How TMetric Measures Whether AI Is Actually Making Your Team More Productive
Most companies didn’t roll AI out as part of some big strategy. It started with someone trying ChatGPT or Claude to get through a task a bit faster. Then someone else started using GitHub Copilot, and another team began experimenting with APIs. Before long there were AI subscriptions, API usage, and a monthly AI bill that kept getting bigger. Very few teams stopped to measure whether the AI was actually making them more productive.
The good news is that measuring it is simpler than most people think. A two-week baseline, a few tags, and one monthly report are enough. This article shows you how to set that up and what to measure once it's in place. Every measurement below is one TMetric already produces as a standard report, so there's no separate dashboard to build.
Why gut feel is not enough — and what TMetric measures instead
One of the best examples comes from a randomised controlled trial by METR. Sixteen experienced open-source developers worked through 246 real issues from repositories they already knew well, with AI randomly allowed on some tasks and not on others. Before starting, the developers predicted AI would make them 24% faster. Afterwards, they estimated they had worked about 20% faster. The measured result was the opposite: tasks took 19% longer when AI was available.
That 19% now comes with a date stamp, and METR says it reflects early-2025 tooling rather than how current models affect developer productivity. However, the gap between what those developers measured and what they believed is the part that still stands. That's the part worth remembering when you're trying to measure AI's impact.
But even when AI does save time, the gains don't always reach the business. Some of that time disappears into meetings, and some into rework.
TMetric found meetings consuming 16.85% of tracked capacity: almost six hours and forty-five minutes per person each week. Fewer than three in ten of those meeting minutes could be linked to a measurable outcome.
: almost six hours and forty-five minutes per person each week. Fewer than three in ten of those meeting minutes could be linked to a measurable outcome.BetterUp Labs and Stanford Social Media Lab found that 41% of 1,150 US desk workers had received AI-generated work that looked finished but lacked substance. Fixing it took just under two hours each time. One person saved time. Someone else spent it.
The three numbers worth tracking — and the TMetric reports behind them
You don't need a complicated dashboard. Three measurements are enough, and each tells you something different about whether AI is actually improving productivity.
AI time against task time
Keep track of the time you spend in AI tools and the time you spend doing the work those tools are supposed to speed up. If AI use goes up and the task takes the same amount of time, you’re using AI without replacing any work. What you want to see is AI time go up while task time goes down. In TMetric, the Apps Usage Report is what shows this: it breaks down time spent in AI tools alongside the tasks they were meant to speed up.
Completion time, before and after
Pick a few tasks that your team does over and over again, such as writing a client report, preparing a brief, reviewing code or doing a QA check. Measure how long they took before the workflow was integrated with AI, then measure them again. Use the median, not the mean, so one unusually slow task doesn't skew the result. This is often the metric that can prove whether AI is actually saving time.. The Detailed Time Report is built for exactly this: filter it to the same recurring task before and after AI adoption to get your before-and-after comparison.
Billable capture rate
If your team bills clients, then track billable hours as a percentage of total tracked hours. If AI is truly driving productivity, then you should be billing more of your team’s time. If AI adoption is on the rise but billable hours are not, then the time is probably being consumed elsewhere, usually in meetings, coordination, or rework. The Tasks Summary Report gives you this rate directly — filter it by Billable status to see billable time as a share of total tracked time.
Setting up the measurement in TMetric
Give yourself one quarter. Spend the first two weeks collecting a baseline before changing anything. Without a baseline, you won't know whether AI improved the work or simply changed how people worked.
1. Collect a baseline
Install the TMetric Desktop app and enable Record apps and sites. TMetric will automatically record which applications and websites people use, how long they spend in them, and activity levels while they work. Leave the workflow unchanged for the first two weeks so you capture normal working patterns before introducing AI into the process.
2. Tag AI-assisted work
Create a tag such as AI-assisted and apply it to the tasks where AI was genuinely used. Continue tracking the same projects and tasks as before. The goal is to compare the same work before and after AI adoption, not to create a separate workflow for AI-generated tasks.
3. Review three reports each month
Rather than building a custom dashboard, review the same reports every month:
- Detailed Time Report to compare time spent on recurring tasks.
- Tasks Summary Report to see where time is being spent across task types.
- Apps Usage Report to understand how AI tools fit into the overall working day.
This gives you a consistent view of both task completion and AI adoption.
4. Compare time with business outcomes
If your team bills clients, don't stop at tracked hours. Compare billable time with the invoices you actually send. AI is creating value only if the time it saves turns into more client work, faster delivery, or higher output, not simply more free capacity. TMetric includes billable time tracking and invoicing features that make this comparison straightforward.
5. Export and review the trends
If you want a deeper analysis, export your reports as CSV or query them through the TMetric REST API. That makes it easy to build your own spreadsheet, BI dashboard, or even ask an AI assistant to summarise monthly trends. The API provides access to Detailed, Tasks, Projects, Team, Activity and Apps Usage reports.
6. Be transparent with your team
Before enabling activity tracking, explain what you're measuring and why. TMetric's productivity tracking is designed to help teams understand how work gets done, and administrators control which activity features are enabled. Make it clear that you're measuring the return on AI tools, not monitoring individual performance.
What a working result looks like in TMetric
TMetric analysed 86 million tracked work minutes from 10,000 opted-in users to understand how ChatGPT was being used at work. On average, employees spent about ten minutes a day in ChatGPT. Based on the tasks where ChatGPT was used, TMetric estimated an average productivity gain of around forty-seven minutes per employee each week.
It helps to scale that up before you decide whether it matters. Forty-seven minutes each, across a team of ten, is just under eight hours a week, or roughly 391 hours over fifty working weeks. On a fully loaded salary of $70,000, TMetric valued the saving at about $28 per employee each week and around $1,458 a year, which works out at close to $14,600 a year for a team of ten. That is the figure to test your own numbers against. (These weekly and annual figures match the numbers published in TMetric's own study.)
What happens next is what will determine if that becomes real business value. If the time you recover results in more client work, faster delivery, or higher output, the gain is real. When this is lost in meetings and coordination or rework, the benefit disappears quickly.
The research also showed that adoption varies by role. Marketing teams spent far more time in ChatGPT than developers or sales teams, suggesting that AI delivers the biggest gains where the work naturally lends itself to drafting, summarising and content creation. The lesson isn't to buy licenses for everyone. It's to see where your own teams get the most value.
When you measure your own results, look at the three metrics together. If AI use is increasing, tasks are taking less time, and billable work is growing, the tools are probably creating real value. If AI use is growing but the other numbers stay the same, people may be using AI without making the business more productive.
Two mistakes to avoid when measuring AI with TMetric
The first is measuring activity instead of results. Time spent in AI tools, activity levels, or keystrokes only tell you that someone was busy. They don't tell you whether the work was completed faster or produced more value. That's why the metrics in this article focus on completed tasks, completion time, and billable work.
The second is looking at AI usage on its own. Knowing the team spent 300 hours in AI tools last quarter tells you very little. The question is what changed because of those 300 hours. Did tasks take less time? Did billable work increase? Without a baseline, there's nothing to compare against, which is why collecting those first two weeks of data matters.
Takeaway: Run the analysis anyway — with TMetric
The hardest part of measuring AI properly is accepting that the results might not match expectations. A tool everyone enjoys using might have little effect on completion times or business outcomes. That's still a useful result because it tells you where the workflow needs improving instead of assuming the answer is buying another AI tool.
Measuring AI productivity doesn't need to be complicated. A baseline, consistent time tracking and a monthly review are enough to replace opinions with evidence.
TMetric's ChatGPT usage study gives you a benchmark to aim at, its activity and app usage tracking captures what your own team is doing, and its reporting features turn that into the before-and-after comparison.
Start a free TMetric trial and measure whether AI is actually making your team more productive.
