A randomized controlled trial found that using Figma Make made design work 20% faster. But you shouldn’t simply multiply that number by your team’s schedule or labor costs. What you should take from it is not the 20% conclusion, but the measurement method: comparing the same tasks while changing only whether AI was used.
How much should we trust the claim that it’s 20% faster?
Bottom line: there is fairly strong evidence that Figma Make reduced time for the three tasks included in the experiment. Figma’s data science team recruited 100 people: 50 product designers and 50 product managers. Participants were randomly assigned to a Make group or a control group that did not use AI, and both groups edited the same social media screen.
The tasks were switching light mode to dark mode, adding a help item to a settings menu, and creating an interaction where comments appear. They mixed three task types with different difficulty levels, from changing UI appearance to implementing a new view and interaction, and refined the tasks three times through a preliminary pilot. They also used a common facilitator script and troubleshooting guide to reduce variation in how much help each facilitator provided.
Total completion time across all participants fell by 20%, perceived task ease improved by 16%, and perceived usability improved by 15%. Looking only at PMs, completion time fell by 23% and perceived task ease improved by 37%. The analysis used hypothesis testing and OLS regression, and Figma says results without a separate notation are statistically significant.
The interesting part is that the effect was shaped more by which task people were assigned than by who was more proficient. PMs gained significant time on the two relatively easy, short tasks, but not on the hardest interaction task. Product designers, in contrast, saw significant improvement only on the difficult task. So simple rollout rules like “it’s a designer tool, so give it to designers first” or “the more complex the work, the bigger the AI benefit” do not fit these results.
What to carry over directly from this study
They did not compare the usual work time of people who often use AI with those who do not. They spread differences such as role and proficiency through random assignment, gave participants tasks of the same difficulty, and changed only AI access. The closer an internal pilot gets to this structure, the more it can reduce the illusion that “fast people also use AI more.”
Task speed is not the same number as team productivity
The most important limitation is that the 20% figure is total completion time for three controlled tasks, not the release speed or cost savings of an entire product team. The public blog post does not provide raw time by task or confidence intervals, and output quality was not included as a primary outcome metric. Figma also left design quality and collaboration effects as topics for future research.
| What the study established | What it did not establish |
|---|---|
| Total completion time for three specified tasks | Project lead time including research, meetings, and approvals |
| Ratings of how easy the tasks felt | Final quality, including brand fit, accessibility, and usability |
| Differences in effects between product designers and PMs | Effects by an individual team’s proficiency, design system, and work mix |
| The causal effect of Make access in a controlled environment | Financial ROI after subscription and AI credit costs |
Results from AI productivity experiments can change even in direction when the tool and work environment change. A randomized trial of 96 Google engineers estimated that AI features reduced work time by about 21% on complex internal tasks, but its confidence interval was wide, and the researchers said the result could not be applied directly to other ecosystems. In contrast, a METR study found that 16 experienced developers working in familiar open-source repositories took an average of 19% longer to complete 246 tasks when AI was allowed. Participants themselves felt they were 20% faster.
Methods that ask only about perceived time savings also need to be evaluated separately. In the UK government’s AI coding tool trial, 424 respondents reported saving an average of 56 minutes per day, but the report itself listed optimism bias and possible overlap in time savings across tasks as limitations. It was also difficult to track change because respondents could not be linked across survey time points. “How much faster do you think you became?” is a satisfaction signal, not evidence equivalent to stopwatch-measured time.
Cost variables also arise when operating Figma Make in practice. AI credits are used each time you send a prompt, and consumption varies with the model, task complexity, and amount of context being processed. Long chat histories, vague first prompts, and models heavier than necessary can increase rework and credit use. Conversely, Figma’s documentation says small color or spacing changes are faster and more efficient with point-and-edit or direct code edits.
Don’t calculate savings as 20% × salary
For saved time to become an actual labor-cost reduction, it needs to be reallocated to other valuable work or reduce external spend, overtime, or hiring demand. Start by comparing total time per accepted deliverable, then convert only recoverable time into costs.
NIST also notes that AI measurement should treat whether results from a specific task generalize to other use cases or domains as a separate question. So Figma’s 20% is best used not as a buying decision, but as a prior hypothesis worth testing with your own team.
How to measure Figma Make’s impact on your team in two weeks
Before making a large rollout plan, choose three frequently repeated tasks with clear completion criteria and run a small crossover experiment. You cannot call a small-scale result a company-wide causal effect, but it is sufficiently useful evidence for choosing seat expansion and workflows.
Choose three tasks with completion criteria
From recent work, select tasks that take 20–90 minutes, such as a dark mode variant, adding a settings screen using existing components, or implementing a clickable interaction. For each task, fix required states, components to use, accessibility requirements, and the allowed number of errors in a checklist. If you rely only on criteria that vary by reviewer, such as “looks good,” the speed comparison falls apart.
Create similarly difficult A/B versions and mix the order
If the same person makes the exact same screen twice, the second attempt gets faster through learning. Create A/B versions with similar functionality and difficulty but different content. Have half the participants do A with Make first and the rest do B manually first. Use the same devices, libraries, and time limits, and standardize the wording for facilitator intervention.
Record quality, rework, and credits along with time
Record elapsed time from start to submission, active work time, generation wait time, number of revisions, and AI credits used in one row. Have reviewers who do not know the work method score the outputs with the same checklist. Do not count a failed output as savings even if it was fast; add the revision time until it passes.
Set rollout scope using medians by role × task
Calculate the savings rate with (manual median time - Make median time) ÷ manual median time, but always view it alongside the quality pass rate. Expand first from combinations that show an effect, such as simple screen changes for PMs or complex interactions for designers. If time falls but pass rates are low or credits and rework increase substantially, keep that task on the manual workflow.
Finally, retain experiment files and prompt records. With Figma Make, results and credit use vary based on inputs such as a clear first prompt, structured frames, plan mode, and edits that specify particular elements. If the tool version or your team’s guidelines change, you need to measure again on the same tasks to compare with prior results.
If you want to dig deeper
Measuring Time Savings From Figma Make — The source study explaining the participant makeup, three tasks, and role-based results from the randomized controlled trial of 100 people. figma.com
Create a Figma Make file — A menu-level guide covering file creation, attaching frames, plan mode, prompt construction, and credit usage. help.figma.com
Best practices for optimizing AI credits in Figma Make — Official documentation on why vague first prompts and long chat histories increase rework costs, and when to use direct editing. help.figma.com
How much does AI impact development speed? An enterprise-based randomized controlled trial — A study that estimates AI productivity effects through a randomized experiment while also noting wide confidence intervals and limits to generalization. arxiv.org
Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity — Useful for understanding the risks of self-reported time savings: participants felt faster while measured time showed the opposite. arxiv.org
