Everyone Runs AI Agents, Few Measure Them: Four Numbers to Track First
"We are already using AI agents." That sentence is heard more and more this year. Ask the follow-up, "How much has it saved or earned the company?", and most people cannot answer. A survey released in August by software company Liferay found that 54% of companies are running or piloting AI agents, yet only 25% measure the impact with clear indicators. McKinsey's global survey, published the same month, shows a similar gap: eight in ten respondents say AI has improved their own productivity, while the share who say AI contributes to company earnings has barely moved in a year. Without AI agent KPIs, budgets are hard to defend and nobody can tell which efforts to expand and which to stop. This article covers the key findings and a measurement approach small and midsize businesses can start right away.
The Numbers: Many Use Agents, Few Measure Them
Liferay's 2026 Agentic AI Adoption and Governance Report surveyed 500 US professionals involved in AI decisions, implementation or daily use. Key points:
- 54% of companies are running AI agents in production or actively piloting them, from 72% in technology to 24% in education.
- Only 25% measure AI's impact with clear KPIs.
- Only 24% have a company-wide AI usage policy.
- 42% say better accuracy is what AI agents need most.
- The leading barriers to getting more value from AI are security or privacy concerns (30%), cost (29%) and lack of training (27%).
McKinsey's The State of AI in 2026 surveyed 1,719 respondents in 97 nations in May and June:
- 80% say AI has improved their individual productivity.
- 37% attribute at least some EBIT impact to AI, essentially unchanged from a year earlier.
- Among organizations with at least US$1 billion in revenue, the share scaling agents in at least one function rose from 27% to 40%. Among smaller organizations it stayed at 22%.
- About one in five say AI operating costs, including token costs, already constrain their use.
Read together, the message is clear: adoption is moving faster than measurement and management.
Why AI Agent KPIs Are So Hard to Pin Down
Faster individuals do not equal a better company
A colleague who uses AI to cut a report from three hours to one has made real progress. But nobody knows where the two saved hours went. If the next step in the process still waits three days, overall speed has not changed.
There is no "before" number
Most teams start using AI first and think about measurement later. By then there is no baseline to compare against, only a feeling that things seem faster.
Costs are scattered
Subscriptions, usage-based token fees, integration and maintenance, plus the time people spend reviewing AI output are rarely added into one figure. In McKinsey's survey, one in five organizations already limit use because of operating costs, so the amount is not trivial.
Four Numbers a Small Business Can Measure First
You do not need an elaborate dashboard. Pick one process an AI agent handles and measure four things:
- Time. How long does one task take from start to finish, including human checking and edits?
- Quality. What share of AI output needs to be redone or heavily revised? This maps directly to accuracy, the top need in Liferay's survey.
- Cost. For each completed task, what is the AI cost plus the cost of human review time?
- Outcome. Has the business number this process is meant to improve actually changed? For example, response time to customers, quotes sent, or complaint rate.
The first three are efficiency. The fourth is the purpose. Looking only at the first three makes it easy to conclude that AI works without being able to say how it helps the business.
Three Steps to Make Measurement a Habit
1. Start with one process
Choose a high-volume, repetitive process with a clear start and end, such as first-draft customer service replies, quote preparation or order data entry. If you are not sure which AI agents are running in your company, begin with an inventory. We explain how in How Many AI Agents Are Running in Your Company? Governance Starts with an Inventory.
2. Record a two-week baseline
Before you roll out or change anything, capture the current state in the simplest way: a spreadsheet noting time per task, whether it was redone and what it cost. Two weeks is enough for a starting point. Teams already using agents can start recording today and treat it as the baseline for the next change.
3. Review monthly and make a decision
The point of measuring is to decide. Review the four numbers once a month and choose one action: expand, adjust or stop. Bryan Cheung, Liferay's co-founder, notes in the report that companies need to decide what data agents can access and which actions require human review. Those decisions need numbers behind them too.
Beyond the Numbers: What People Must Judge
Metrics can tell you how fast and how cheaply AI works. They cannot tell you whether the work is worth doing. A useful division of labor:
- AI handles: logging time and usage for every task, compiling reports, flagging unusual figures.
- People handle: deciding what to measure, setting the standard for good quality, interpreting the reasons behind the numbers, and making the final call to invest more or stop.
Some value is hard to express in numbers at all, such as the care a customer feels, or whether colleagues have been freed from repetitive work to do something more meaningful. Managers need to ask and observe those in person.
Conclusion: Measure First, Then Scale
The barrier to adopting AI agents keeps falling. What separates companies is who can say clearly what the agents delivered. Starting with one process, four numbers and a monthly review already puts you ahead of three quarters of companies. That is what TitApeX means by "AI delivers perfection; people make it moving": let AI do the work well, and let people judge what "well" means.
If you want a way of measuring your company's AI workflows that is easy to read and act on, take a look at our services or get in touch.
Comments