A two-site florist that moved tonnes of fresh stock and could not see what it lost

A busy floral operation runs two locations: a flagship shop and a separate arrangement studio that fills online and wholesale orders. Together they move a high volume of perishable stock through roughly 80 to 120 staff. Inventory was counted by hand once a month and typed into a spreadsheet the team had built itself. Supplier deliveries, fifteen to twenty vendors with two large ones, were checked against paper and entered by hand at receiving, with little validation along the way.

The volume was high enough that small losses hid easily. The team could keep the floor stocked without obviously running short, but they had almost no view into what was actually used, wilted, or walked out the door. A one to two percent error sat in the numbers every month, and because the count was monthly, by the time anyone saw a gap the trail was four weeks cold. Transfers between the shop and the studio were noted informally, so even the two sites could not be reconciled against each other.

We started where the loss was hiding: the gap between what sold and what was on hand. Every sale at the point of sale carries a recipe, the stems and materials an arrangement consumes, so we deducted those automatically to keep a theoretical stock count running in real time against the physical one. The difference is shrinkage, and now it surfaces the same day instead of at month-end.

On the receiving side, supplier deliveries are reconciled on arrival and adjust stock directly, so the same numbers are not retyped from invoices into a spreadsheet by hand. Both sites sit on one platform with transfers tracked between them, so the shop and the studio finally reconcile against each other. A single dashboard shows stock state and variance by site at a glance. The extraction behind the receiving step is AI-driven with a deterministic fallback, so a messy delivery note still produces a usable line rather than failing.

Loss stopped being a monthly surprise. Variance is visible the day it appears, by site and by category, so a spike in one product or one location gets caught while the cause is still recoverable. The one to two percent error that used to live unexamined in the monthly count is now something the team can actually act on.

The manual inventory entry that ate about fifteen hours every month is largely gone, folded into receiving and point-of-sale instead of a separate hand count. The same team runs both sites without adding an inventory clerk, and for the first time the shop and the studio read off one set of numbers.