits a distribution synchronously and shows a fit-quality verdict. The fit is saved as a Library entry automatically, so it's reusable elsewhere. |
| From library | Pick an existing Distribution Library entry — useful if you've already fit this product's demand for another purpose. |
| Manual | State a distribution and parameters directly, or describe demand in plain English ("usually around 50 a day, sometimes spikes to 80") — the AI maps this to a normal or triangular distribution and confirms its interpretation before running. |
Demand is expressed per day or week — whichever period your data uses. Lead time is expressed in the same unit.
Lead time
Same three sources as demand. The common case is a single point estimate ("about a week"), which maps to a constant distribution. If you give a range ("5 to 10 days"), it maps to triangular or uniform depending on how you phrase it — the AI states its interpretation back to you before running.
Target service level and costs
Set a target service level, or a pair of cost inputs to unlock cost-minimizing mode. See Reference for the full field list and defaults.
Cost-minimizing mode only activates when both cost fields are present. If you give just one, the AI asks for the other or proceeds in service-level-only mode and explains why.
Order quantity
By default, the simulation uses order-up-to: each reorder replenishes inventory back to the reorder point plus expected demand during one lead time — a reasonable default that doesn't require knowing an economic order quantity. If you supply a fixed order quantity (e.g. "I always order in pallets of 200"), the simulation uses that instead.
Computing an optimal order quantity (EOQ) is out of scope for this tool — see Core concepts and the spec's Out of Scope section. Q is either what you supply or the order-up-to default.
See Reference for what happens when an input is invalid, out of range, or missing.
Using the AI assistant
The AI tab is the primary way to set up and iterate on an Inventory Reorder Simulator project. It's focused by default on a new project, with a prompt stub asking you to describe the product you want to plan reorders for. See Core concepts for how the assistant reads your project context and validates a run before submitting it.
What the AI can do
| Action | Example prompt |
|---|---|
| Set up the initial simulation | "I sell about 50 coffee bags a day, lead time is a week, I want 95% service level" |
| Fit demand from data | "Use my Shopify export, the units_sold column" |
| Adjust an input and re-run | "What if lead time is 10 days instead?" |
| Add cost inputs mid-conversation | "Holding costs me about $0.50/unit/month, stocking out costs me a $5 lost sale" |
| Interpret the curve | "Why does the cost go back up after 300 units?" |
Proactive narration
After every completed run, the AI automatically posts a plain-language interpretation to chat — the recommendation, the achieved vs. target service level, and (if cost inputs were given) how much cushion the cost-minimizing point has relative to the target-service-level point. You don't need to ask for this; it's not gated on a question.
The AI also proactively flags issues: if the achieved service level deviates notably from your target (which can happen due to discreteness in the underlying periods), or if the fitted demand distribution had a poor fit quality, it tells you.
Iterating
Each follow-up that changes an input — a different lead time, a different target service level, newly added cost inputs — enqueues a new run that replaces the project's previous results. There's no need to start a new project to test a "what if."
Be specific. Instead of "make it safer," try "show me the curve at a 99% service level instead of 95%."
Limitations
- The AI edits the current project only; it can't compare across projects in the same conversation.
- Response times depend on an external AI API and may vary; occasional slowness or unavailability is possible.
- Fitting demand from data still requires you to pick a Data Hub file and column — the AI guides you to the right one but doesn't infer it from nothing.
Exporting results
The Download CSV button appears in the Results view once a run has completed — it isn't shown for a run that's still pending, running, or that failed. It downloads the full reorder-point sweep as a CSV, built directly from the results already loaded in your browser — there's no server round-trip. See Reference for the column list.
The file is named inventory-sim-results-{timestamp}.csv and downloads immediately with no additional dialog.
The headline recommendation (reorder point, safety stock, achieved service level) and the AI interpretation aren't part of the CSV — they stay visible in the Results view. A PDF summary report was considered for v1 and deferred.