Most online sellers end up with several numbers for the same SKU. Shopify says 14, Amazon says 9, the warehouse spreadsheet says 11, and the shelf holds 6. Each system updated its own counter at a different moment, and none of them can say why it disagrees with the others.
The cost shows up as oversold orders, cancelled marketplace sales, stock sitting unlisted because nobody trusts the figure, and a month-end inventory value your accountant has to rebuild by hand.
When custom beats an off-the-shelf app
Multichannel inventory apps, the stock modules in ERP and accounting packages, and the inventory built into store platforms cover most sellers well. Use one if it fits.
Custom inventory software earns its cost when:
- You sell through several stores and marketplaces from more than one warehouse, and each has its own rules for what can ship from where.
- Bundles, kits, lots or expiry dates matter, and the apps model them differently.
- You run two or three apps that each think they are the source of truth.
- Your accountant needs stock valued with a cost method the apps can’t show at the order level.
How we would build it
A ledger of movements, not a counter
The core is a stock ledger. Every change is a row: goods received, an order reserved or shipped, a return, a transfer between locations, a count adjustment. Each row records the SKU, the location, the quantity, the time, the reason and the document behind it (purchase order, order number, count sheet).
Stock on hand is the sum of those rows. Available stock is on hand, minus reservations for open orders, minus a safety buffer you set per channel. Because nothing overwrites a total, any figure can be traced back to the movements that make it up.
The data model
- Products and SKUs, with bundles defined as lists of component SKUs and quantities
- Locations: warehouses, and bins inside them if you pick by bin
- Lots or batches, where expiry or traceability matters
- Movements and reservations, as above
- Channel listings, linking each store or marketplace listing to a SKU or bundle
Channel sync
The system would integrate with each channel through its own API: Shopify’s Admin API and webhooks, Amazon’s Selling Partner API, and WooCommerce’s REST API. Incoming orders become reservations. Outgoing updates push each channel’s available quantity.
Every incoming event is stored with its channel ID before it is applied, so a webhook delivered twice is applied once. A scheduled check reads back what each channel reports and compares it with the ledger. Drift goes into a review queue with both numbers. It is never overwritten silently in either direction. For Shopify-specific screens or logic, see our Shopify app development work.
Checks that keep the numbers right
- Whole units in integers. Quantities and money are stored as integers (cents, or the smallest unit you sell), never floating point.
- No negative stock by accident. A shipment that would take a location below zero is refused and queued, unless you allow backorders for that SKU.
- Cost layers that tie. Under first in, first out, each sale consumes the oldest receipts first, and the cost of goods sold for a period equals the cost of the layers consumed.
- Counts as adjustments. A cycle count records the difference as a movement with the counter’s name and reason. It does not reset the total.
- Closed periods stay closed. Once a month’s stock value is reported, changes dated in that month are refused, and corrections are posted in the open period.
Proof from our own code
We have not shipped an inventory system for an online store. The closest engine we run in production is the investment register in Lekhavo, our bookkeeping product for Indian accounting practices, which tracks holdings as quantities and cost layers. It works the way a stock ledger should:
- Purchases are lots with a date, a quantity and a cost. Sales draw on lots first in, first out, or on lots a person chooses where the rules allow it.
- A sale for more units than the holding had on that date is refused, with that reason.
- Quantities are held as integers in ten-thousandths of a unit, and money in paise. Only the cost of a partial sale is rounded, and when a sale takes all the units left in a lot, it takes all the cost left too, so no stray paisa stays behind.
- Each change to the register is written to an audit trail with its actor.
The same design would carry over to SKUs and warehouses. The products would be new.
What the first version includes
- The stock ledger, with locations and bundles
- Sync with your main store and one marketplace
- Reservations, per-channel buffers and the drift review queue
- Purchase order receiving and cycle counts
- An inventory valuation report for your accountant, tied to the ledger
- Source code, a written spec and runbooks your team can maintain
If you also need picking, packing and bin-level control, that is a warehouse management system on the same ledger. If stock value has to reach your books, it connects to the accounting integration.
How the work is run
Tembrant builds with an AI software factory: many agents work in parallel, held to fixed gates. Every change starts from a written spec, tests are written first against samples of your real orders and stock files, each change gets independent reviews, and a person signs off before it ships.
If your channels disagree about what you have in stock, tell us how many SKUs, warehouses and channels you run.
