The Gap Between Using AI and Running On It

Count your business tech subscriptions before you read on. Most owners find more than they expected, and some are more obsolete as time moves on.

Goldman Sachs asked both questions in its 2026 10,000 Small Businesses Voices survey, which polled 1,256 owners who had been through the Goldman Sachs small business program, and the two came out very differently: 76% of the owners surveyed use AI in some form, and 14% have it embedded across core operations.

Why the adoption number looks so good

The same survey found 93% of AI users describe the impact as positive. Read that alongside the 14%, and the two stop looking contradictory.

A positive-impact question asks a person to compare their work now against their work before. Drafting a client email in three minutes instead of fifteen is positive. Cleaning a spreadsheet in five minutes instead of forty is positive. The person doing that work is right, and they'd say so honestly.

What that question doesn't reach is the distance between where the firm is now and where it could be. In a business where everybody built their own private workflow, with no shared context and no agreement about what the tools may touch, the people are genuinely better off and the business has claimed a fraction of what's available.

Which is why the survey's third finding matters. 73% of those owners say they would benefit from additional training and implementation resources. They can tell what they have is makeshift. They built workarounds and would like a system.

Five tools, bought quickly, with capabilities that now overlap

The SBE Council's 2026 small business technology survey found that 82% of small-business employers now invest in AI tools, and that the typical small business runs a median of five of them.

Just in the past six months, review the capabilities of the AI services you’ve added to your workflow. Only one may have been able to manage the creation of photo assets; now three can. One was incredible for application work; now four are. Each was initially purchased to give your firm an advantage, but the competition between each other’s services is worth a second look at their current value to your business. There is no need to increase tech debt for every member of your team.

You can check the cost of that today in three places. Are these capabilities duplicative or are they nuanced for different use cases? Is your company's knowledge base now fractured, and is that causing agent drift? Are you seeing the end product diminish as a result of handoffs at each step in the process?

Automation is growing, which changes what a missing system costs

Marc Zao-Sanders's June 2026 Harvard Business Review study of more than 12,000 real AI use cases found people turning AI toward a wider mix of tasks, including more agent-like work where the whole task is handed over.

We'd draw the consequence out. Under augmentation, somebody reads every output and catches the errors. As tasks move toward automation, fewer eyes sit between a mistake and a client.

Evidence from a different setting is worth holding next to that. Datadog's 2026 State of AI Engineering report, which observes teams building AI products rather than firms buying them, found around 5% of production AI requests failing in February 2026, most of them because they hit capacity limits. Engineering teams instrument for exactly this. A five person studio doesn't.

What the 14% did

They ran the same tools through a system.

One place the work lives. A source of truth holding the canonical version of every client, project and decision, so your tools read from one record.

A written page on what AI may touch. You name the approved products, the prohibited categories, the data rules, and a route for exceptions. One page, and it makes shadow tool use unnecessary instead of forbidden.

One workflow integrated end to end, then the next one. You pick a single process where the handoffs connect, prove it, and repeat. Nobody needs a platform migration to do that.

None of this is a second job laid on top of the business. You do it inside the work you already have, one workflow at a time, and the firm keeps running while you do.

What you can do this week

Write down every AI subscription the business pays for, who chose it, what it does, and what information lives inside it. Take an afternoon to review the new capabilities of each service since you signed on, and maybe have a conversation with your rep to understand their future release schedule.

All Why Adoption Stalls Briefings