Ninety-one percent of small and medium businesses using AI said it boosts their revenue, according to a December 2024 Salesforce survey. That number is technically accurate, but it only tells us what businesses that had already adopted AI reported. It tells us much less about the businesses that haven't adopted it yet.
Census Bureau data covering December 2025 through May 2026 puts overall AI usage among US businesses at 17% to 20%. Among firms with fewer than 20 employees, the number is lower still, and it barely moved over that six-month period. U.S. Census Bureau, Large Firms With at Least 20 Employees Biggest AI Users. A Federal Reserve FEDS Notes analysis published in April 2026 by economist Jeffrey S. Allen cross-checked that figure against two other national surveys and landed close by, at roughly 18% adoption as of year-end 2025. Federal Reserve, Monitoring AI Adoption in the U.S. Economy.
So the honest starting point for most small and mid-sized businesses isn't "everyone else is already ahead of us." AI adoption is still relatively early, especially among smaller firms. The Census data makes that clear.
If you're an SMB or ecommerce operator, that's actually the useful part of this data. There's no wave to catch up to, just a short list of places where AI can measurably cut manual work, and a way to prove it moved a real number instead of taking someone's word for it.
Having AI Isn't the Same as Operational AI for Small Businesses
A lot of these numbers count things differently than you'd expect. A manager who occasionally asks ChatGPT to draft a product description counts as an AI user in most surveys. So does a company that has automated its order-to-fulfillment pipeline with a system that reads incoming orders, checks inventory, flags exceptions, and updates three other tools without anyone touching a keyboard. Only one of those is actually changing how the business gets work done.
| Workflow Dimension | Basic / Ad Hoc AI | Connected Operational AI |
|---|---|---|
| Core approach | Isolated prompts (e.g. ChatGPT) | Automated system wired into existing tools |
| Data integration | Manual copy-paste between apps | Direct read/write sync (Shopify, ERP, Xero) |
| Decision making | Human handles every transfer step | System applies business rules & flags exceptions |
| Team impact | Slight personal convenience | Measurable reduction in hours & data entry errors |
| ROI measurement | Unclear or difficult to prove | Baseline vs 60–90 day before-and-after metrics |
Operational AI means the second version: AI connected to an actual business process, working with real data, following rules someone set, and taking action in the tools you already run. This distinction matters most for smaller operators, who rarely have a data science team or six months to spend before seeing a result. For a smaller business, starting with one narrow, connected workflow can be a more practical way to test whether AI is actually creating value, with a number attached to it before anyone builds anything.
The 2026 Operational AI Playbook: 4 Steps to Prove ROI
●1. Pick One Measurable Workflow, Not a Department
"Automate customer service" isn't a project, it's a wish. "Automate the return-authorization step where someone manually checks order history and approves or denies a refund" is a project.
●2. Write Down the Baseline Before Touching Anything
Hours per week, volume of orders or tickets, current error rate. Without this, there's no way to prove the automation did anything.
●3. Connect the Tools Instead of Adding a New One
The manual work usually exists because two systems don't talk to each other and a person is the connector, copying data from Shopify into a spreadsheet, then into QuickBooks. The fix is rarely a new app. It's a system that reads where the data already lives and writes back into the tools you already use.
●4. Measure the Same Number Again After 60 to 90 Days
Don't switch to a new metric just because it looks better. Use the exact same one from step two. If it moved, that's your result. If it didn't, that's useful information too, before you expand the project.
If you can't explain the workflow step by step on a whiteboard, don't automate it yet. Map it first.
Real Examples, Not Enterprise Case Studies
CloudFO's finance team was pulling numbers from Shopify, Xero, QuickBooks, Amazon, and several bank accounts, then manually stitching everything together every month, with mistakes creeping in along the way. Axonari connected all of these into one system that could pull and reconcile the numbers automatically. According to CloudFO, reporting became 62% faster, forecast accuracy improved 45%, and reporting errors fell 70%.
Sidechain's HR team had a version of the same problem: candidate information retyped across tools that didn't share data, with strong applicants slipping through because matching took too long. Once the systems were connected, Sidechain reported an 80% drop in manual data entry, 50% faster candidate matching, and a 30% improvement in hiring accuracy.
Neither company started with "we need AI." Both started with a workflow slowed down by disconnected tools, and both got fixed by connecting those systems rather than adding a new standalone tool on top.
A few common mistakes to avoid: starting with the tool instead of the workflow, skipping the baseline measurement so there's no way to prove anything changed, and treating one pilot as a full AI strategy instead of the first proof point.
How Axonari Helps
We use the same approach with every client. We don't start by asking which AI tool to buy. We start by mapping the workflow, identifying where time and accuracy are actually being lost, and finding the specific point where a connected system can read the data, make a decision inside clear rules, take the action, and hand off to a person only when it should. CloudFO and Sidechain, above, are both ours. See how we approach AI systems projects →
Where to Start
AI adoption is still relatively early, especially among smaller businesses. And even among companies using AI, many are starting with a limited number of tasks or business functions. That gap creates an opportunity: instead of trying to automate everything, pick one slow, manual workflow and prove the value first.
Pick one slow, manual workflow. Write down what it costs today. Connect the systems instead of adding another one. Check the number again in 60 to 90 days.
Quick Checklist
Sources
U.S. Census Bureau, Large Firms With at Least 20 Employees Biggest AI Users, Business Trends and Outlook Survey, May 2026. Read the data
Federal Reserve, Monitoring AI Adoption in the U.S. Economy, FEDS Notes, April 2026. Read the analysis
Salesforce, New Research Reveals SMBs with AI Adoption See Stronger Revenue Growth, published December 2024. Read the survey
Axonari, Case studies: CloudFO and Sidechain. Read our casework