Five systems, one version of the truth.
Businesses grow one system at a time, and people become the integration layer between them. Re-keying, mismatched records and 'which number is right?' are the result.
People as the integration
- A new customer typed into the CRM, then finance, then the support desk
- The same company spelt three ways across three systems
- Sales and finance arriving at a meeting with different revenue figures
- Integration tools that break the moment a record doesn't match exactly
An agent as the integration
- A record created once and carried into every system that needs it
- Records matched even when names, addresses and codes differ
- One agreed figure, with the trail back to its source
- Exceptions queued for a person instead of silently failing
How systems are kept in step
- 1 Map We agree which system owns each piece of data, and who is allowed to change it.
- 2 Connect Agents link your systems through their APIs, or through the screen where there is no API.
- 3 Match AI resolves the messy middle: duplicates, different spellings, missing codes.
- 4 Guard Changes above a threshold, and anything that touches money, wait for approval.
- 5 Watch Every sync is logged, and drift between systems is reported before it matters.
Watch systems sync.
An illustrative run of an integration agent at work. The customers are invented; the way it works is not.
The record, after.
One customer as five systems saw it before, and what the agent left behind. The numbered markers show what it did.
- Matched across systemsRecords are joined even when the names, codes and addresses differ, with the evidence shown.
- Changed once, carried everywhereA change made in one system reaches every other system that needs it.
- Asked when unsureWhat the agent cannot settle from the records goes to a person, not into a guess.
One customer, five systems
How each system held the customer
| System | Before | After |
|---|---|---|
| CRM | Customer 07 Pty Ltd | Matched, owner of the record |
| Finance | CUSTOMER 07 P/L | Matched, address updated |
| Support desk | Customer 07 (Brisbane) | Matched, queue created |
| Billing | Old delivery address | Updated from the portal |
| E-signature | Contract signed 10:02 | Linked to all four |
What the agent did
- Recognised three spellings as one customer, from the ABN, the address and the contact's email.
- Carried the new address from the portal into finance, billing and the CRM.
- Asked the account manager one question: the billing start date was not on the contract.
Illustrative. Names and figures are invented.
Watch one invoice, handled two ways.
A 66-second film, with no sound to turn on: a supplier invoice processed by hand, the same invoice with an agent, and then the rest of a finance portal built the same way.
Read what the film shows
- By hand. A supplier invoice lands in a shared inbox. It is typed into the purchasing spreadsheet, typed again into the accounting system, then waits for an approval by email. It is paid on day nine, late, and nobody notices that the unit price is 6% above the purchase order.
- With an agent. The same invoice is read as it arrives and checked against the supplier file, the purchase order and the goods receipt. The agent flags the price difference and asks the purchasing manager, who chooses to query the supplier with one tap. The agent sends the email, receives the credit note, posts the invoice at the corrected amount and schedules payment for the due date, logging every step.
- The whole portal. The same approach across payables, bank reconciliation, debtors, expenses, month-end and the board pack: one portal on top of the accounting system already in place, with people approving what matters.
Where the AI comes in.
Moving a field from one system to another is plumbing. Deciding what the field means, and what to do when it doesn't match, is where integrations usually fall over.
Matching messy records
'Acme Pty Ltd', 'ACME P/L' and 'Acme (Brisbane)' are recognised as one customer, with the evidence shown before anything is merged.
Reading what systems can't
Emails, PDFs and portal screens become structured data, so the systems that never had an integration can finally be connected.
Handling the exceptions
When something doesn't fit the rules, the agent explains the problem in plain English and asks a person, rather than failing quietly.
Built to run in production, like everything we make
- Inside your systems Agents work in the CRM, finance and operations tools you already run, with the same permissions as a member of staff.
- A person approves what matters Payments, customer messages and anything above a limit you set wait for a human yes.
- Every action logged Each step is recorded, so you can answer what happened, when and why.
- Measured and handed over Results are measured against today's baseline, and your team is trained to run and change it.
Questions people ask us
How do you connect business systems that don't talk to each other?
Spurbuck Advisory agrees which system owns each piece of data, connects systems through their APIs or through the screen where there is no API, and uses AI to match records that don't line up exactly. Changes above a threshold wait for approval, and every sync is logged.
What happens when records in two systems don't match?
The agent recognises when 'Acme Pty Ltd', 'ACME P/L' and 'Acme (Brisbane)' are the same customer and shows the evidence before merging. If it isn't confident, it explains the problem in plain English and asks a person rather than failing quietly.
Is this the same as an integration platform?
It works alongside one or replaces the need for one. The difference is the AI layer that handles documents, inconsistent data and judgement calls that rule-based integrations can't.