There’s an idea spreading among business owners that’s worth looking at calmly: rushing to adopt AI is worse than not adopting it at all.
That’s not just our opinion. Domo’s Chief Data Officer has said it publicly: enough with the AI FOMO, it’s time to slow down. And while that may sound counterintuitive coming from someone who sells data platforms, the argument makes a lot of sense for any SME thinking about taking the leap.
In this article, we’ll explain what’s going on, why so many companies are failing in their AI projects, and how to avoid those mistakes in your own business.
The context: a quiet rebellion against urgency
Over the last two years, the dominant narrative has been clear: "adopt AI now or get left behind". That pressure has pushed thousands of companies to buy Copilot licenses, ChatGPT Enterprise subscriptions, all kinds of "AI-powered" tools… only to realize months later that the ROI is nowhere to be found.
This isn’t just a gut feeling. There’s serious data behind it:
- A 2025 MIT study found that 95% of generative AI pilots in companies were not producing measurable returns.
- Gartner predicts that by the end of 2025, at least 30% of generative AI projects will be abandoned after the proof-of-concept phase.
- McKinsey’s State of AI 2024 found that although 72% of companies use AI in some function, only a minority report a significant impact on EBITDA.
The question is not whether AI works — it does, and very well. The question is why so many companies are failing to implement it.
Why AI projects fail in SMEs
After working with dozens of businesses, the failure patterns are surprisingly repetitive. Here are the most common ones:
1. Starting with the tool, not the problem
The classic mistake: "We need to bring AI into the company." That’s not a goal; it’s a trend. It’s like saying "we need to bring electricity" in 1910 without knowing what you want to light up.
AI is a cross-cutting technology. Without a specific problem to solve, it becomes an expense with no return.
2. Confusing "trying ChatGPT" with "implementing AI"
Giving your team access to a generic tool is not digital transformation. It’s like handing someone a hammer and expecting them to build a house.
AI applied to a business requires:
- Integration with your current systems (CRM, email, calendars, ERPs)
- Understanding of your company’s specific context
- Processes designed around the new capability
- Control and oversight mechanisms
3. Underestimating the hidden cost of DIY
Many founders try to build their own workflows with Zapier + ChatGPT + a spreadsheet. It works for two weeks. Then it breaks over a weekend, no one knows how to fix it, and they go back to the old method.
AI in production needs:
- Error handling and retries
- Logs and observability
- Maintenance when APIs change
- Continuous improvements as real feedback comes in
4. Adopting without measuring
If you don’t know how long it takes you to respond to a lead today, you won’t be able to tell whether AI made you faster tomorrow. Most companies don’t have a baseline, so they can’t prove the value of what they’ve implemented.
The slow-mo approach: go slow to go faster
The idea is not not to adopt AI. The idea is to adopt it with judgment. In practice, that means:
Step 1 — Audit before you automate
Before touching a single line of code or buying a tool, you need to understand where time is really being lost. An honest audit usually reveals things like:
- 4 hours a day lost to manual lead follow-up
- 3 hours a day managing the inbox
- 1–2 days for each sales proposal
- 20–30% of the team’s time spent copying and pasting between apps
Without those numbers, any AI investment is an act of faith.
Step 2 — Start with a concrete, measurable problem
A good first AI project meets these criteria:
| Criteria | Why it matters |
|---|---|
| Repetitive | AI shines at tasks that repeat many times |
| Well defined | If you can’t describe the process step by step, AI won’t be able to either |
| Data available | Without data from the current process, there’s nothing to train on or measure |
| High opportunity cost | The more expensive the wasted time, the faster the ROI |
| Low risk if it fails | Start with something where a mistake won’t sink the company |
Step 3 — Build, measure, iterate
The first version won’t be perfect. That’s okay. What matters is:
- Launch something functional quickly (days, not months)
- Measure the real impact using the same KPIs you had before
- Iterate based on data, not intuition
Step 4 — Expand only when the first case works
Once you have a measurable win, you now have:
- Internal confidence in the technology
- A team that knows how to work with AI
- A repeatable pattern for the next use case
Only then do you expand. Not before.
Cases where "going slow" produces fast results
This isn’t theory. When this approach is applied to specific problems, the results come quickly:
Lead follow-up: From hours of response time to under 60 seconds. Response rate +35%, pipeline +22%. 4 hours of manual work eliminated per day.
Inbox management: From 3 hours to 15 minutes a day. Urgent emails handled in 8 minutes instead of 2–3 hours. Escalations reduced by 90%.
Sales proposals: From 1–2 days to 10 minutes per proposal. Capacity tripled (from 3–4 to 10–12 proposals per week).
What all the cases have in common: they started with a concrete problem, measured before and after, and built a tailored solution — they didn’t buy a generic tool and hope for miracles.
The trap of "AI will do everything"
There’s a message going around that’s worth shutting down: the idea that AI will soon replace everyone and we need to run just to avoid falling behind. The reality is much more sober.
Today, AI is excellent at:
- Processing natural language at scale
- Classifying, enriching, and routing information
- Automating decisions with clear rules
- Freeing humans from mechanical tasks
Today, AI is not a good option for:
- Critical strategic decisions without supervision
- High-level creative tasks without human guidance
- Processes where mistakes are costly and there’s no control
- Fully replacing professional judgment in regulated areas
Understanding this difference is what separates an implementation that adds value from one that adds chaos.
What you should do this week
If you’re thinking about adopting AI in your business, before buying anything, do this exercise:
- List the 5 tasks that take the most time from you or your team each week.
- Estimate the real cost: hours × salary × 52 weeks. It usually gets your attention.
- Mark which ones are repetitive and well defined. Those are your candidates.
- Choose just one. The one with the highest cost and lowest risk.
- Define the result you expect in 30 days. Specific, measurable.
With that, you already have a foundation that’s a thousand times better than 90% of companies that are just "doing things with AI."
In summary
The rebellion against AI is not really against AI — it’s against bad AI implementation. Against hype, against generic tools sold as silver bullets, against projects that promise revolutions and deliver expensive subscriptions.
Going slow doesn’t mean standing still. It means taking firm steps in the right direction, instead of running in circles chasing the latest trend.
At Studio SmartWork, we work exactly that way: we start by understanding the problem, design a tailored solution, deploy it in less than a week, and keep it running. No generic templates. No vendor lock-in. No promises we can’t prove with numbers.
AI is not magic. It’s engineering applied to real business problems. And like all engineering, it works best when it’s done with a clear head.