We’ve been having the same conversation on repeat here at Sketch. A business owner sits down with our team, and within five minutes, they say some version of the same thing:
“I know I need to do something with AI. I just don’t know where to start.”
They’re not uninformed. Most of them have read the breathless LinkedIn posts about how AI will replace everyone by next Tuesday. They’ve seen competitors mention AI on their websites. They’ve tried ChatGPT a few times. They’ve experienced analysis paralysis while reading about “91 AI tools to try today so you don’t fall behind.” Somehow, the mass hysteria hasn’t helped them get started on their first custom AI solution.
These are perfectly competent business leaders, and they’re in the majority. A 2025 survey by Reimagine Main Street and the NSBA found that most small businesses are stuck in what I’d call the “AI curious” stage. They’ve stalled in the experimentation phase. They’re extremely interested, and believe they stand to benefit, but can’t find a practical way to move forward.
Compare this to a stat from the U.S. Chamber of Commerce, which reports that 96% of small business owners plan to adopt AI. The intent is nearly universal, but most people haven’t moved past basic chatbots.
If you look back at the quote above, you might see why it’s so hard to make progress: This person has started with the solution, and is trying to work backwards to the problem. Our clients tend to have better luck when they start with a business problem, then build the right solution for their problem.
When someone comes to Sketch and says they want to build software, we don’t hand them a tech stack. We ask what they’re trying to accomplish. What’s slowing their team down? What’s costing them money? Can they show us how their system works today? What manual process makes their best people want to quit? AI should be treated the same way.
Using a certain tool is not the end goal. If you have pictures to hang, a hammer can help you achieve better outcomes. If you build a hammer and look for things to hit... We’ve all seen SaaS companies make things worse by cramming AI in where it doesn’t belong. Start with a problem worth solving.
The benefit of AI is that it gives us more ways to address business cases. It changes what’s now feasible, and how much friction you have to accept. We still evaluate investments in similar ways, but problems that were too expensive to solve six months ago might not be so daunting because of how AI accelerates our development workflows.
Based on the AI projects we’ve delivered, we’ve seen a natural progression in how businesses adopt AI. These three options represent rising levels of maturity. Most companies can get concrete wins the fastest on the first level. If one of the other levels presents an obvious use case, though, there’s no need to pass up a good opportunity.
The simplest place to start is a chatbot trained on your company’s specific knowledge. You can build a proof of concept for this with a custom GPT or a Claude project, and maybe it can be a good way to practice building your first AI agent, but you’ll get more control by taking it further than that. That means a custom tool built on your internal documentation, product specs, policies, or procedures.
We built something like this for a regional bank. Their employees were spending hours each week tracking down answers to the same recurring questions about internal processes. A custom AI chatbot gave them instant, accurate responses pulled directly from the bank’s own knowledge base.
The time savings were immediate, but the client didn’t stop there. They advanced through greater levels of AI adoption, adding functionality like a loan underwriting AI agent. Then they became even more efficient, completing loan narratives 8x faster.
The next step up is identifying work where employees spend significant time pulling information from multiple sources and cross-referencing it manually. Think compliance checks, product specification lookups, vendor comparisons, or regulatory research.
Related Reading: AI Tools Across Job Functions (2025 Guide)
For one chemical manufacturing client, we built an AI solution to automate a research process their sales team was doing by hand across multiple technical sources. In the past, a salesperson would have to go to a computer for manual cross-referencing. Now they can get a comparison in seconds (on their phone, while they’re still with the customer).
Once you’re comfortable with AI handling questions and research, look for places where your team generates the same types of content over and over:
We don’t recommend outsourcing human judgment entirely. However, AI is great at handling the first draft and time-consuming formatting. For example, one client with a job search consultancy came to us for an AI resume app to free themselves up for coaching and mock interviews.
A lot of the AI content out there is designed to make you anxious. “You’ll be left behind your competitors if you don’t have 20 AI agents running at all times!” The best custom AI developers I know aren’t resorting to metaphors about earthquakes and tidal waves to stir up hysteria on LinkedIn.
This isn’t an attempt to invalidate concerns about AI. That would be insane. The point is that there’s a lot to be excited about, too, and handwringing isn’t helping anyone.
The Salesforce SMB Trends Report found that 91% of businesses surveyed already using AI say it boosts revenue. We’ve seen it with our own clients, too (and in our own business). If you want to get excited, choose a process where you can measure the time, cost, or revenue, then watch how you can improve things with AI. The companies leaning into this new technology are growing.
Start by identifying a specific, recurring problem that eats up your team’s time. Common first projects include internal knowledge chatbots, automated research workflows, and repetitive document generation. Ideally, pick a use case where you can measure the time or cost savings clearly.
No. You’d need a data scientist for some intense AI/ML applications, but certainly not to get started with AI. A software developer who understands AI APIs and integration patterns can build all kinds of practical AI solutions. You’re best off with a team that understands both the technology and your business context. That way you end up with something people actually use.
We’ve had engagements under $50,000 and over $1,000,000. The best approach is to start with a focused use case, prove the value, and expand from there. In other words, your first AI project should be way on the small end of the range.
A custom chatbot built on your internal documentation is a different investment than an enterprise AI solution that integrates with multiple databases, workflows, and systems. At Sketch, we deliver working software every two weeks with no change orders, so you can see progress, adjust direction quickly, and stop whenever you have enough.
ChatGPT and similar tools are general-purpose. They’re useful for personal productivity, but they don’t know your business, your data, or your workflows. Custom AI software integrates directly into your operations and is trained on your specific information. General AI tools give similar, generic answers to more than a billion people. Custom AI solutions are your competitive advantage.
Hopefully you feel more prepared to get started with custom AI for your business. If you’re still a little hesitant, well, one blog post probably wasn’t going to fix everything anyway. In either case, you don’t have to move forward alone. We built our AI Launchpad for exactly this reason, and we’re giving away free consultations.