That’s the question worth asking before you greenlight the next AI project — a chatbot, a reporting dashboard, a tool that tracks reviews or writes social captions or answers the phone. Not “can we build this with AI,” because at this point almost anything can be built with AI. The real question is whether building it yourself gets you more customers or saves you money compared to using something that already exists.
For a lot of business owners lately, the honest answer has been no. And the businesses finding that out the hard way are a lot more recognizable than you’d expect.
Why Does Building It Yourself Feel Like the Smart Move?
It’s an easy trap to fall into. AI tools are cheap to start, the marketing around them promises “no-code” simplicity, and there’s a real appeal to owning something instead of paying a subscription for it. For a business owner used to being scrappy, DIY feels like the responsible choice.
The problem is that “cheap to start” and “cheap” aren’t the same thing. Once you factor in the hours spent testing, fixing, and babysitting a homemade tool — plus the hours not spent running the business — the math often flips.
What Does the Data Actually Say About DIY AI Projects?
This isn’t just a hunch. MIT’s Media Lab spent 2025 studying how companies were actually using generative AI, and the results were sobering: the vast majority of corporate AI pilots never make it past the experimental stage or deliver a measurable financial return, according to the resulting report, The GenAI Divide: State of AI in Business 2025. Researchers reviewed hundreds of public AI deployments and interviewed dozens of executives to reach that conclusion.
The most telling detail in the report wasn’t about the technology itself — it was about who built it. Tools built by outside specialists succeeded roughly twice as often as tools built in-house, a gap the researchers tied to experience with deployment and integration rather than to the underlying AI models being any different. Meanwhile, most of the money companies poured into AI went toward sales and marketing projects specifically, because they’re the easiest to pitch internally — even though the report found stronger returns showed up elsewhere.
In plain terms: the businesses winning with AI generally aren’t the ones building their own tools from scratch. They’re the ones plugging into something that’s already been built, tested, and refined by people who do that one thing for a living.
What Happens When Big Brands Build Their Own AI Tools?
McDonald’s is the example everyone remembers, and it’s a useful one because it wasn’t a scrappy startup cutting corners — it was McDonald’s, working with IBM, with years and real money behind it. Starting in 2021, the chain piloted a custom AI voice-ordering system at drive-thrus in more than 100 locations. By 2024, it was a punchline: TikTok filled up with videos of the AI adding bacon to a customer’s ice cream, piling hundreds of dollars of chicken nuggets onto a simple order, and confidently ignoring corrections. In June 2024, McDonald’s ended the IBM partnership and pulled the technology, saying it wanted to “explore voice ordering solutions more broadly” instead.
Three years of investment from two of the biggest companies in the world, and the result was a tool that couldn’t reliably take a drive-thru order.
Air Canada had its own version of the same lesson. Its website chatbot told a customer about a bereavement discount policy that didn’t actually exist. When the customer tried to use it, Air Canada refused — then argued in a tribunal that the chatbot was “a separate legal entity” responsible for its own words. The tribunal didn’t buy it, and the airline was ordered to honor what its own AI had promised.
Neither of these companies lacked resources. What they lacked was a tool built specifically for the job it needed to do, by people who’d already worked out where things go wrong. That’s a hard thing to build from scratch, no matter how big the brand behind it is.
What’s the Real Cost of “Free” AI Tools?
The sneaky part of building your own AI tools is that the bill doesn’t show up as a line item. It shows up as the afternoon someone spent prompt-testing a chatbot instead of following up with a lead. It shows up as the reviews that went unanswered while a team tried to get a homemade dashboard to pull the right numbers. It shows up as a customer-facing tool that embarrasses the business the one time it matters most.
None of that shows up on an invoice. But it’s real, and for a business whose actual job is selling something other than software, it’s a cost that compounds.
So Before You Build It, Ask Yourself This
Before the next AI project gets greenlit internally, it’s worth running it through one simple filter: is building this thing actually going to get you more customers or save you money — compared to using something that already exists?
If the honest answer is no, that’s not a failure. It’s a sign the project belongs with people who build and refine that exact tool every day, instead of as a one-off experiment bolted onto a website. The businesses treating AI as infrastructure someone else specializes in — rather than a DIY hobby — are the ones actually seeing the return.