Two-thirds of small firms using AI now build their own tools, Pax8 finds

Small and mid-sized businesses have moved from buying AI tools to building their own. In every market surveyed, most do it without a formal, documented review, Pax8 found in a survey published on Monday.

Pax8 is a Denver-based marketplace. It sells AI and cloud software to small businesses through IT service providers. It surveyed 1,150 business leaders in the US, the UK, Germany and Sweden. The results form the Q3 2026 SMB AI Pulse Report. It is the third edition of the quarterly series, and the first to cover Europe.

“SMBs are moving from using AI tools to building AI into the way their businesses operate,” Nick Heddy, Pax8’s president and chief commerce officer, said in a statement. “The opportunity is significant, but so are the new demands around governance, security, cost, and accountability.”

What small businesses are building

Of the businesses that use AI, 67% have built at least one AI solution that is now in regular use. Another 19% are prototyping. Only 13% rely solely on off-the-shelf tools.

Among the builders, 52% are developing automated workflows that connect AI to other business systems. Close to half (46%) are building or testing AI agents that complete multi-step tasks on their own. Another 39% have built custom integrations using an AI provider’s API, and a third are customising or fine-tuning models.

The US leads, with 81% of businesses that use AI running solutions they built. The figure is 76% in Sweden, 54% in the UK and 52% in Germany. Sweden has the most builders working on AI agents, at 55%.

Last month, Meta launched Muse for small businesses, linking its AI agent to Shopify.

The AI nobody approved

In the past six months, 35% of businesses that use AI discovered shadow AI. Pax8 defines it as tools, workflows or agents running without official approval. Sweden had the highest rate, at 55%. The US followed at 42%, then the UK at 24% and Germany at 19%.

Formal, documented review of AI built in-house ranges from 23% in the UK to 45% in Sweden. In the US, 36% of builders have formal review. Another 41% use review that is consistent but undocumented.

Policies on how staff use AI are more common. Some 40% of businesses have a documented policy and 35% have informal guidelines. Pax8 noted that such policies do not cover the agents and workflows a business builds itself.

Respondents named the risks of AI tools that employees build. In every market, data security or privacy exposure topped the list, cited by 30% to 35%. Last month, a Dataiku survey found that 81% of CIOs lack full oversight of AI agents built outside IT.

Bills higher than expected

Of the businesses that paid for AI in the past year, 49% got a bill or usage charge higher than they expected. That rose to 57% in the US and 65% in Sweden. Germany was lowest, at 33%.

US businesses pay through usage-based pricing (46%), fixed subscriptions (41%) and per-seat pricing (40%). Pax8 said usage-based charges are the hardest to forecast. Some 69% said they would be more likely to expand AI with outcome-based pricing at a similar cost.

Pax8 also asked about data control. Seven in ten businesses would pay at least 6% more for greater control over how providers store and use their data. In the US, the figure was 81%.

Time saved, time spent checking

Some 69% of businesses that use AI said it had freed up at least some capacity. But 55% of decision-makers personally spend three or more hours a week reviewing, editing or fact-checking AI-generated work, and 7% spend ten hours or more. In the US, 68% spend at least three hours.

Last month, a PwC survey of workers found most are falling behind on AI skills.

In all four markets, businesses that use AI extensively were more likely to feel ahead of competitors than those still experimenting. The gap was 63% against 40% in the US, and 80% against 24% in Sweden. The US is the only market where fear of falling behind (38%) outweighs fear of costly AI mistakes (32%).

How the survey was run

Pax8 ran the survey online among owners, executives and managers involved in technology decisions at companies with 5 to 499 employees. It polled 400 in the US, 350 in the UK, and 200 each in Germany and Sweden.

The margin of error is about 2.9 percentage points for the full sample and 6.9 points for Germany and Sweden. Pax8 said results for smaller subgroups should be read as directional.

Pax8 works with more than 47,000 IT partners. Its report says the findings point to a growing role for partners that can help small businesses inventory, secure and govern the AI they build.