Most companies see no return on their AI investment. The approach is wrong, and the technology is rarely the cause. MIT NANDA studied 300 AI implementations, spoke with 52 organizations, and surveyed 153 executives. The conclusion: 95% sees no measurable business results from AI. Only 5% does.

Researchers call this gap the GenAI Divide: the gulf between companies that actually make AI work and the vast majority that keeps getting stuck in pilots.

For SMBs, there is a concrete lesson here. These are the five insights from the research most relevant to your business.

INSIGHT 01 · Using AI tools is not the same as getting results

More than 80% of the organizations studied have tested ChatGPT or Copilot. Nearly 40% say they actively use it. But those tools deliver individual productivity gains, not measurable impact on business results.

The feeling of progress is there. The results are not.

Sound familiar? Colleagues using ChatGPT for emails and summaries, but when you ask what it delivers for the business, silence. That is the difference between experimenting and implementing. And that difference quietly costs you time and money.

INSIGHT 02 · AI pilots fail because of poor integration, not poor models

The second most cited reason AI pilots fail is output quality. But that does not hold up. The same people complaining about poor AI results at work use ChatGPT at home with great satisfaction, on the same types of tasks.

The real problem: the tool does not fit the process. No memory, no context, no connection to how work actually flows. Every session, you explain again who your client is, what the tone of voice is, what was decided last week.

AI does not fail from a lack of intelligence. It fails from a lack of integration.

INSIGHT 03 · Your employees are already using AI, just not through official channels

While official AI projects stall, employees have long moved on. The MIT research shows that at more than 90% of the companies studied, employees use personal AI tools for work, while only 40% of those companies have an official subscription.

They just use it. Multiple times a day. Without IT knowing.

That is a signal. The need is there, the tools work, and your employees are ready. What is missing is an approach that turns that energy into something that moves the business forward, instead of fragmented productivity gains per person.

INSIGHT 04 · Building in-house works twice as poorly as working with a partner

Many companies believe they have more control when they build AI tools internally. The research says otherwise: external partnerships achieve a successful rollout twice as often as internally built solutions.

Not because external parties are smarter. They bring tools that already work in comparable processes, know the pitfalls, and implement faster.

For SMBs, this is especially relevant. You do not have a team of five developers on staff. You have a business to run. The companies that successfully deploy AI do not choose the most impressive demo. They choose a partner who understands their process.

INSIGHT 05 · The biggest AI savings are in the back office, not in marketing

Around 70% of AI budget goes to sales and marketing. Makes sense: those results are visible and easy to report.

But the biggest savings in the research came from the back office. Companies that deployed AI for document processing, customer contact, and administration saw savings of 2 to 10 million per year. External services became redundant. Agency costs dropped by 30%.

For an SMB, that translates directly: the hours lost to manual work, double data entry, and chasing information. That is where AI makes the fastest difference.

What should you do differently as an SMB to make AI work?

SMBs that do get results from AI do three things differently: they buy existing tools instead of building, start with one or two processes, and choose tools that learn their way of working.

That is how Best Byte approaches it at TIG Sports & Events, an events agency of 40+ people. Not starting with building, but first understanding what is going on: through conversations with every department, we mapped where time was being lost and how people were already using AI. What stood out: AI was already being used actively in several places, but without shared agreements on accounts, data, and tone of voice. Some employees were holding back because they did not know what was and was not allowed. The first thing we delivered was therefore not a tool, but an AI policy: business accounts, clear agreements about data, one toolstack for the whole team. Only then the processes.

For AI to work, you need a foundation: processes on paper, data in one place, systems that can talk to each other. Without that foundation, you are giving AI more chaos to work with, not less.

Where do you start with AI in your business?

You do not have to start big, but you do have to start somewhere. At TIG we chose three starting points: invoice checks in accounting, a searchable knowledge base for the thick rulebooks sports federations send with every event, and AI support in guest communication. Everything connects to systems that were already there.

Choose the processes where time leaks or mistakes happen. Map them first, then move to tooling.

Most entrepreneurs we speak with know something is off, they feel it, but cannot pinpoint where. That is exactly what an AI scan helps with: an honest look at where you stand, without a sales pitch.

Kickstart the AI scan →