Build it, buy it, or fix the spreadsheet?

Goldman’s survey of 1,256 owners puts a number on the gap: nearly everyone who uses AI likes it, and almost nobody has it wired in. For an office of five to ten people, closing it usually means buying the subscription and connecting it, so nothing’s typed twice.

The timing helps. On February 24, Intuit and Anthropic announced a multi-year deal to build custom AI agents into QuickBooks (BusinessWire, CPA Practice Advisor), the first due this spring. They follow Intuit’s own agents, pitched last July as “a virtual team working on their behalf while the customer is always in control” (Insightful Accountant). AI is arriving inside software you already pay for.

In Thryv’s January outlook, 67% of its clients expected no impact from AI in 2026 (Thryv). It’s a small sample, and we think many will feel it anyway, the day their books start suggesting the next step.

Three choices, not two

The build-versus-buy guides are written for a hundred staff and an IT manager, and their advice holds for eight (The Systems Effect, Smarterflo): buy for standard work, build where the workflow is your edge. They leave out the option most small offices live with.

• Fix the spreadsheet. Clean the data, lock the columns, connect it to the tool it keeps getting copied from. A few hours of your time, or a small project.
• Buy and connect. Pay per seat for software a vendor maintains, wired to your other tools.
• Build and own. One fixed price for a tool shaped around your office, in your accounts, with an optional care plan behind it.

One guide’s 20-person example has a $25,000 build barely beating a $50-a-seat subscription over three years. Right comparison, old price: our Replace a subscription build starts at $5,000 for one tool, with a second tool priced in the written plan.

How many seats before building wins?

Put that $50 seat against a one-tool build from $5,000, plus upkeep at $370 a month: our Care support plan and typical hosting. We recommend a replacement only when it pays for itself within 18 months.

The ten-seat row is most of our clients, and even over three years it only breaks even. At $50 a seat the line sits near 13 seats. That’s why the rule is 18 months and not three years: a build should never be recommended when it’s marginal for you.

Below the line, the better spend is a QuickBooks integration, from $1,500 for one connection, two systems, so the booking tool talks to the books.

Your price isn’t $50 and your office isn’t exactly ten people, so put in your own seats, price per seat and build price. The calculator counts upkeep the same way.

When the spreadsheet should stay

More often than a software shop likes to admit. A good automation audit asks whether work is frequent, follows the same steps, and costs something visible when it goes wrong (Databrain). A sheet that fails all three is a filing cabinet. Our own test:

A sheet that fails a test is usually cheaper to fix than replace. Our Data cleanup and migration, from $1,500, merges the duplicates and keeps the history. Or we connect it to the tool it keeps getting copied from. The sheet stays in your account, and the rows stay yours.

To find the task worth automating first, January’s piece has a one-week method.

What to buy, and when to build

Books and payroll are always a buy: the rules change every year, and the QuickBooks agents will do more of that work for the same subscription. So are email, calendars, e-signature and the generic parts of a customer list.

So is drafting a flyer. OnDeck and Ocrolus found 56% of small businesses using AI, most of it in marketing, at 63% (Barchart). Building earns its place in two cases:

• The workflow is your edge. How you price a job or get a patient from referral to chair is how you beat the shop down the road. A product built for ten thousand offices flattens it; one built for yours keeps it.
• No product fits. You’ve tried three tools and each keeps a spreadsheet beside it. You’re paying per seat for most of a solution and doing the rest by hand.

Buying is still the default, but “AI now does much of the building, so custom work is affordable where off-the-shelf tools don’t fit” (The Systems Effect). Our Custom software starts at $5,000 for a Client portal and $7,500 for an Operations app, such as a job tracker; the written plan fixes the price.

Count your seats and pull up the invoice for the tool you resent most. Between them, they tell you which of the three choices is yours.