AI Stock Ordering with Claude and Excel: SOH + 6-Month Movement

Why SOH plus six months of movement is enough to start?

Full demand planning suites exist because enterprises juggle promotions, seasonality, multi-echelon inventory, and supplier calendars. Many independent pharmacies and SMB stockrooms still decide weekly orders from a short list of fast movers, slow bleeders, and “we always get six” habits. For that middle reality, a trailing six-month movement window (by month or by week rolled up) usually captures enough signal to see acceleration, drift, or a flatline — especially when you layer current SOH on top.

What this means on the floor

The goal is not a perfect forecast. It is a consistent first pass: if nothing moved for four months and you still hold twelve units, maybe you do not order twelve more. If movement climbed every month and lead time is noisy, maybe you add buffer. AI’s job is to do the boring aggregation, flag outliers, and phrase trade-offs you can sanity-check in minutes.

Dense packs and mixed strengths are where hand counts lie — unless you are using a camera to count them for you →

What to export into Excel (or Google Sheets) before you prompt?

Keep one row per SKU (or per pack size if that is how you buy). Minimum useful columns: identifier (SKU or internal code), description, quantity on hand, and six numeric columns for outbound movement — e.g. issues to patients, sales, or transfers out — one per month for the last six months. If your system only gives totals, use monthly net movement instead, but be consistent month to month.

How to validate this in your next stock review

Add optional columns if you have them: standard pack size from the supplier, minimum order quantity (MOQ), typical lead time in days, and next expiry date for the oldest lot on hand. Expiry is not required for velocity math, but it stops you from ordering into a shelf that is already time-stressed — ExpiryDesk exists precisely so that date sits next to quantity in daily work, not only in the monthly export.

Name the sheet clearly and freeze the header row. Ambiguous labels (“M1”, “M2”) confuse both humans and models; prefer YYYY-MM or month names everyone recognises.

Spreadsheets age faster than stock — most people track this wrong. Here is the smarter way →

Why A Claude-ready prompt pattern that stays grounded matters for cash and service levels

Paste your table, then ask for structured outputs — not vibes. Example structure: (1) assume the role of a pharmacy inventory analyst; (2) use only the numbers provided; (3) compute average monthly movement and simple trend (up, flat, down); (4) estimate days of cover as SOH divided by average daily movement from the six-month total; (5) propose a suggested order quantity for a two-week or four-week cover target, respecting MOQ if supplied; (6) list assumptions and caveats explicitly.

Why this signal should reach finance the same week

Ask for a table back: SKU, suggested order qty, days of cover at current SOH, trend summary, and risk notes (e.g. “movement dropped last two months — confirm before reordering”). Request that the model flag rows where data is missing or movement is zero so you do not autopilot blind spots.

If your organisation allows file upload instead of paste, the same logic applies: one CSV, one prompt, one review step. Rotate API keys and data policy according to your employer — never paste patient-identifiable data.

Rotation only works when the soonest date is visible before the truck arrives — here is how teams close that gap →

How to read “days cover” and trend without fooling yourself?

Days of cover is a flashlight, not a law. It divides what you hold by how fast stock has been leaving recently. If last month was a spike because of a local flu season, your average will swing; note that in the prompt or smooth with a second line of analysis (“ignore the highest month if labelled as promo”).

When this turns from noise into write-off risk

Trend labels — improving, stable, declining — help prioritise which lines get human attention first. Pair that with expiry: a declining mover with short-dated stock is a different meeting than a rising mover with long-dated stock. ExpiryDesk’s dated line list complements this macro export by keeping per-batch risk visible day to day.

Always reconcile the model’s suggested order against supplier calendars, cold-chain limits, and regulatory minima for scheduled medicines. AI suggests; licensed buyers and responsible pharmacists approve.

If your reminder lives on a sticky note, it does not survive a busy service — this is what an expiry reminder looks like when it scales →

What operators should do about Where this fits with ExpiryDesk — and what to do next week

Use Claude (or similar) to stress-test your Friday order list from SOH and movement. Use ExpiryDesk to ensure the stock you already hold is consumed in the right order and that short-dated lines do not silently duplicate what you are about to buy. Together, velocity and shelf-life answer different questions; both reduce expensive surprises.

What this means on the floor

If the exercise works, document your prompt and column layout once — then every export becomes a five-minute review instead of a forty-minute spreadsheet tour. That is the habit that scales without another enterprise licence — and it still starts with honest numbers in a single table.