Short answer: many companies now have AI tools running, but far fewer can say what those tools earned or saved. Three surveys published in 2026 point to the same gap: too few companies name an owner for the result, or count the work before the AI arrives. The fix costs almost nothing. Give each AI project one owner and one number, and write the number down before launch.
What the surveys found
Genpact, which sells AI-run operations services, surveyed more than 2,000 senior executives at companies with annual revenue of $1 billion or more and released the findings on 12 November 2026. 79% of the companies have AI agents in production, though only 20% use them broadly and consistently across functions. Only 15% have fully tied their AI investment to profit-and-loss outcomes. On average, 49% of the AI initiatives they funded in the past 24 months delivered the value expected. The answers are self-reported.
Closer to home, Adobe and Incisiv surveyed 475 senior executives in India, all with authority over AI investment, for Adobe's 2026 State of AI Readiness research. Only 18% of the Indian companies have a designated AI owner with budget authority and accountability for outcomes. Adobe also says only 10% of companies measure AI return on investment comprehensively, and that most track usage rather than quality, rework or the real cost of reviewing AI-generated output.
The State of AI in Indian Enterprises 2026 report from ET Edge CIO&Leader drew on 300+ senior technology leaders in May and June 2026. As CXOtoday reported, just 12% of organisations can point to significant, measurable return on their AI investment, and 57% either have no measurable return or can't tell whether returns exist. CXOtoday says the report treats this as a measurement failure rather than a performance failure.
Genpact's sample is all large companies, and two of the three surveys come from firms that sell AI products or services. Still, the lesson carries over to a 40-person business, and it's easier to apply when you're small.
Why usage numbers fool you
"Logins went up" or "the team sent 3,000 prompts this month" tells you people opened the tool. It doesn't tell you the work got faster, cheaper or better.
Take an AI that drafts quotations in two minutes. If your senior salesperson then spends twenty minutes checking and fixing each one, it may cost more than it saves. A usage report will still look great. Adobe's list is the useful one here: quality, rework, and the time spent reviewing what the AI produced.
Give every AI project one owner
Pick one person whose own result depends on the project. The sales head owns the enquiry follow-up tool. The accounts lead owns invoice checks. It shouldn't be the person who set the tool up, and it shouldn't be "the team".
The owner decides whether the tool stays, changes or gets switched off. In Adobe's India data, 93% of the 30 companies it calls Leaders have a designated AI owner with budget authority, against 18% of companies overall.
Pick one number, and record it before launch
Choose a number the owner already cares about, and write down today's value before the tool goes live. When Adobe asked executives what they'd do differently, they mentioned setting clearer metrics and naming an owner earlier. By the time they did, the chance to set a baseline had passed.
Some examples, by business process:
- Sales: time to first reply on new enquiries; share of leads followed up within 24 hours.
- Marketing: hours spent each week building the marketing report; cost per qualified lead.
- Operations: invoices processed per person per day; share of documents sent back for correction.
- Leadership: days taken to close the monthly MIS; hours of manager time spent chasing updates.
A week of honest counting in a spreadsheet is enough for a baseline. You don't need a dashboard.
Count what it costs to keep running
Next to the number, write down what the AI costs each month:
- subscription or usage fees;
- time people spend reviewing and correcting its output (the one most teams leave out);
- time spent fixing it when a template, price list or process changes.
If review time eats the saving, either the checks sit in the wrong place or the task isn't ready for AI yet.
Check at 30 and 90 days, then decide
At 30 days, look at the number and the review time together. At 90 days, the owner makes a call: expand it, fix it, or switch it off.
Switching off a tool that didn't move its number is a good outcome. You learned it in three months for the price of one subscription. The ET Edge report recommends that boards insist on a minimum, rigorously measured return benchmark before authorising more AI investment. In a small company, that can be as simple as a rule: no second AI project until the first one shows its number.
If you're already paying for several AI tools and nobody can tie them to a number, a one-hour tool audit is a sensible first step.
A one-page scorecard
Keep one row per AI project:
| Project | Owner | Number | Baseline (and date) | Target | Monthly cost incl. review time | 30 days | 90-day call |
|---|---|---|---|---|---|---|---|
| Enquiry first replies | Sales head | Median reply time | 3 h 10 min (week of 2 Nov) | Under 15 min | Fees + 2 h review a week | Expand / fix / stop |
The example row is made up to show the format. Put your own numbers in it, and keep the sheet where the owners can see it.
Arthat AI
Working on something similar at your company?
Arthat AI builds these automations for Indian businesses every week. If you want to talk through whether this fits your situation, we'll give you a straight answer — even if the answer is "don't build it."
