← All Insights

What is AI actually worth? Start with hours, quality, and cost.

4 min readGuildBuild Team
AI business valueProductivityAI evaluation

A useful number has a denominator

In a study of 5,172 customer-support agents, AI assistance increased issues resolved per hour by 15% on average. Results differed across workers. That is evidence about a specific workflow, not a universal saving that can be applied to your whole payroll.

Another experiment spanning 66 firms and 7,137 workers found that active users spent two fewer hours on email each week in its later months. It did not establish a broader change in task output. The working paper also discloses that some authors worked for Microsoft, the maker of the tool studied.

Estimate capacity before you promise savings

Take a recurring task and record how many people perform it, how long it takes, and how frequently it happens. For an illustrative planning example, 10 people spending five hours a week on reporting would recover 15 team-hours a week if a pilot achieved a 30% net time reduction. Over 48 working weeks, that is 720 hours of capacity.

The 30% is an assumption to test, not a GuildBuild result. Include time checking and correcting output. Recovered capacity only becomes financial value when you know how it will be used; it does not automatically remove payroll costs.

Give the pilot four acceptance measures

  • Time: minutes per completed task, including review and exceptions.
  • Quality: share accepted without rework, plus the severity of mistakes.
  • Cost: delivery, model usage, licences, maintenance, and human supervision.
  • Adoption: the share of eligible work people actually complete through the new workflow.

Compare similar tasks across the baseline and pilot. Record the sample size and observation period, and investigate whether harder cases were left out. Set the acceptance threshold with the process owner before the pilot begins.