When your CRM and accounting system disagree, start with one real customer journey and decide which system owns each fact. Trace a sale through customer creation, invoice, payment and refund. Record where information is copied, delayed or changed. Then test the awkward cases before switching on a connector. A shared screen is useful; a reliable handoff is what keeps the numbers trustworthy.
This is a timely problem. A Sage survey published on 23 September 2026 asked 1,500 UK SME owners and senior decision-makers about disconnected tools. Respondents reported spending an average of 2.9 hours a week moving and re-entering information. Sage used that answer to model 19 working days a year. It is a vendor survey and a model, not a measured saving your business will automatically receive.
A UK question posted to the HubSpot community this week puts the practical difficulty plainly: how do you keep invoices and payments aligned when the sale starts in a CRM and the money is recorded elsewhere? The post mentions partial payments and refunds, exactly the cases that a simple “create invoice when deal closes” automation can miss. One question is not evidence of search volume, but it is a useful description of the job.
Which system should own the customer and the invoice?
Give each field a named owner. A CRM is usually the working home for sales contacts, deal stages and conversations. The accounting system should be authoritative for issued invoices, credit notes, payment allocation and the balance owed. Your arrangement may differ, but it needs to be written down.
Take the customer name. Sales may first enter a trading name; finance may need a legal billing name. If both systems can edit the same field and changes travel in both directions, which value wins? “Latest update wins” sounds tidy until somebody corrects the legal name in accounts and an old CRM record overwrites it five minutes later.
Make a short handoff sheet with five columns: business fact, owner, receiving system, trigger and exception owner. A row might say: “issued invoice number; accounts owns it; CRM displays it; send after invoice issue; finance resolves missing numbers.” Write the rule in language the sales and finance teams can check without reading code.
Some information should not cross the boundary. A finance system does not need every CRM note. A sales colleague might need to see whether an invoice is paid, but not to edit the accounting ledger. Limiting the data in motion makes the integration easier to explain, secure and maintain.
Where does the disagreement begin?
Follow one journey from accepted quote to money received. Sit with the people doing the work and record the actual steps, including spreadsheets, email and manual checks. Do not start with a vendor’s diagram of how its connector is supposed to work.
Look for four common breaks:
- The customer is matched by name or email alone. Names change, people share inboxes and one organisation can have several contacts. Keep the identifiers from both systems and an explicit mapping between them. Decide who can merge duplicates.
- The invoice is created twice. A user raises it manually while an automation is retrying. A safe handoff checks whether the source event has already produced an invoice before creating another.
- “Paid” means different things. A CRM deal marked won is not an invoice paid. A payment may cover part of an invoice, several invoices, or arrive with fees deducted. Return the accounting status with its date and source rather than guessing from the sales stage.
- Failures disappear. An expired authorisation or changed field can stop a connector. If nobody sees a failed queue, the systems drift quietly until month end.
The Department for Business and Trade’s SME technology-adoption research describes the value businesses place on reliable, personalised support when adopting software. That applies after purchase too. Someone must own the handoff when a normal Tuesday transaction does not fit the happy path.
What should a CRM-to-accounts handoff specify?
I would ask for a one-page contract for the data flow before choosing a connector or commissioning code. It should answer these questions:
- What starts the handoff? Is it a deal marked won, an approved order, or a finance review? A sales stage alone may be too early if the billing entity, tax treatment or price is still uncertain.
- What is created? A draft invoice gives finance a chance to check it. An issued invoice has a different consequence. State which action is permitted and who approves it.
- How are records matched? Store the source and destination IDs. Define what happens when an existing customer looks similar but has a different legal name or address.
- Which way does each field move? Customer contact details might move from CRM to accounts, while invoice number and payment status move back. Avoid two-way editing unless there is a clear conflict rule.
- What counts as complete? “Sent to connector” is not the same as “accepted by accounting”. Require a destination ID, a timestamp and a visible outcome.
- Who sees a failure? Decide where failed records appear, who receives an alert, when they must act, and how to replay a record without duplicating it.
Keep the document with the system owner, not just the developer. If the finance lead cannot explain the rule for a corrected invoice, the specification is unfinished.
Which cases should you test before going live?
Test with synthetic records in a safe environment where the products support one. If a supplier only offers a live test, agree the controls with your finance team first. Do not create a pretend paid invoice in production and hope somebody remembers to remove it.
Use a small set of cases that expose the real business rules:
- A new customer buys once and pays the invoice in full.
- An existing customer buys again, using a different contact email.
- The billing address changes after the first invoice.
- A deal is corrected after an invoice draft exists.
- A payment covers only part of an invoice.
- A credit note or refund changes the outstanding balance.
- The receiving system rejects one record, then the handoff is retried.
For each case, write down the expected customer ID, invoice status, payment status and person responsible if it fails. Check both applications and the audit trail. The last case is especially revealing: a retry should finish the original handoff rather than create a second customer or invoice.
You do not need to automate every rare exception. A visible queue and a clear manual decision can be safer than clever logic nobody can inspect. What matters is that the exception is found, owned and closed.
Should you use a native connector, a platform or custom code?
Choose the smallest approach that passes the tests above and can be operated by your team. A built-in connector is a good starting point when it supports your actual invoice and payment flow. Read its field list and failure behaviour, not just its integration logo.
An automation platform can be useful for a straightforward handoff with modest volume. Include subscription charges, usage limits, administrator time, monitoring and support in the running cost. Ask whether a change to either product’s API will be noticed before transactions pile up.
Custom code earns its place when the workflow has specific matching, approval or exception rules that standard connectors cannot express reliably. It brings build, hosting, security review and maintenance work. Ownership of source code helps, but it does not remove the need to operate the service. Our system integration service starts by mapping the business journey and its exceptions before proposing a build.
Replacing either main application is a larger decision. If the CRM and accounting products otherwise work well, a controlled handoff may cost less and disrupt fewer people than replacing both. If the processes themselves have changed, an integration can also preserve a bad workflow. Check that first. A five-year comparison should use your own purchase, build and running costs, not somebody else’s headline saving.
How do you know the fix is working?
Count mismatches before making a change. For two weeks, record how many customer records need manual matching, invoices need correction, payment statuses lag, and failures need a person to intervene. Record the minutes spent on those tasks. A small team can do this in a shared sheet if it has an owner and a review date.
After the change, use the same definitions. Compare the number of clean journeys, exceptions and time spent fixing them. A lower licence bill is useful, but it does not help if finance spends longer reconciling payments. Nor is a fast demo proof that the integration will survive a partial payment or an expired token.
Set a regular reconciliation: sample recent completed transactions, check the customer and invoice identifiers in both systems, and inspect unresolved failures. Keep a fallback for raising an invoice when the handoff is unavailable. Write down who can switch to that fallback and how records will be caught up afterwards.
At renewal, keep the one-page handoff sheet and the test cases. They make it easier to judge a proposed upgrade, switch connector or change supplier. They also make the cost of maintaining the integration visible, which is part of the real cost of owning software.
Recommended reading
- System Integration for UK SMEs covers the wider choice between APIs, automation platforms and custom middleware.
- Custom Software vs SaaS: The True 5-Year Cost Comparison helps frame the cost of keeping or replacing a tool.
If your CRM and accounts disagree today, bring one real handoff and its exceptions to a conversation about system integration. I can help map the failure, test the options and decide whether a connector, a small build or a process change is justified.