Pilot a site material reconciliation workflow

Select one site store and a manageable material family, such as contractor-owned reinforcement steel. Define the stock unit, included locations, subcontractor custody boundaries and reporting cutoff. Include returns and transfers touching that boundary even if the other site is outside the pilot; otherwise the balance will be incomplete.

The stores head owns operational acceptance. A QS validates the installed-work basis if consumption is in scope. Finance owns reconciliation to the stock ledger and adjustment policy. An integration owner is responsible for failures and recovery. Record these names in the pilot charter before selecting software.

Keep the first scope precise: consolidate accepted movement records and propose differences. Do not use it to infer technical consumption, classify loss automatically or decide recoveries. Those decisions need their own evidence and authority.

The proposed flow is: source records → validation → movement balance → count comparison → owner review. Invalid units, missing acknowledgements and conflicts go to the exception queue before being treated as settled movements. Show the balance as provisional when an unresolved record could change it; excluding a row must never make an incomplete reconciliation look final.

Reconcile each extract as well as the final stock figure. Record its source cutoff, number of movements and receipt/issue totals in each valid unit, then account for accepted, rejected and superseded rows. For a site sheet without permanent record IDs, agree an ID convention at capture and preserve revisions. Row number alone is unsafe if users can sort or insert rows. Check that an issue copied into both the site sheet and ERP represents one movement, not two independent sources of quantity.

Confirm the actual interfaces in the installed systems. An export may be sufficient for a first trial. ERPNext’s stock-entry model is a useful movement reference, but the connector must be tested against the project’s actual version and configuration.

Illustrative test: the opening stock is 35 tonnes. A storekeeper records a 5-tonne issue offline. The device uploads it, but loses the acknowledgement and sends it again. The expected balance is 30 tonnes, not 25.

Assign the issue a stable identity before the first upload. The receiver should recognise a replay of that event and return its existing result. If the same identity arrives with a different quantity, hold it as a conflict rather than treating it as either an ordinary retry or a new issue. The AWS Builders’ Library discussion of safe retries explains the underlying use of request identity; the construction acceptance condition here is the single movement in the balance.

Now add a genuine second 5-tonne issue. It must have a new identity and reduce stock to 25 tonnes. Suppressing every row with the same item, date and quantity would incorrectly remove legitimate work. This pair of tests exposes both duplicate posting and excessive deduplication.

Keep event time and entry time separately. A return physically received before cutoff but approved later needs an explicit treatment agreed with stores and finance. The pilot should show the late evidence and its effect on a revised statement, preserving the previously issued version.

An exception needs a quantity or transaction, a reason, supporting references, an owner and a next action. “Transfer not reconciled” is incomplete. “2 tonnes dispatched from Store A; receiving acknowledgement missing; Store B supervisor to confirm by tomorrow” is actionable.

Use separate reasons for missing data, unit mismatch, duplicate identity, disputed quantity and accounting correction. The integration team can repair a failed export, but it cannot decide whether the subcontractor received the steel. The stores team can confirm custody, but should not change a financial period without finance approval.

Prevent cosmetic closure. An item resolved by a manual correction should retain the original discrepancy and approval. Track reopened cases and repeat causes so the pilot does not merely move reconciliation effort into a different screen.

Measure at least one representative completed reconciliation cycle, including late and corrected records. Record preparation time, review time, unresolved quantity by material, exception age and the number of missing source references. Agree the comparison period and inclusion rules before seeing pilot results.

After synthetic tests pass, run a complete reconciliation cycle alongside the existing process. Include movements received late and the count-adjustment review. Compare preparation and review hours, the age of unresolved differences and the number of records needing manual repair. List the slow cases with their causes; a percentile based on a handful of monthly closes adds little.

A stock variance found in the source records is a business exception for investigation. A balance that differs from those same records because the integration omitted, duplicated or changed a movement is a pilot failure. Stop expansion for that failure, an authority bypass or unauthorised disclosure. Pause the affected connector, retain its checkpoints and return reviewers to the last accepted pack. Repair and replay in a controlled test, then reconcile before resuming.

For a later write-back phase, require approved transaction types, permissions, target references and a demonstrated reversal process. Do not delete accepted ledger history as a rollback technique. Finance should approve correction entries in the authoritative system.

The handover should include field mappings, rejected-record reasons, an operating guide, support ownership, export instructions and the tested recovery procedure. A pilot dependent on its original developer to explain every difference is not ready to expand.

The pilot succeeds when the balance is reproducible and the remaining differences have owners. Reduced preparation effort is valuable only if correction work and unresolved quantities have not simply moved elsewhere.

1. Agree one material boundary and sign off the opening balance.
2. Collect a representative baseline and map each source of authority.
3. Pass the duplicate, transfer, correction and access tests.
4. Run a reviewed live cycle and reconcile every output to its inputs.
5. Decide whether to improve capture, extend the read-only scope or separately approve limited write-back.

A material-pilot enquiry can begin with the material group, number of stores, current records and the difference that keeps returning at month-end. That gives the discussion a specific balance to investigate.