An enquiry lands in the inbox. Someone reads it and types the details into the CRM. When the job is agreed, the same details are typed again into the invoicing software. If there is a job sheet, that is a third time. Nobody planned this. It grew, one tool at a time.
Why it happens
Nearly always because the tools were bought at different times to solve different problems, by people who were right to buy them. The accounting package was chosen by whoever does the books. The CRM was chosen by whoever does sales. Neither decision was wrong, and nobody was asked to make them talk to each other.
So the integration became a person. That is the actual state of most small businesses, and it works, right up until that person is on holiday.
What it really costs
The typing time is the smallest part. The expensive parts are:
- Errors. Every retype is a chance to transpose a number or misspell an email, and those are found later, by a customer.
- Disagreement. When two systems hold the same fact, they eventually disagree, and then nobody trusts either.
- Delay. Work waits for the person who does the copying.
- Concentration. Interrupting skilled work to do data entry costs more than the entry itself.
If you want a number, count how often somebody has to check two systems to answer one question. That is the tax.
The realistic fixes, cheapest first
Use a native integration
Check first. Plenty of mainstream tools already connect to each other, and if yours do, the answer is an afternoon of configuration and no build at all. Anyone who does not check this before quoting you is not being straight.
Connect them with a workflow tool
Tools like Zapier or Make handle simple, well-behaved flows well. They are a genuinely good fit when the data is clean and the logic is a straight line. They get expensive and fragile when volume grows or the logic gets conditional.
Build a direct integration
Worth it when the volume is real, the logic has branches, or one of the systems has no API and needs the data pulling out of documents. More work up front, cheaper to run, and it does what your business actually does rather than what a template allows.
These are in order deliberately. If a native integration solves it, take the native integration. We would rather tell you that than sell you a build.
Where AI is and is not the answer
If both systems have an API and the data is structured, this is plumbing, not AI, and should be priced as such. Anyone reaching for a language model to move a field from one database to another is adding cost and unpredictability to a solved problem.
AI earns its place when the input is unstructured: an email written in prose, a PDF invoice with a different layout per supplier, a photographed delivery note. Reading messy input and producing clean structured data is exactly what it is good at, and it is where most of our document work sits.
The important part is that it must never invent a value it could not find. A field it cannot read should come back empty and flagged, not filled with a plausible guess, because a wrong value that looks right is worse than an obvious gap.
Where to start
Pick the single pair of systems that causes the most retyping, and fix that one. Not the whole stack. One connection, working reliably, tells you more about whether this is worth continuing than any amount of planning.