AI ROI for Small Business: 7 Lessons from Enterprise Leaders That Actually Apply to You
Kiernan McGuire | CEO, TAM Strategy Inc. (Fractional COO Services)
March 2026
I work with small and medium businesses — most of whom tell me some version of the same thing: "I know I should have an AI strategy, but I can't find the time to prioritize it and I think I’m falling behind."
Last week I attended the AI ROI Conference hosted by Section AI, a full day with senior AI leaders from IBM, Cisco, Northwell Health, Zapier, Salesforce, Siemens, and others.
The through-line from nearly every session was essentially this:
"We all keep reading about the big swings — apocalypse or abundance, hype vs. slop. The reality is a lot of work in the trenches."
Even the biggest, most well-resourced companies are figuring this out in real time. The technology is evolving at a furious pace — every three months brings meaningful change. That's both humbling and oddly reassuring for smaller organizations: you're not as far behind as you think, and the playbook is still being written.
Here's what I took away that I believe applies whether you're running a 15-person firm,15,000-person company or even are a solo-preneur.
Seven things enterprise AI leaders taught me — and what they mean for your business:
1. 97% of organizations are still “AI snacking” — Meeting summaries and email drafts aren’t transformational. Real ROI lives at the workflow layer.
2. Your domain experts unlock the value — The person who knows the workflow finds where AI fits.
3. Crawl, walk, run — in that order — One hypothesis, one sprint, one clear outcome. Evolve from there.
4. Vertical beats horizontal — Go deep on one high-friction problem before trying to AI-enable everything at once.
5. Manage AI — don’t be managed by it — Human judgement must not just be “in” the loop, but “on” it.
6. Leadership has to own it — Without executive modeling and protected time to experiment, it stays a side project.
7. Revenue Per Employee is the north star metric — The goal isn’t fewer people. It’s more output from the people you have.
Each section below goes deeper with enterprise proof points and a specific SMB application.
1. 97% of Organizations Are Still "AI Snacking"
One of the most clarifying statistics of the day: 97% of employees who use AI at all are what one speaker called "AI snacking" — meeting summaries, email drafts, the occasional research query. Individually useful. Collectively, not transformational.
The gap between AI dabbling and meaningful AI ROI is enormous. The organizations seeing real returns have moved past the snacking phase into what one framework called three distinct stages:
• Augmenting the workforce — giving employees AI tools and the skills to use them
• Accelerating workflows — redesigning how work actually gets done
• Reinventing core processes — rethinking the business model itself
That third stage takes years. Most organizations aren't there. And there is plenty of value to be mined in the first two.
Moving beyond snacking starts with one deliberate commitment. Pick a workflow with high repetition — client reporting, proposal drafts, intake processing, meeting follow-ups — something your team does constantly and that follows a predictable pattern. That’s your pilot. Then do three things: develop a prompt framework specific to your organization (not a generic ChatGPT prompt, but one built around your terminology, your clients, your standards); build a custom assistant (e.g. a Gem, GPT or Claude Project); experiment with two or three platforms to find what fits your workflow best; and measure results against a baseline you establish before you start — time saved per task, quality consistency, turnaround speed. Without a baseline, you’ll never know if it’s working, and you’ll never be able to make the case to your team that it is.
2. Your Domain Experts Unlock the Value.
Every speaker who talked about successful AI deployment said some version of this: the people who know the workflow have to own the problem-solution.
IBM structures this explicitly — 7-to-9 person "fusion teams" that pair AI technologists with business domain experts, all focused on a single workflow outcome. One example: what would happen if AI reviewed 100% of M&A contracts? The hypothesis, the pilot, the iteration — all driven by people who actually know what a problematic contract looks like.
For SMBs, this doesn't require a dedicated AI team. It means identifying your functional champions — the person in finance who knows where the manual reconciliation pain is, the person in marketing who knows which content creation tasks eat 40% of their week. They're the ones who can identify where AI actually fits, not in the abstract, but in the specific.
Give those people time and permission to invest in AI skills and run experiments. That's where the value gets found.
3. Crawl, Walk, Run. In That Order.
You don't need an enterprise AI platform, a fully loaded governance committee, or an agentic workflow on day one.
The framework that came up repeatedly: start with one specific problem, a small team (or just yourself), a clear outcome you'd recognize if you saw it, and a 1-2 week sprint to test it. Refine. Iterate. Then decide if it's worth scaling.
Northwell Health ran a nine-month pilot with 30 physicians before committing to Ambient, an AI-powered documentation tool integrated with their EPIC system — and went from pilot to 1,100 clinicians using it within five months.
For a solopreneur or small shop, this might look like: spend two weeks testing whether an AI assistant can handle 80% of a specific client deliverable. Measure the time saved. Decide if it's real. That’s the movement from crawl to walk.
All the hype engenders the temptation to skip to running. Every speaker cautioned against it.
4. Vertical Beats Horizontal. Narrow Beats Broad.
Many speakers noted that narrow agents focused on a specific problem are where the most measurable ROI shows up.
One Siemens customer built a virtual engineer agent focused on a single problem: diagnosing production downtime. The result: fix time cut by 85%, 6,000 hours saved annually. Not because they deployed AI everywhere — because they went deep on one problem that really mattered.
For SMBs, the translation is simple: resist the urge to "AI-enable" your whole operation at once or get paralyzed by mistakenly thinking that’s what’s required to unlock value. Think lean agile. Pick your highest-friction, highest-volume workflow. Conduct pilots there first.
5. Manage AI. Don’t Be Managed By It.
One of the mindsets that most resonated from the day came from Cisco’s AI governance team: “I’m learning this at the same time you are. We’re going to manage AI — not be managed by it.” That framing matters more than it sounds from an organizational transformation perspective. It removes the fear, opens up experimentation, and positions AI as a teammate rather than a threat or a magic answer.
It also acknowledges the reality: you’re going to have to manage slop. AI makes it easy to generate something. Making sure that something is actually good still requires human judgment and context. There always needs to be a human on the loop — not just in it. “In the loop” means you’re aware. “On the loop” means you’re accountable for the output.
One practical suggestion that stuck with me: give your AI tools performance reviews. Just like you’d evaluate a team member — what worked, what didn’t, where it fell short — do the same for your AI assistants. Then load that feedback into your knowledge base so the tool actually learns and improves over time. It’s a simple habit that turns a generic AI assistant into something that gets better the longer you use it. (Full disclosure, I haven’t trialed that yet, but it seems to comport with the general strategy of continually improving your system prompts and knowledge base.)
6. Leadership Has to Own It. Or It Stays a Side Project.
This came up in nearly every session. AI strategy that lives in the IT department or gets delegated to a working group tends to stay theoretical. The ones that stuck had explicit executive commitment — not cheerleading, but actual resource allocation and behavioral modeling.
When OpenAI released a major new model in 2023, Zapier's CEO called a "code red" and paused normal operations for a week so every employee — not just engineers, but marketing, finance, and ops — could experiment hands-on with the new tools. Weekly AI users jumped from 10% to over 50% in one initiative.
Northwell Health's CFO sits on their AI executive committee. That sends a signal: this is a financial strategy, not a technology experiment.
For smaller organizations, this is actually simpler. You don't need a committee. You need the owner or CEO to dedicate time for the team to experiment, talk about what’s working (or not) and support transformation. Give your AI champions airtime to share their successes and drive peer-to-peer adoption.
7. Revenue Per Employee May Be the Metric of the AI Era.
There was a lot of discussion about how to measure AI ROI — workflow velocity, cost reduction, revenue growth, employee satisfaction, training adoption rates. Each matters in context.
But one metric kept surfacing kept surfacing as the potential north star:
Revenue Per Employee (RPE).
Not headcount cuts. Companies doing more with more. Napster’s CEO flew to his offshore team in India to spend a day training them on new AI workflows. The result: his entire team produced what used to take a month – in single day. No layoffs. Just more output.
If your best people are spending 25-30%+ of their time on rote, administrative work at the expense of higher-level functions, that’s a leverage opportunity. RPE is one way to measure whether you're solving it.
The Bottom Line
Does this require a massive budget? No.
Does this require a team of AI engineers? No.
What do I need? It requires clarity about where your friction is, domain experts who know your workflows, a willingness to run small experiments, and leadership that takes it seriously enough to invest in clear-eyed organizational transformation.
Pick one workflow. Build one hypothesis. Run one sprint. See what you learn.