Why this matters
A point of sale system sees more transactions in a week than your finance team posts in a month. Each sale carries tax, a tender type, a cost of goods and a stock movement. Retype any of it and the errors pile up quietly until month end turns into an argument about which number is right.
The two systems also describe the same event in different words. The POS records a ticket, a shift and a tender. The ERP wants a journal entry, a cash receipt and an inventory issue. Connecting them is less about moving data than about agreeing on which side owns each field and what one record means.
Done well, the link produces a daily sales summary per location, tenders mapped to the right accounts, stock reduced at item level, and a short exception queue that someone actually reads. Done badly, it produces a second set of books nobody trusts, plus a store manager who still keeps a paper tally in the drawer.
Before you start
Pick an owner for the item master. Most companies keep items, cost and price in the ERP and let the POS read them. If staff can create items or change prices at the till, you get two versions of the truth within days. Choose one direction and switch off editing on the other side.
Build the mapping tables before you touch code. You need a line for every tender type, including cash, card, gift card, store credit, mobile wallet and purchase order. Each maps to a GL account or a clearing account. Do the same for tax codes, store to branch, and SKU to item record. Every missing line becomes a rejected record later.
Get credentials and a sandbox. You want API keys scoped to reading sales, a test company in the ERP, and a test location that is not your busiest store. Ask whether the POS offers webhooks or only a polling endpoint, and how far back the API returns history.
Agree on a cutoff: a date and time when the new flow becomes the only flow, plus opening till floats for each location. Without a hard cutoff, someone will process the same day twice. Most of the effort in How to connect point of sale to ERP sits in these decisions, not in the code.
How to do it
Start with the daily sales summary rather than the item lines. It is the record finance needs, it is small, and when it breaks you can see why. Once the summary reconciles for two weeks, add item detail for stock and margin, then returns, then additional locations.
Keep the summary idempotent. Key each record on store, business date and shift so a retry updates the entry instead of posting a duplicate. Duplicate journal entries are the most common way these projects lose trust in week one.
Expect the first live week to be about exceptions, not volume. Check the rejection queue twice a day and fix mappings as they surface, because a missing tax code will hit every store at once on the next business day.
- Map fields and owners. Write a table with three columns: POS field, ERP field, owning system. Include tender type, tax code, store id, SKU, quantity, unit price, discount and timestamp. Get sign off from whoever closes the books.
- Set up a sandbox and credentials. Create API credentials in the POS back office with the narrowest scopes that still read closed tickets or shift reports. Add a test company or branch in the ERP. Never point the first build at production.
- Post the daily sales summary first. Pull each closed shift or business day, total it by tender and tax code, and post one entry per store per day. Send card tenders to a clearing account so the bank deposit clears against it separately.
- Add item level lines. Send sold items with quantity, unit price, discount and cost, and let the ERP issue inventory. Decide whether stock moves from the sale feed or from a physical count, and do not run both for the same store.
- Route returns and voids the same way. Returns post as negative lines against the original business date when that period is open, or as a separate adjustment when it is closed. Voids never happened, so exclude them from the summary.
- Make every write idempotent. Key line level records on the POS transaction id and summaries on store, date and shift. Store those keys in the ERP or a small mapping table so retries update rather than duplicate.
- Add retries and a rejection queue. Retry network failures with backoff. Send validation failures to a queue with the raw payload and the error text, and give one person the job of clearing it each morning.
- Go live one store at a time. Run the new flow beside the manual process at a single store for two weeks and compare daily. Cut over, repeat at the next location, and keep the old path available for a month.
How to check it worked
Reconcile in three directions every day for the first month. Sales per tender in the POS against the ERP cash receipt and clearing account. Tax total against the tax report. Item quantity sold against the inventory movement report. Small timing gaps are normal. A gap that survives to the next day is not.
Then follow the money. Take one card batch and trace it from POS settlement to the clearing account to the bank line. If that chain takes more than a minute to explain, the tender mapping needs another pass.
Read the exception queue for a week and look for patterns: one store always failing, one tender always rejected, one tax code missing. Count the failures, not the successes.
Last, ask the store manager what they stopped doing. If they still type totals into a spreadsheet at close, the job is not finished.
Mistakes to avoid
Sending only a grand total per day. It posts quickly and looks tidy, but you lose item level cost, so margin reports drift and stock counts never match. Keep the summary for finance and the lines for operations.
Letting both systems edit prices or stock. Two writers means conflicts, and conflicts usually get settled by whichever system saved last. Make one side read only.
Treating a failed sync as a network blip. Most rejections are mapping errors: an unknown tax code, a closed accounting period, a store code that does not exist in the ERP. Log the full payload and the error text or you will be guessing.
If the build is more than your team wants to take on, the Cloudverg Systems integration build runs $12,500/month, fixed scope over three to four weeks, and covers the integration map, retries with dead letter handling, a field mapping document and alerting when a sync fails. It is one option among several. A middleware tool, your POS vendor's native connector or a small internal service can all do the job.
Frequently asked questions
Should the point of sale or the ERP own inventory quantities?
Pick one system as the record for stock and let the other mirror it. Most companies keep stock in the ERP and have the POS read availability. If tills must keep selling when the network drops, the ERP reconciles from the sale feed at sync time rather than blocking sales.
Can I export a CSV from the POS each night instead of using an API?
Yes, and it holds up for a single store with steady volume. The trade-off is manual handling, no automatic retry and no protection against a half written file. Move to an API or webhook feed once you have several locations or want the reconciliation to run unattended.
How should sales tax be handled in the sync?
Map each POS tax code to a tax jurisdiction or tax group in the ERP and post tax as its own line. Do not let the ERP recalculate tax on imported sales, or the totals will differ from what the customer paid at the counter.