Short answer: Gartner's 2027 CIO survey found only 13% of CIOs say their company has seen significant or transformational value from AI, and only about one in five AI use cases succeeds. The firms doing better write down the expected value before they start and agree when to stop. Smaller businesses can copy that: one process, clear numbers, a short pilot and a stop rule.
This guide is for owners and managers of small and mid-size businesses who want AI projects that pay off, not more experiments.
What did the Gartner survey find?
Gartner surveyed 2,002 CIOs and technology executives from May to July 2026 and released the results on 19 October 2026.
| Finding | Share or effect |
|---|---|
| CIOs reporting significant or transformational AI value | 13% |
| Chance an AI use case succeeds | About 1 in 5 |
| Have rules for stopping a pilot before production | 31% |
| Have rules for dropping low-value AI already in production | 17% |
| Feel more pressure to cut costs with AI | 58% |
| Expect AI to raise total cost of ownership | 51% |
Gartner calls the best performers "vanguards". Companies with rules for stopping a pilot before production were 2.3 times more likely to be vanguards. Companies with rules for dropping low-value AI already in production were 2.6 times more likely.
These are large enterprises, not small businesses. But the lesson travels well, because a small firm has even less room to keep paying for a tool that is not working.
Why do so many AI projects fail to pay off?
Usually it is not the model. It is the setup around it.
- No number to beat. "Save time" is not a target. "Cut invoice entry from 6 hours to 2 hours a week" is.
- Too many pilots at once. Gartner's analyst put it simply: volume is not value.
- No exit plan. Without a stop rule, a weak pilot drifts on because nobody wants to call it.
- Running costs nobody planned for. Gartner notes AI costs are probabilistic and set at run time. Usage based pricing can grow as more staff use the tool.
- Not connected to real systems. A tool that needs copy and paste from your ERP or CRM rarely saves the time it promised.
How do you pick an AI project that pays off?
1. Start with one process and its current numbers
Pick a task that is frequent, rule based and measurable. Examples: sorting incoming support emails, drafting quotes from a price list, or matching purchase orders to invoices. Record today's time, cost and error rate before anything changes.
2. Write the value and the stop rule before you start
Put both on one page. For example: "If the pilot does not cut handling time by 30% with no rise in errors after 6 weeks, we stop." Agree who makes the call. Gartner stresses this is a team decision, so involve the person who owns the process, not only IT.
3. Run a short, time-boxed pilot
Four to eight weeks with a small group is usually enough to see a trend. Keep a person checking outputs during the pilot.
4. Track running costs from day one
Log subscription fees, usage charges, staff time spent fixing outputs and any integration work. Compare the total with the value you wrote down in step 2.
5. Connect it to your systems only after it passes
Integration with your ERP, CRM or helpdesk is where real savings come from, but it also costs the most. Do it once the pilot has earned it. Our checklist for connecting AI to business systems covers the technical checks before go-live.
6. Review again after 90 days in production
A tool that worked in the pilot can lose value as volumes, prices or processes change. Set a second stop rule for production, the practice only 17% of CIOs have.
Common mistakes to avoid
- Measuring activity (prompts sent, users signed up) instead of outcomes.
- Letting a vendor demo define the target.
- Skipping the baseline, so there is nothing to compare against.
- Treating a stopped pilot as a failure. A quick, cheap "no" is a good result.
Where Ascent24 fits
Ascent24 builds and connects AI workflows and business software, including ERP, CRM and custom apps. When we scope an AI project, we agree the baseline, the target and the running cost estimate up front, and build in the measurement so you can see whether it is working. If the numbers do not justify a build, we will say so.
FAQs
What is an AI stop rule?
A stop rule is a condition, agreed before a project starts, that tells you when to end a pilot or switch off a tool in production. It usually combines a target, a deadline and a cost limit.
How long should an AI pilot run?
Long enough to see a clear trend on real work, often four to eight weeks for a single process. Set the length before you start so it cannot drift.
Do small businesses need a formal AI strategy?
Not a long document. A one-page list of candidate processes, each with a baseline, a target and a stop rule, is enough for most small firms.
Why do AI costs rise after launch?
Many AI tools charge by usage, so costs grow as more people use them. Gartner also found 51% of CIOs expect AI to increase total cost of ownership.
Sources
- Gartner, "Gartner Survey Finds Only 13% of CIOs Say Their Enterprise Has Realized Significant or Transformational Value from AI" (19 October 2026): https://www.gartner.com/en/newsroom/press-releases/2026-10-19-gartner-survey-finds-only-13-percent-of-cios-say-their-enterprise-has-realized-significant-or-transformational-value-from-ai