I don’t swear often, but I’ll admit to saying words I’d never repeat in a meeting while trapped in an automated phone tree, pressing “4” for the third time, trying to reach a human being who could answer a question the system was never designed to handle.
Somebody built that system, and they almost certainly hit their targets: call volume, headcount down, cost per contact—all down. Every dashboard number moved the right way, but the customer’s experience got measurably worse. That gap between what got measured and what happened is what I’d most want an SMB leader to avoid when bringing AI into their business.
If you’re an SMB leader, you’ve likely heard the same message repeatedly: adopt AI or fall behind. Salesforce found that three out of four SMBs are investing in AI, showing adoption is now expected, not differentiating.
But investment alone doesn’t guarantee impact. Nearly 70% of SMBs remain in the experimental stages of AI maturity, and larger organizations that have used AI for years still struggle to make it a repeatable way of working.
Part of this challenge is the sheer pace of change. As new tools, features, and platforms enter the market, businesses feel pressure to layer AI onto existing processes. Others hesitate because the options make it difficult to know where to start. In both cases, businesses struggle to realize AI’s full value because they don’t know where it can have the greatest impact.
SMBs succeed when they identify where repetitive tasks create bottlenecks and where AI can solve a specific operational problem. Once that’s clear, AI can save time, reduce friction, or improve output.
AI Only Creates Value When the Workflow Is Ready
The most common pitfall in AI adoption is plugging into tasks one at a time before asking where it can create value. A business might save a few minutes here and there, but isolated instances don’t change how work gets done.
The deeper issue is that most SMBs run on fragmented workflows. Customer data lives in one platform, marketing campaigns in another, project management in a third. When those systems don’t talk, employees spend more time moving information than using it. Bolting AI onto that fragmentation adds another place to manage (and it doesn’t help that using AI well can feel like a second job, turning leaders into part-time prompt engineers).
To create lasting impact, AI should move work forward by connecting information, surfacing insights, and automating repetitive tasks.
This doesn’t require a massive overhaul. The goal is to move beyond isolated use cases and build unified workflows that consistently improve efficiency and deliver measurable business value. These six steps provide a practical framework.
6 Steps for SMBs to Meaningfully Implement AI
1—Start with the workflow problem, not the tool
Identify where routine tasks create bottlenecks or pull your team away from higher-value work. The strongest use cases are recurring, time-consuming and tied to predictable outcomes. Customer support inquiries, project-management follow-ups and competitor intelligence qualify, because they follow recognizable patterns and can be standardized without sacrificing quality.
If you’re unsure where to start, ask AI how to use AI. Describe your business, team and the parts of the week that eat time, then ask where it would look first. It’s a low-cost way to get a map of the territory before committing to a tool.
Then define success before turning anything on. Decide what “better” means and record where things stand today: how long the task takes, how often it goes wrong. That baseline cannot be reconstructed later, and companies that set success metrics before deployment consistently see better returns than those measuring afterward. With the problem and finish line defined, you’ll know which capabilities and tools you need.
2—Evaluate the technology already in place
When we first looked, half the AI capability we needed was already inside tools we paid for but weren’t using. The buying reflex is strong; the audit reflex isn’t.
Our experience isn’t unusual, so once you’ve named the use case, audit what you have before making a shopping list.
Then assess how well those systems work together. Can information move directly between platforms, or are employees copying data from one tool to another? Technologies such as Model Context Protocol (MCP) make it easier to create pathways between tools that couldn’t previously communicate, reducing costly replacements.
While you’re there, examine the data, because its quality determines AI outcomes. Inventorying existing tools, systems, and records can reduce spending and prevent further fragmentation, creating a more connected stack that moves information efficiently.
3—Experiment deliberately and build trust
Our first pass at an AI support draft got the tone wrong in a way that would’ve embarrassed us in front of a customer. Instead of killing the project, we kept a human in the loop until the system earned its way out.
Expect that pattern: the first output will rarely be perfect, but the use case may still be right. It may need better context, clearer instructions or refinement. Adjust how AI drafts a follow-up or flags a missed task before relying on it broadly.
Human oversight remains essential. Rather than giving AI full autonomy, begin by having it observe the workflow and recommend actions while people own the final decision. Then identify gaps, refine and build confidence before expanding AI’s role.
For example, AI might review refund requests, determine whether they meet policy, and draft a response for approval. As recommendations improve, AI can handle routine requests, allowing employees to monitor outcomes and step in on exceptions.
4—Measure results and refine
AI has to create measurable value. This is where the baseline from step one pays off: you’re comparing against captured numbers, not a memory. Rather than tracking AI usage, measure revenue-generating activity, human output and customer experience. Most businesses get the first two and forget the third, which is how you build a phone tree.
Take a massage therapist. If AI handles bookings, reschedules and reminders, she might fit in one more appointment. Revenue up, output up. But if the system frustrates people into booking elsewhere, she’s optimized her way into a smaller business. All three measures must hold.
Another example is how we rebuilt our own customer support. We trained an AI agent on support and education materials, and customers connect with it first. They describe the problem, the agent works through it, and only unresolved issues reach a human. The ticket volume our team touched dropped dramatically: that’s the output leg.
The experience leg improved too, but could have gone south. We’re not staffed to give every customer a dedicated success manager, so people previously waited on email. Now they get immediate help. Satisfaction and resolution rates held and often improved. If they’d moved the other way, the efficiency gain wouldn’t have been worthwhile.
Internally, the same approach moved output: AI-enabled workflows increased our pull request volume by 200% and cut cycle times from six days to roughly 12 hours.
Weak results are a signal, too. If a workflow isn’t improving anything, revisit the problem or the data feeding it.
5—Watch where the bottleneck moves
Here’s something I didn’t expect: AI doesn’t always remove a bottleneck. Sometimes it relocates one.
When our engineering team got significantly faster, product and design had to decide what to build at a pace they weren’t set up for. If we hadn’t caught that, we would have just built chaos and paid for it later.
So, when you accelerate one part of your business, look immediately at what sits on either side of it. The work has to go somewhere. Capacity created in one place becomes pressure in another, and the win only counts if the whole system got better.
6—Start small and let your early adopters lead
Pick one problem and one outcome (don’t try to overhaul the business at once).
Then, think about who runs the pilot. Every team has people who will try anything and people who wait to see evidence. Both are valuable, and the cautious ones are often the reason you haven’t taken a bad risk yet, but they need different things. Give the first workflow to an early adopter. When it works, their result is what moves everyone else. Evidence persuades faster than enthusiasm, and that’s exactly how adoption spread through our engineering team.
If you’re a business of one, that’s simpler and harder: you’re the early adopter. Pick the workflow that annoys you most and start there.
Turning AI Into Business Value
For SMBs, AI can feel like a moving target. New tools emerge constantly, expectations keep shifting, and the pressure to move fast is real.
But adoption was never about keeping up with every new release. It’s about the less glamorous work underneath: finding where your processes can be improved and applying AI there, deliberately, one proven use case at a time.
Be skeptical of the noise. There are still very few legitimate examples of AI running repeatable, complex workflows end-to-end. Most companies are figuring it out, even those on stage, and a fair number of the loudest voices are selling a course or a consulting engagement. Confidence is part of the product. Believe none of what you see and half of what you hear.
Instead, measure progress against your own business three months ago, not against someone else’s keynote. You will never know where they actually are. Is AI helping remove friction, create capacity, and improve how your team serves customers? Those are the outcomes that matter.
Aleks Bass is a product and technology leader known for bringing a research-driven lens to AI, product development, and decision-making. With a background spanning consumer insights, product leadership, and enterprise-scale platforms, she challenges conventional AI narratives by focusing on where automation actually delivers value, where it breaks down, and why human judgement is the connective tissue between efficiency and automation and rich, customer-centric outcomes.
At Typeform, she is helping redefine how businesses move from data collection to action, while building a more disciplined, human-centered approach to AI adoption.