At a glance
- The workflow matches each store's perpetual inventory to receipts, transfers and usage, flags variances above a tolerance you set, and packages the evidence for review.
- The pharmacy operations manager investigates every variance and approves every adjustment. The agent never posts one.
- Controlled-substance inventory is a separate, more carefully governed workflow, and this one excludes it by default.
- The agent works alongside the inventory software you already run; it does not replace it.
Can AI reconcile pharmacy inventory across multiple locations?
Yes, for the preparation and the first-pass matching, with people deciding. A perpetual inventory is the running on-hand count that a pharmacy system updates as stock is received and used. Inventory reconciliation compares that count with physical counts, purchase records and transfers. An inventory variance is a gap between them that the records do not yet explain.
One store-system vendor already treats this as routine work. McKesson Canada's Propel Rx checklist tells pharmacies to spot check the on-hand count against what is on the shelf and, when discrepancies are found, to adjust the on-hand counts [1]. Across many stores, that work multiplies: each store needs its own records pulled, lined up and explained.
- The agent: collects each store's records, maps items and units, computes the expected count, flags variances and assembles the evidence.
- Store staff: count the physical stock and recount when asked.
- The pharmacy operations manager: decides the cause and whether an adjustment is justified.
The inventory reconciliation workflow, step by step
The workflow below covers non-controlled items only. Run it per store at each count cycle, then roll the results up across the chain. You can copy it into a runbook and adjust the cutoffs and tolerances to your own systems.
- Set the scope and the cutoff
List the stores and the non-controlled items in scope. Fix one cutoff time per store, so counts, receipts and usage all cover the same period. Controlled-substance items are not on this list.
- Collect the records
For each store, pull the perpetual on-hand report, the physical count sheet, wholesaler receipts, purchase orders, inter-store transfers and the adjustments already posted. The agent lists any store with a missing file.
- Standardize the items
Map every record to one item key and one unit of measure, so a case, a pack and a single unit are never compared as if they were the same. Flag any item with no clean mapping.
- Compute the expected count
For each store and item: expected on hand = opening count + received + transferred in - used - transferred out +/- adjustments already posted. Compare it with the system's on-hand count and the physical count.
- Flag the variances
Flag any item where the system count, the expected count or the physical count differ by more than the tolerance the operations manager sets. Rank by units and by item value, and suggest a likely cause from the patterns in the table below.
- Package the evidence
Attach the records behind each flag: receipt lines, transfer lines, earlier adjustments, count timing and the suggested cause. The agent drafts the explanation; it does not conclude it.
- Investigate and decide
The pharmacy operations manager investigates each variance, asks store staff to recount where needed, and decides the cause and whether an adjustment is justified.
- Adjust and record
Authorized staff enter each approved adjustment through your existing inventory system, with the reason and the approver recorded. Log the cause of every variance so the tolerances can be tuned. The agent never posts an adjustment.
| Variance pattern | How it is flagged | Evidence attached | First check by |
|---|---|---|---|
| Receipt not entered | Wholesaler invoice quantity is higher than the receipts entered for the item | Invoice line and receiving record | Store receiving contact |
| Transfer recorded at one store only | A transfer-out has no matching transfer-in at the other store | Both stores' transfer records | Store contact at the receiving store |
| Unit or pack-size mismatch | The variance is a multiple of a pack size, or the item maps to more than one pack | Item master record and pack sizes | Inventory data owner |
| Count timing gap | The count was taken before or after the cutoff, with usage in between | Count timestamp and usage during the gap | Store staff who counted |
| Expired or returned stock not removed | The variance sits on an item with return or expiry activity | Return record and expiry list | Store inventory contact |
| Unexplained shortage after recount | The variance persists after the recount and every explanation above | Recount sheet and all earlier evidence | Pharmacy operations manager |
Does this include controlled substances?
No. Controlled-substance inventory reconciliation is a separate, more carefully governed workflow, and this one excludes it by default. The agent here touches recordkeeping accuracy for non-controlled stock only. It makes no access or dispensing decision of any kind, and every controlled-substance access and dispensing decision stays with pharmacy staff.
One vendor's documentation separates the two as well. The Propel Rx checklist says to "Conduct a regular cycle count of your narcotics, controlled drugs, and target substances (NCTs)." It also lists "Review NCT Inventory" apart from "Review Regular Inventory" [1]. We recommend the same separation in your own design.
- Tag each item's class in the item master, and exclude controlled items from the extract before the agent sees it.
- Route any item that looks misclassified to your compliance team, not to the agent's queue.
- Treat any later wish to bring controlled-substance records into an automated comparison as a decision for your compliance team and pharmacy leadership, made before any build.
Who investigates an inventory variance?
The pharmacy operations manager, always. That is an operating choice we recommend, not a requirement. The agent assembles the evidence and suggests a cause, store staff recount when asked, and the manager decides the cause and whether an adjustment is justified.
- Agent: assembles the records, flags the variance and drafts a suggested cause.
- Store staff: recount the physical stock when the manager asks.
- Pharmacy operations manager: investigates, decides the cause and approves or rejects the adjustment.
- Authorized system users: enter the approved adjustment, with the reason recorded.
The Propel Rx checklist describes the same investigate-then-adjust order: it tells users to investigate on-hand discrepancies by reviewing the Shipment window in the Drug Folder Inventory tab [1]. Your own system will have its own screens; the point is that someone with authority looks at the evidence before a count changes.
Set the tolerance yourself, and size the review load with your own numbers: if each variance takes N minutes to investigate and a store raises V variances a cycle, the load is N x V minutes per store per cycle. For the review-and-approve pattern behind this step, see document review with human approval.
Does this replace existing inventory software like LEAFIO?
No. The design is an agent layer around the inventory system you already run. Your system keeps the stock records and the audit trail, and the agent prepares what goes in and packages what falls out. Where your system already does a step well, keep using it.
- LEAFIO's pharmacy software page says its AI-powered software "reduces errors by standardizing processes and flagging inventory exceptions early" [2].
- The same page describes ERP integration as "connecting straight into the ERP your chain already operates" [2].
- The Propel Rx checklist tells users to "Review and ACTION the Auto-Reconcile Exceptions report" for the purchase orders it receives [1].
Those are the vendors' own descriptions, and we make no claim about either product beyond them. The LEAFIO page we read (October 2026) covers forecasting, replenishment and merchandising. It doesn't describe reconciling physical counts against receipts, so confirm scope with the vendor. We don't claim an integration with any named product.
How does this fit with multi-location AP?
Inventory reconciliation and accounts payable read the same purchase events from two sides. The invoice says what you owe, the receipt says what arrived, and the count says what is on hand. A receipt that was never entered shows up as an inventory variance and may also surface as an invoice mismatch.
One store system ties these together too: the Propel Rx checklist tells users to send and reconcile purchase orders electronically through the system [1]. The same records feed both workflows, so fix a receiving error once and tell AP.
- Shared records: wholesaler invoices, receiving records, purchase orders and store coding.
- Shared owners: the operations manager for the count side, the AP lead for the invoice side.
- Shared method: matching under written tolerances, with exceptions reviewed by a named person.
The matching and exception mechanics are the same ones covered in financial reconciliation automation. Keep the AP workflow itself separate, and let the two exception queues share evidence.
When this doesn't apply
- You run a few stores and your system's own reports already settle each count. A spreadsheet or the built-in report may be enough.
- Nobody takes regular physical counts. There is nothing to reconcile against until counts exist, so start there.
- Item data is messy. LEAFIO's page makes the point for replenishment: "Start with clean master data (SKUs, pack sizes, suppliers), clear reorder rules, and store/warehouse parameters." [2] The same applies to matching.
- Your exception queue is already short and well explained. An agent layer adds little there.
- The records are controlled substances. That is a separate workflow, as covered above.
Where to start: a first scoped workflow
Start with a small group of stores and a limited set of non-controlled items, not the whole chain.
- Pick a group of stores that already take regular counts, and one product category with clean item data.
- Run the workflow in parallel with your manual process on a closed count cycle, and compare the variances each finds.
- Log the cause of every variance, so the tolerances and the owner for each pattern come from your own data.
- Name the operations manager, the recounting staff and the approvers before the first live cycle.
When the parallel run is clean, add stores and categories. For how a first workflow is built and put into use, see AI agent implementation. For general match rules and tolerances, see reconciliation match rules and exceptions.
Frequently asked questions
Can AI reconcile pharmacy inventory across multiple locations?
Yes, for preparation and first-pass matching. An agent collects each store's records, maps items and units, computes expected counts, flags variances and packages the evidence. People still count the stock, investigate each variance and approve each adjustment, so the agent supports the pharmacy operations manager rather than replacing that review.
Does this include controlled substances?
No. Controlled-substance inventory is a separate, more carefully governed workflow, and this one excludes it by default. The agent addresses recordkeeping accuracy for non-controlled stock only, and never makes an access or dispensing decision. Confirm the requirements that apply with your compliance team or counsel.
Who investigates an inventory variance?
The pharmacy operations manager, always. That is an operating choice we recommend, not a requirement. The agent assembles evidence and suggests a cause, store staff recount when asked, and the manager decides the cause and whether an adjustment is justified. Authorized staff then enter it in your system.
Does this replace existing inventory software like LEAFIO?
No. The agent works around the inventory system you already run: it prepares records going in and packages the variances that come out. Your system keeps the stock records. We describe products only from their own documentation and don't claim an integration with any named product; confirm scope with each vendor.
How does this fit with multi-location AP?
Both workflows read the same purchase events: invoices, receiving records and purchase orders. A receipt that was never entered can appear as an inventory variance and as an invoice mismatch. Keep the workflows separate, but let their exception queues share evidence so a receiving error is fixed once.
Related reading
- Financial reconciliation automation covers the matching and exception workflow this article applies to inventory.
- Document review with human approval covers the review-and-approve pattern behind the variance step.
- AI agent implementation covers how a first scoped workflow is built and put into use.
- Reconciliation match rules and exceptions covers tolerances and exception handling in general.
- Pharmacy chains covers where AI agents fit across multi-store pharmacy operations.
More in Finance operations
Deal Jacket and Title/Registration Automation for Multi-Rooftop Dealer Groups
Deal jacket and title/registration automation for multi-rooftop dealer groups: a completeness checklist table, audit-gap handling and cross-store standards.
Auto DealershipsFranchise Brand-Standards Audit Documentation: Automating the Paper Trail Behind the Inspection
Franchise brand-standards audit documentation: how an agent turns photos, PDFs, forms and emailed fixes into one packet per location, ready when asked.
Multi-Unit Franchise OperatorsGroup and Folio Billing Audit: Catching the Leak Types a Normal Error Check Misses
A group and folio billing audit for hotel management companies: a leak-type table of rate code mismatches, missing closes, adjustments, orphaned charges.
Hotel Management Companies