QuickBooks knows exactly what you sold last March. It has no idea what you could have sold.
That gap is the whole problem with forecasting inventory from an accounting system, and it is the reason a good answer from Claude on your QuickBooks exports can still be wrong in a way nobody notices until the next stockout.
This is a method for doing it anyway, honestly: which reports carry the signal, what they cannot tell you, the handful of formulas that turn history into a reorder point, and how that reorder point lands in cash. It also says when you should stop and buy a dedicated tool instead.
What QuickBooks already does, and what it does not
QuickBooks Online tracks inventory quantities on the Plus and Advanced plans. You can set a reorder point per product, and QuickBooks uses it to tell you what is running low or out of stock, then lets you batch those items into a purchase order. The reorder point is a number you type in. Deciding what it should be is left to you.
On the desktop side, QuickBooks Desktop Enterprise lists "automated inventory forecasting" in its Advanced Inventory module, on the Platinum and Diamond tiers. On Intuit's own QuickBooks Online help pages, I could not find a demand-forecasting feature at the time of writing. The January 2026 inventory release added item receipts, moving average cost and prefilled count adjustments, all useful, none of them a forecast.
So the job is: compute the reorder point QuickBooks asks you for, per item, from data QuickBooks already holds. That is a good job for Claude, as long as you know what the data is missing.
The four exports that carry the signal
Export these over the longest window you have. Twenty-four months is the practical minimum if seasonality matters, because one year gives you one observation per month and no way to tell a pattern from an accident.
- Sales by Product/Service Detail. Every sale transaction by item, with date and quantity. This is your demand history, and it is the one that matters most.
- Inventory Valuation Detail. Every transaction per inventory item and how it affects quantity on hand, value and cost. The running quantity is what lets you find the days an item sat at zero.
- Purchases by Product/Service Detail, or Purchases by Vendor Detail. What you bought, from whom, when.
- Open Purchase Order Detail. What is still on order and what has been received so far. Without it, a forecast will happily reorder stock that is already on a boat.
Add the Inventory Valuation Summary as of today for the starting position: quantity on hand, value and average cost per item.
If you have read how to analyze QuickBooks data with AI, the logic is the same: detail over summary, monthly over annual, and a question with a right answer.
Three things the exports cannot tell you
Lost sales are invisible. QuickBooks records a sale when there is an invoice or a receipt. A customer who wanted the item in a week it was out of stock and bought elsewhere left no trace. The month shows low sales, and a naive average reads it as low demand, which lowers the reorder point, which makes the next stockout more likely. The error feeds itself.
The fix is mechanical. From the Inventory Valuation Detail, find every period where the running quantity hit zero, and treat that period's sales as censored: exclude it from the averages, or flag it and decide by hand. Ask Claude to do this first.
Lead times are not recorded. The product record in QuickBooks Online has a preferred vendor, a purchase cost and a reorder point. It has no lead-time field. If the business raises purchase orders and receives against them, the gap between the order date and the receipt date is an observed lead time, and its spread matters as much as its average. If purchases are entered straight as bills, the lead time lives in the buyer's head. Ask them, write the answer down, and treat it as an assumption rather than a fact.
Service level is a decision, not data. How often you are willing to run out is a commercial choice. Nothing in the ledger holds it, and it moves the cash more than anything else in the calculation.
The method, one item at a time
Four steps. Every formula is standard and none of them needs anything fancier than a spreadsheet.
1. Clean monthly demand per item. Sum quantity sold per item per month from the sales detail, then drop or flag the stockout months found above. Merge items that are the same product under two names, which in a QuickBooks file that has lived a few years is more common than anyone admits. Claude is quick and good at spotting those pairs; the merge itself is your call.
2. Seasonality, if you have the history. For each calendar month, divide that month's average demand by the average across all months:
Seasonal index (month m) = average demand in month m / average monthly demand
Forecast demand (month m) = baseline monthly demand × seasonal index (m)
The twelve indices average to 1. For slow movers, compute the indices at category level and apply them to each item, because an item that sells 6 units a month does not have a seasonality of its own, it has noise.
3. Safety stock. The buffer that covers demand being higher than average while you wait for the delivery:
Safety stock = z × σ(monthly demand) × √(lead time in months)
z comes from the service level you picked: 1.65 for a 95% chance of not running out during a replenishment cycle, 2.33 for 99%. This version assumes the lead time is stable. If it is not, the formula has a second term for lead-time variability, and that is usually the moment to stop doing this by hand.
4. Reorder point.
Reorder point = expected demand during the lead time + safety stock
A worked example. An item sells 400 units a month on average, with a standard deviation of 90. Lead time is 1.5 months, about 45 days.
- Demand during the lead time: 400 × 1.5 = 600 units
- Safety stock at 95%: 1.65 × 90 × √1.5 = 1.65 × 90 × 1.225 ≈ 182 units
- Reorder point: 600 + 182 = 782 units
That 782 is the number you type into the reorder point field in QuickBooks. For a seasonal item, use the forecast demand for the months the lead time covers, not the annual average.
A prompt that gets this done across a few hundred items in one pass:
From these exports, build monthly demand per item for the last 24 months. Flag every month where Inventory Valuation Detail shows the item at zero on hand and exclude it. Compute mean and standard deviation per item, then safety stock and reorder point with the formulas below, at a 95% service level and the lead times in the attached table. Show me the 20 items where the new reorder point is furthest from the one in QuickBooks.
The last line is the one that earns its keep. It turns a table of numbers into a conversation with whoever does the buying.
Where the forecast meets cash
The reorder point decides when you buy. The order quantity decides how much. Together they decide how much cash sits in the warehouse:
Average inventory ≈ safety stock + order quantity / 2
Keep the example item, ordered one month of demand at a time (400 units) at a unit cost of 25. Average inventory is (182 + 200) × 25 = 9,550.
Now raise the service level to 99%. Safety stock becomes 2.33 × 90 × 1.225 ≈ 257 units, 75 more than before. That is 75 × 25 = 1,875 of extra cash tied up, for one item. Across 200 items with a similar profile, it is 375,000. The P&L does not move by a cent.
That is the point to take to the business owner: service level is a cash decision dressed up as an operations one.
For the balance-sheet side, the question becomes how this per-item view rolls up into the inventory line of a three-statement model. That is a different choice (percent of COGS, days, unit build, or flows), and it is covered with numbers in four ways to model inventory in a financial model. The per-item method here is the operational input; the model's inventory line is where it turns into days and cash.
Timing matters too. Cash leaves when the supplier bill is paid, not when the stock arrives, so the purchase plan has to go through your supplier terms before it reaches the cash forecast.
When to buy a dedicated tool instead
The method above is right for a business with tens to a few hundred active items, one or two locations, and a buyer who places orders weekly or monthly. It is a spreadsheet-sized problem.
It stops being one when you have thousands of SKUs, several warehouses, supplier minimum order quantities, bills of materials, or purchase orders that should be drafted automatically every morning. At that point you want software whose whole job is replenishment, and the QuickBooks ecosystem has several: StockTrim, GMDH Streamline and Inventory Optimizer all describe QuickBooks integrations, with demand forecasting, lead times, service levels and purchase order generation built in. On desktop, Advanced Inventory in QuickBooks Desktop Enterprise is the native option.
Those tools answer "what do I order this morning". They are not built to answer "what does this buying policy do to cash in November", and that is the question the finance side usually owns.
Keep the forecast out of the chat
Here is the failure that shows up in month two. Claude computed 782 in a chat. You typed it into QuickBooks. Next month demand shifts, and the 782 is still there, because it was a value, not a formula. It is the same trap as AI hardcoding in a financial model: correct when written, silently stale after.
The fix is the one from using AI with QuickBooks for forecasting: the logic lives in a structure that persists, and the exports refresh the data underneath it. Demand, safety stock, reorder point and cash should be formulas that recompute when a new month of sales lands, with the service level and lead times as named assumptions you can change.
Layerz is one way to hold that: a structured spreadsheet, ready for AI. You import the QuickBooks exports as data sources, and next month's export refreshes the same source in place. The forecast lives as a model with live formulas that Claude drives over MCP, instead of a chat answer, and it exports to Excel with those formulas intact. There is no live QuickBooks connection today, so the export step stays. If you would rather do this in Excel, the rule is the same: formulas, not pasted values, and the assumptions in cells someone can read.
The takeaway
QuickBooks holds enough to forecast inventory for most small businesses: sales by item, the running quantity on hand, purchases, open orders. It does not hold lost sales, lead times or the service level you want, and the forecast is only as good as how you handle those three.
Clean the stockout months, write the lead times down, pick the service level on purpose, and let Claude do the arithmetic across every item. Then read the cash line, because that is where the decision really lands.