When two tools can’t talk, a person becomes the API

---

TL;DR

• When two tools can’t talk, a person becomes the API. Someone does the integration by hand, every week.
• That manual bridge has hidden costs: labor hours, error rates, and data that can drift.
• You can estimate your own waste with one line of arithmetic. Don’t trust the scary number, run the math on your own workflow.
• The fix starts at procurement: ask how a tool integrates before you sign, not after.

---

When two tools can’t talk, a person becomes the API.

When two tools don’t share an API integration, a human fills the gap by exporting and re-importing data.

There’s a beginning of the week ritual in most ops teams that is kind of taken for granted. Someone opens Tool A, exports a CSV, deletes the columns that don’t matter, renames the ones that do, and imports the file into Tool B. Every week. Sometimes every day.

If that’s you, you’re the API.

Tools need to have an agreed way to exchange data (that’s all an API is). And if they don’t, then a person gets to stand in the gap and move the data themselves. We’ve all done it.

A real integration is infrastructure a company pays for, and that cost shows up on an invoice. When a person does the same work by hand, the company pays the same integration-infrastructure price in salary and hours. You are the API, and you’re being expensed as one, quietly.

This post is about dragging that cost onto the invoice.

The three costs of a human bridge

So what are you paying for?

The hours themselves. Exporting, reformatting, importing, spot-checking.

Manual transfers introduce errors and each error costs twice: once to detect, once to fix. And detection is the expensive half, because errors from manual syncs don’t announce themselves, they are noticed when something screws up.

The moment the same business record lives in two unconnected systems and each copy gets updated separately, they start to diverge. There’s no obvious error to catch, because neither copy is “wrong” in isolation. They just disagree. Over months, that can turn into wrong reports, wrong forecasts, wrong headcount decisions.

Calculate your own waste

Price one manual sync — just one — from your own week.

Write down four numbers: minutes per run, runs per month, your loaded hourly rate, and how many bad imports you get in a typical month (plus the hours to fix each). Then:

labor  = minutes per run × runs per month × loaded hourly rate
errors = bad imports per month × hours to fix each × loaded hourly rate

monthly cost of one manual sync = labor + errors

A worked example. Your CRM and your billing tool don’t talk, so someone reconciles them by hand: about 30 minutes a run, twice a week. Call it 8 runs and 4 hours a month. At a loaded rate of $60/hour that’s $240 a month in labor.

Now for the errors. Say one bad import a month slips through: a mispriced line item, a duplicate account. Tracing and fixing it eats two hours across the people involved. Another $120.

So one unglamorous half-hour sync runs about $360 a month, or $4,300 a year.

A rule of thumb for what to do with that. If a manual sync costs more than a few thousand dollars a year and it touches a core system — CRM, billing, support — you’re in the territory where buying an integration or funding the API work is probably going to be cheaper than paying for it by hand.

And now you can name it, instead of carrying it around as vague frustration. Integration tax: the recurring, hidden cost of moving data by hand instead of by API.

Once you can price that tax for a single sync, you can justify buying a real integration or funding the API work. And you can push back on “we’ll just export a CSV for now” before it becomes someone’s job forever.

Why am I mentioning this? I’ve been there too. I have to keep my own working information in line with the rest of the team. If I forget to update some copy it’s not immediately a big deal. Until someone has to scramble before that email goes out to figure out why the link isn’t right. Or why the text wasn’t changed. All because I may not have copy pasted something properly. I try to use integrations whenever I can because I don’t like context switching. And it helps.

A sanity-check ceiling

Vendor surveys put the ceiling in the millions. MuleSoft’s 2026 Connectivity Benchmark surveyed 1,050 IT leaders globally and found the average organization now runs 957 applications, with only about 27% of them connected, which is down from 29% the year before. Earlier research from MuleSoft and Vanson Bourne, back in 2021, put custom integration labor at around $3.5M a year for the average large enterprise.

Remember that this is directional. MuleSoft sells integration software, and it’s their survey. It’s proof the problem is real at scale, and that’s all you need it for. The $4,300 you calculated is the number that matters, because it’s yours and it holds up in the room.

The self-audit

You can map your own gaps in 15 minutes, no technical skills required. List the tools that share data — CRM, billing, support, marketing automation, analytics — and draw an arrow everywhere data moves between them (CRM → billing, support → CRM, and so on).

Then label each arrow by asking one question: how does the data actually move?

• Connected by API: data flows on its own, on a schedule or in real time. No human touches it.
• Connected by a person: someone exports a spreadsheet on Fridays. This is the human-as-API, whatever the vendor’s website said about “integrations.”
• Not connected at all: the same information gets typed into both tools separately. This is drift’s natural habitat.

Anything in the second and third buckets is a line item you can now price with the formula above.

What to do with the number next time

Most of these gaps were avoidable. They got locked in at procurement, when a tool got chosen on features and price and nobody asked how it talks to the rest of the stack. Your priced sync is what changes that conversation next time.

Before you sign, ask the vendor four things:

1. Is there a public API, and can I see the docs before buying?
2. Does it push data as changes happen, or do I have to keep asking (webhooks vs polling)?
3. What are the rate limits on my tier?
4. How are breaking changes handled — is there versioning and a deprecation policy?

Don’t worry about grading the answers. The signal is whether the vendor has clear ones at all. A vendor who can’t point you to their API docs is telling you who’ll be doing the integration: someone on your team, every Monday, by hand. If the tools don’t talk, you are the API.

We’ll go deeper on procurement in a separate post. Here it’s enough to know that “we’ll just export a CSV for now” is rarely free.

FAQ

How much do disconnected systems cost a company?

It depends on your workflows, but you can estimate it: time per manual sync × frequency × loaded hourly rate, plus the cost of finding and fixing the errors it introduces. Vendor surveys and related research from MuleSoft put custom integration labor in the multi-million-dollar range per year for large enterprises, and find only a minority of enterprise apps actually connected, but your own bottom-up math is more defensible in a budget conversation.

What is data drift?

In strict machine learning terms, data drift is a change in the statistical distribution of the data a model sees in production versus what it was trained on. In this post I’m borrowing the idea for business systems: when the same record lives in two unconnected tools and each copy gets updated separately, they slowly diverge. That silent divergence poisons reports and decisions over time without ever throwing an error message.

What should I ask a SaaS vendor about integration before buying?

Whether they have a public API, whether it uses webhooks or polling, what the rate limits are on your tier, and how they handle breaking changes and deprecation.