A Practical Strategy to Manage Your Small Business AI Stack

Grant Calloway

Here's a number that should make you sit up. The JPMorgan Chase Institute, in a report from April 2026, tracked payments to AI services across 4.6 million small businesses. Firms that started in 2019 took 77 months, more than six years, before ten percent of that cohort was paying for AI. Firms that started in 2025 got there in about six months. That's nearly thirteen times faster. And employer firms adopted at close to twice the rate of businesses with no employees.

Grant Calloway

Now, note what that data actually measures. It's payments, not survey answers. Money leaving the account is the most honest tape there is. The report also says it only counts paid tools, so it may understate real usage. Then add the SBE Council's 2026 Small Business Tech Use Survey. Eighty two percent of small business employers have invested in AI tools. The median business uses five of them. Ninety three percent plan to keep investing, and sixty two percent plan to spend more. Owners aren't asking whether anymore. They're managing a stack.

Grant Calloway

Which brings me back to an earlier episode, 'Turn AI's Saved Hours Into More Sales,' where early credit survey data showed owners feeling productive while sales sat flat. The Philadelphia Fed has since looked at Delaware, New Jersey, and Pennsylvania, using a survey taken in fall 2025. Seventy one percent of AI users there reported higher productivity. Yet over seventy percent said sales hadn't changed. Fifty seven percent described themselves as still experimenting, and only three percent said AI was fully integrated. That's one region, mind you, not the whole country. But the pattern fits. Five tools with no map is just a junk drawer with a monthly bill.

Grant Calloway

There is a map, and it comes from Harvard Business School. Professor Andy Wu and Bharat Anand of NYU laid it out in the Harvard Business Review. Wu's starting point is that a small business owner holds an advantage over the giants. Quote, the CEO in the smaller organization would have a more direct outlook of all the tasks that are taking up time in the organization. No IT committee, no ancient system to protect. You can see the bottlenecks from your own chair.

Grant Calloway

The framework is two questions. First: how costly would a mistake be? Think brand damage, a financial penalty, a customer who walks and doesn't come back. Second: does the task run on straightforward, explicit information, or on human judgment and nuance, the kind of knowledge that lives in an experienced person's head? Cross those and you get four zones.

Grant Calloway

Start in the easy corner. Low stakes, explicit data. First drafts of routine social posts. Transcribing and summarizing calls. Parsing a vendor's inventory sheet. Sorting receipts and incoming leads. Let AI run on these with a light spot check from you. If it botches a hashtag, the republic stands.

Grant Calloway

Now the opposite corner. High stakes, heavy on judgment. Pricing strategy. A sensitive customer dispute. Signing a contract. The final client proposal. Wu and Anand wrote that if an error would lead to serious harm, financial loss, or reputational damage, firms must be far more cautious about using gen AI without human oversight. In this corner AI supports, and you sign. The two zones between are mixed cases, and the sensible rule is to match your oversight to how bad the mistake could be.

Grant Calloway

Consider Andy's Pizza, fourteen shops across Washington, Maryland, and Virginia. Owner Andy Brown used AI coding tools to build internal apps for kitchen screens, phone order tracking, and staff scheduling, plus a voice inventory app that works in English or Spanish. But customers? Quote, we will never replace a cashier with a kiosk. That's my reading, not his words, but it looks like the matrix in practice. Automate the back of the house, keep humans at the counter.

Grant Calloway

Here's the part I like best. Tasks can move on the map. Wu points out that nearly every Zoom call he has now starts with someone asking to record it. What used to be tacit knowledge in the ether can become explicit data. Record the sales calls. Capture the field service notes. Suddenly expertise that lived in one person's head is searchable, and AI can lean on it for back office work without touching customer trust.

Grant Calloway

So what do you do with a median of five tools that may not talk to each other? My suggestion, and this is interpretation, not a finding: anchor every tool to a zone. This week, list every AI tool you pay for. Beside each one, write what a mistake would cost and whether the knowledge it needs is written down. A tool you can't place is a gadget in the drawer, and it's costing you money. Then pick one task that feels like judgment only because nobody ever wrote it down, and start recording it.

Grant Calloway

Wu and Anand warn that your fiercest competition may soon be solo entrepreneurs and micro teams built around AI from day one. Fine. The answer isn't a bigger stack. It's a disciplined one. A tight book beats a big one every time, and that goes for your tools too.