# Best AI adoption analytics tools in 2026: logins are not impact

Best-of list · By Nolli research, The Nolli team · Published 2026-10-17 · Updated 2026-10-17 · Prices and features checked 2026-10-16

Canonical page: https://withnolli.com/blog/best-ai-adoption-analytics-tools-2026

**Short answer.** To see who uses AI and how often, Microsoft’s Copilot Dashboard (if you run Copilot), Worklytics and Larridin do it. To see whether the work changed, only Jellyfish (software teams), ActivTrak (time use) and Nolli say they measure it. Elsewhere, “impact” is usage times an assumed number or a survey. We read the public pages on October 16, 2026.

## Quick picks

- **Best if you run Microsoft 365 Copilot: Microsoft Copilot Dashboard.** Documents its method down to the minutes per action, and needs no separate license to view. It counts Microsoft’s own AI only.
- **Best across several AI tools, with a price: Worklytics.** Reads usage metadata from Copilot, Gemini, ChatGPT and Claude, and publishes a plan from $2,500 a month. Its “impact” is a correlation.
- **Best for software teams who want outcomes: Jellyfish.** The one entry that says it joins AI use to delivery records such as throughput and quality. Software teams only, quote only.

## Key takeaways

1. All eight tools count who uses AI; only Jellyfish, ActivTrak and Nolli describe measuring a change in the work itself, and Jellyfish does it for software teams only.
2. Microsoft’s “Copilot assisted hours” is a count of actions multiplied by six minutes, so it moves when use moves, whether or not the work improved.
3. At Okta, 74 engineers reported working about four fewer hours a week after Copilot, while monthly pull requests, review cycle time and lines of code did not change significantly.

Every tool on this list answers one question well: who is using AI, and how often. The question most buyers carry is a different one: is it changing the work? The two sit far apart. In a case study on its own site, ActivTrak describes a software company of about 200 people where license data showed more than 95% had accessed AI tools and session data showed more than 30% of users barely used them. A login counted. A session of a few seconds counted.

We read each vendor’s public pages on October 16, 2026, and ran none of the products. For each one we asked how it measures: usage logs, surveys, collaboration data, or the work itself. Then we wrote down what the page says that measurement can show about a change in the work. Our guide to measuring AI impact explains why that second step matters.

## Four ways to measure, and what each cannot say

Every product here uses one of four methods, and most use two. The method decides what the dashboard can know, so it is worth reading before the feature list.

The methods are our grouping of what each vendor’s page describes. Only the cited sources support the “cannot say” column.

| Method | What it counts | Used by | What it cannot say |
| --- | --- | --- | --- |
| Usage logs | Active users, sessions, actions, tokens. Pulled from admin APIs, browser extensions, SSO or email metadata. | Microsoft, Worklytics, Larridin, Torii, Nudge Security, ActivTrak, Jellyfish | Whether the work got better, faster or cheaper. In the ActivTrak case study a session of a few seconds counted as use. |
| Surveys | What people say about time saved and satisfaction. | Microsoft (Viva Pulse and Glint), Larridin, Jellyfish | What actually changed. METR says its early-2025 study found people overestimated AI’s effect on their time on tasks by 40 percentage points on average. |
| Collaboration data | Meeting load, chat, calendar and email metadata, and who works with whom. | Worklytics (a customer quote on its site names organizational network analysis) | Whether the work product changed. It sees the pattern around the work, not the work. |
| The work itself | Output, quality, cycle time or time per task on a defined workflow, from system records or observed work. | Jellyfish (software delivery), Nolli (by workflow), ActivTrak (time by activity) | Work with no clean record, and anything before you have a baseline. It needs a defined workflow to compare. |

## The 8 tools, in order

> **How we chose.** We read each vendor’s public product pages, documentation and pricing pages on October 16, 2026. We tested nothing and talked to no vendor. One rule sets the order, applied to what the page itself says. First, does the page describe a measure of the work (time use, output, quality, cost per task), or only a measure of use? Second, within the same group, how much of the method is published: the signals, the formula, the limits. That gives three groups. Jellyfish, ActivTrak and Nolli describe measuring the work. Microsoft, Worklytics and Larridin turn usage into impact with an estimate, a correlation or a survey. Torii and Nudge Security count use, spend and risk. Nolli is ours and follows the same rule. It lands third, behind two vendors that publish more of their method. The order is a judgment from documents. A security team or a CFO would order it differently. Where a page was silent, we say so.

### 1. Jellyfish AI Impact

Best for software engineering organizations that want AI use tied to delivery results. Jellyfish is an engineering management platform, and AI Impact is its module for AI coding tools such as Copilot, Cursor and Claude Code. Its page says it tracks adoption, usage, spend and outcomes (speed, quality, productivity), and links usage to throughput, quality and delivery speed using signals from engineering systems.

Strengths: The only entry that says it joins AI use to outcome records. The integrations listed on its page include Git, Jira, GitHub and Azure DevOps.; Adoption comes from automatically detected system signals, with no survey needed to see who uses what.; Compares several coding assistants and agents on one measurement model, and shows token spend by tool, team or initiative..

Limits: Software teams only. Nothing on the page covers finance, support or operations work.; Quote only. The page quotes customer results with no method or sample behind them, so we repeat none of them..

Pricing: Quote only. Jellyfish says pricing depends on the number of seats and the modules you select.

Vendor page: https://jellyfish.co/platform/jellyfish-ai-impact/

### 2. ActivTrak AI Insights

Best for seeing how time at work shifts as AI use grows, with a price you can read. ActivTrak is workforce activity software, and AI Insights is its AI layer. Its page says it shows which AI tools are in use, approved or not, where adoption has stalled, and how AI-assisted work changes where employees spend their time. Its case study says it captures time in application at the session level.

Strengths: Separates a login from real use. In its case study of one software company with about 200 employees, license data showed more than 95% had accessed AI tools and session data showed more than 30% barely used them. That is the vendor’s own write-up of one customer.; Places teams on a maturity scale from Stage 0, non-users, to Stage 5, end-to-end AI orchestration.; Publishes a price: Productivity Optimization is $17 per worker per month billed annually, and the pricing page lists AI Insights as a $4 add-on..

Limits: Time in an app says how long, not whether the result was good. The AI Insights page names no output or quality measure.; Its base product tracks worker activity, and the pricing page lists screenshots as an add-on. Agree with legal and HR what you switch on before rollout..

Pricing: $17 per worker per month (Productivity Optimization, billed annually) plus $4 per worker per month for AI Insights, per its pricing page.

Vendor page: https://www.activtrak.com/solutions/ai-insights/

### 3. Nolli (made by Nolli)

Best for measuring adoption and efficiency by team and workflow from the work itself. Nolli learns how work runs from people’s work on company devices, the email and calendar the company connects, and the logs of the AI agents people use. For adoption it shows who uses which AI tools, skills and agents, how often and on which work. For results it measures the same workflow done by people, by people with AI and by agents, on time, quality and cost per task.

Strengths: Adoption and results come from one source, the work, so a team’s AI use can be read next to what happened to its time, quality and cost.; Agents run in shadow mode first and are scored against what your people did, before anything goes live.; Your company decides what it learns from. The trust page says passwords and login, checkout and billing pages stay out, keys, card numbers and ID numbers are removed before any AI model sees the work, and leaders see workflows and patterns across teams..

Limits: It sees company devices, the apps your admin allows, and the mail, calendar and agent logs you connect. A personal account on a personal device is outside its view.; No public price, and less public method detail than Microsoft. Nolli has no published customer results, so there is no number here to check.; The before-and-after comparison works per workflow. Ask how it treats work it has not mapped..

Pricing: No public price. Request a demo.

Vendor page: https://withnolli.com/solutions/ai-enablement

### 4. Microsoft Copilot Dashboard (Viva Insights)

Best for companies that run Microsoft 365 Copilot and want a first look this week. The Copilot Dashboard sits in Viva Insights and covers four groups of metrics: readiness, adoption, impact and sentiment. Adoption counts active users over the past 28 days and sorts them into power, habitual and novice users. The impact page shows Copilot actions, an estimate of assisted hours and a satisfaction rate from thumbs-up and thumbs-down reactions.

Strengths: The most open method on the list. Microsoft’s page spells out how assisted hours are computed, down to the minutes per action.; No paid Viva Insights or Copilot license is needed to view it. Tenants with 50 or more Copilot or Viva Insights licenses also get benchmarks, agent insights and survey data.; Sentiment can come from Viva Pulse or Glint surveys, or a file you upload. Advanced reporting lets analysts upload their own business data to measure impact on the metrics they choose..

Limits: It counts Microsoft Copilot, Copilot Chat and agents inside Microsoft 365. Nothing on the pages we read covers ChatGPT, Claude or Gemini use.; Assisted hours are an estimate, and Microsoft says the method will change. The data is a rolling 28 days with up to a six-day delay. A known issue under-counted email-draft and summary metrics from June 1, 2025 to February 18, 2026..

Pricing: No separate price to view it. Microsoft says neither a paid Viva Insights license nor a Copilot license is required.

Vendor page: https://learn.microsoft.com/en-us/viva/insights/org-team-insights/copilot-dashboard

### 5. Worklytics

Best for HR and IT teams that want AI use across several tools next to collaboration data. Worklytics is workplace analytics software with an AI adoption product. It connects to Microsoft Copilot, Gemini, ChatGPT, Claude and others, tracks usage by team, tool and role, and sorts AI activity into work categories such as coding, research, drafting and email. It says it links usage to calendar, email and project tool signals to compare high and low adoption teams.

Strengths: Works from metadata. Its site says it does not store or analyze work content, only log data, and that it aggregates and pseudonymizes to the group level.; Publishes plans: a free tier for up to 100 users on calendar data only, and a Business plan from $2,500 a month.; Offers raw AI usage data for export to your warehouse or BI tools. The pricing page puts data pipeline access on the Enterprise plan..

Limits: Comparing high and low adoption teams is a correlation. Early adopters may differ in other ways, and the pages we read describe no baseline or staggered rollout.; The pricing page disagrees with itself: one card says Business is for 100 to 1,000 employees, and the FAQ says it includes up to 200 users. The AI page also still dates peer benchmarking to summer 2026, so confirm it exists..

Pricing: Published. Free for up to 100 users (calendar data only); Business from $2,500 a month; Enterprise custom.

Vendor page: https://www.worklytics.co/measureai

### 6. Larridin

Best for finding every AI tool in use, approved or not, and steering people to company accounts. Larridin deploys a browser extension and a desktop application that detect AI tools and agents, classify what people use them for, and report adoption by team, role and location. Its Scout product says it covers more than 300 AI tools, and combines passive usage detection with user surveys to compare time savings by team.

Strengths: Tells personal accounts from enterprise ones and, per its page, can allow, warn or block in the browser or on the desktop.; Per its FAQ, most sensitive data is processed in the browser, and only summary statistics go to Larridin’s cloud.; Shows spend and unused licenses next to adoption..

Limits: The productivity side rests on users’ own reports and surveys. Its FAQ lists business impact intelligence as roadmap.; Tracking is computer-based only and misses iOS, and browser detection works at the URL level. The FAQ also mixes shipped features with what customers are interested in, so ask which are live..

Pricing: Not published on the pages we read.

Vendor page: https://larridin.com/ai-adoption/

### 7. Torii AI Management Platform

Best for IT and finance teams that want AI spend and seat use in one view. Torii sells SaaS management and an AI Management Platform that its page marks as early access. Its dashboards show estimated token value, active users and sessions, adoption by department, and engineering signals such as commits and code changes.

Strengths: Ties tokens to a department, manager or cost center, and flags licenses nobody uses.; Its own June 2026 article says it finds AI tools through browser activity, SSO logs, finance data and OAuth grants..

Limits: Dollar figures are an estimated token value at standard API rates, not your invoice, per its support page.; The only measure of work is engineering activity. A click on a person opens that person’s own usage and productivity view, so decide who may see it..

Pricing: Not published.

Vendor page: https://www.toriihq.com/products/ai-management-platform

### 8. Nudge Security

Best for security teams that need to find AI tools and accounts first, with usage as a side benefit. Nudge Security finds AI apps, accounts, agents and data integrations. It starts from email metadata in Microsoft 365 or Google Workspace to spot sign-ups, then offers a browser extension for deeper signals such as how often AI is used, file uploads and paste events.

Strengths: Discovery starts with an email integration, with no proxy or endpoint agent.; Publishes a price: $750 a month up to 150 users, then $5 per user per month from 150 to 1,500 users, billed annually..

Limits: It is a security product. The pages describe use, risk and policy, and no measure of whether work improved.; The browser extension monitors AI conversations for sensitive data, such as file uploads, paste events and API keys. Settle that policy with legal and HR first..

Pricing: Published. Essential $750 a month up to 150 users; Growth $5 per user per month for 150 to 1,500 users, billed annually; Enterprise custom.

Vendor page: https://www.nudgesecurity.com/use-cases/ai-security

## What each tool measures

**What each tool’s public pages say it measures** (none, partly or fully)

| | Jellyfish | ActivTrak | Nolli | Microsoft | Worklytics | Larridin | Torii | Nudge |
| --- | --- | --- | --- | --- | --- | --- | --- | --- |
| Usage by AI tool | fully | fully | fully | partly | fully | fully | fully | fully |
| Usage by team or role | fully | fully | fully | fully | fully | fully | fully | partly |
| Time use or time per task, from logs | partly | fully | fully | partly | partly | partly | none | none |
| Output or quality, from work records | fully | partly | fully | partly | partly | none | partly | none |
| Compares users with non-users, or before with after | partly | partly | fully | partly | fully | partly | none | none |
| Survey or self-report built in | partly | none | none | fully | none | fully | none | none |
| Privacy controls described | none | partly | fully | fully | fully | fully | none | none |

Our reading of each vendor’s public pages, read on October 16, 2026. Full: the page says it plainly. Half: it says part of it, or the figure is an estimate, as with Microsoft’s assisted hours. Empty: we found nothing, which does not prove the tool lacks the feature. “Usage by AI tool” is half for Microsoft because it counts Microsoft’s own AI. Nolli is ours and is read by the same rule.

## Use against change

**How much each tool measures use, and how much it measures change in the work** (x: How much it measures use, 0 to 100; y: How much it measures change in the work, 0 to 100; corners: top-left Little use data, much change, top-right Use and change, bottom-left Neither, bottom-right Much use data, little change)

- Nudge Security: x 78, y 6
- Torii: x 85, y 20
- Larridin: x 90, y 45
- Microsoft: x 80, y 44
- Worklytics: x 68, y 52
- ActivTrak: x 58, y 62
- Jellyfish: x 66, y 84
- Nolli: x 52, y 74

Positions are our reading of the public pages we read on October 16, 2026, not measurements. Left to right: from tools that report little about who uses AI, to tools that detect every AI tool and every account. Bottom to top: from counts of use, to estimates, correlations and surveys, to measures of the work. Nolli is ours and is placed by the same reading. The order of the dots matters more than the exact spot. Jellyfish measures change for software teams only.

## How a usage count becomes an “impact” number

Microsoft shows the cleanest example, because it publishes the formula. Its impact page reports “Copilot assisted hours”, and its documentation says the metric is computed from employees’ actions in Copilot and multipliers drawn from Microsoft’s research. Here is the chain for the two biggest buckets.

**From actions to dollars in the Copilot Dashboard**

1. **Count the actions**: Searches, summaries and creation actions in Word, Outlook, Excel, PowerPoint, Teams and Copilot Chat.
2. **Multiply by minutes**: 6 minutes per search or summary action and 6 minutes per creation action. Meeting summaries count the meeting’s length.
3. **Add up as hours**: The sum over the past 28 days is “Copilot assisted hours”.
4. **Multiply by a rate**: Default $72 an hour, set from US Bureau of Labor Statistics data. You can change it.
5. **Show as value**: “Copilot assisted value”, an estimate of what the assistance was worth.

From Microsoft Learn, read on October 16, 2026. Microsoft ties each six-minute figure to a study it cites: 163 knowledge workers for search and summaries, and 147 for a writing task. It calls the result a general estimate and says the method will change.

Nothing in that chain looks at the work. Double the number of summaries and the hours double, whether anyone read them. This is not a flaw Microsoft hides; the page says so. It does mean the figure answers “how much was Copilot used?” in hours. The same shape appears elsewhere: Larridin’s Scout product combines passive detection with surveys (its newer Workflow Intelligence page describes comparing AI-assisted and non-AI runs of the same workflow, and its FAQ still lists business impact as roadmap), and Worklytics correlates usage with other signals. Only the tools in the first group of our order describe reading the work.

## What the studies say about the gap

The closest public look at what usage-plus-survey misses is a field study at Okta, published by Fei Liu and Todd McKinnon in Communications of the ACM on September 17, 2026. They followed engineers through a GitHub Copilot rollout. They invited 261 people who received licenses between July and September 2024, and 97 answered. Of those, 74 had complete before and after data. The authors compared April to June 2024 with September to November 2024, joining survey answers to pull request data from the company’s engineering tools.

- **44.77 to 40.83**: weekly work hours the Okta engineers reported before and after Copilot, a drop of about four hours. The authors also report 1.06 fewer hours a week on coding and 1.56 fewer on testing. Self-reported, 74 engineers. (source: Liu and McKinnon, Communications of the ACM, September 17, 2026, https://cacm.acm.org/research/beyond-the-hype-the-efficiency-throughput-gap-with-github-copilot/)
- **No significant change**: in monthly pull request count, pull request cycle time or lines of code for the same engineers, measured from the company’s engineering systems. Only the number of reviews performed rose significantly. (source: Liu and McKinnon, Communications of the ACM, September 17, 2026, https://cacm.acm.org/research/beyond-the-hype-the-efficiency-throughput-gap-with-github-copilot/)
- **7%**: of the survey respondents reported no Copilot use in the three months after getting a license. The survey group used Copilot more than license holders as a whole, so the authors say the picture may be too rosy. (source: Liu and McKinnon, Communications of the ACM, September 17, 2026, https://cacm.acm.org/research/beyond-the-hype-the-efficiency-throughput-gap-with-github-copilot/)

Read what each number is. The hours are the engineers’ own report. The pull request figures come from the company’s systems. The authors call the distance between them the efficiency-throughput gap. The caveats are theirs and ours. It is one company, Copilot as it was in late 2024, a before and after comparison and not a randomized trial, and a small non-user group that cannot carry a conclusion.

A second study asks the same question on small tasks. Sunny Yu, Myra Cheng, Robert Hawkins and colleagues at Stanford, NYU, MIT and Princeton ran three pre-registered studies with 2,691 people recruited on Prolific, a US sample. In one, 600 people predicted how much time AI would save on short tasks that take under five minutes, and 1,001 others did the tasks with or without a chatbot.

**Seconds saved per task by AI: predicted against measured**

- What people predicted AI would save: 55.7 s
- What AI saved when people did the tasks: 7.5 s

Source: Yu and colleagues, arXiv 2605.22687, May 2026, https://arxiv.org/html/2605.22687

These are simple tasks and a general sample, so do not carry 7.5 seconds into a finance close. The point is the gap between the guess and the clock. In the same paper, people also predicted they would use AI less often than they did. A survey can miss use in one direction and benefit in the other.

METR’s survey, run from February to April 2026 with 349 technical workers (87 software engineers, 71 researchers, 129 academics and PhD students, and 48 founders and managers), is what a survey-based impact figure looks like. The median respondent reported a 3x change in speed and a 1.4x to 2x change in the value of their work. METR’s own caution: survey results are not necessarily grounded in reality, the sample is a convenience sample, and response was about 2% among people it emailed. Treat any dashboard that reports hours saved from a survey as this kind of number.

## Pick this, skip that

- **Pick Microsoft if** you already run Microsoft 365 Copilot and want adoption by team this week. Read assisted hours as an estimate, and add another tool only if people use other AI.
- **Pick Worklytics if** you use several AI tools, want group-level data with no work content, and want to read a price before a sales call.
- **Pick Larridin if** your first problem is shadow AI and personal accounts. Scout’s impact side leans on surveys; check how far its newer Workflow Intelligence product has come.
- **Pick ActivTrak if** you want time in AI tools by team on a published price, and legal and HR accept activity tracking.
- **Pick Jellyfish if** most of your AI use is in software teams and you want it joined to delivery records.
- **Pick Nolli if** you want adoption and results by workflow from the work itself, across more than coding, and a demo is fine in place of a price.
- **Skip Torii unless** IT and finance own the question and it is about spend. Its AI product is marked early access.
- **Skip Nudge Security unless** security owns the question. It will tell you what AI is in use and what data went in, not what the work became.

Whoever carries the question at your company is often a Head of AI Enablement. What that job owns is worth a read before you choose, because the buyer decides which of the two questions gets answered.

## What to ask every vendor

**Eight questions, and get the answers in writing**

- [ ] Which number on the dashboard is measured, and which is estimated? Ask for the formula.
- [ ] What does “active” mean: one action in a month, or repeated use? Microsoft’s power user is 15 or more actions a week in at least 9 of the last 12 weeks. Another vendor may use another bar.
- [ ] Does it count only the vendor’s own AI, or every tool, including personal accounts?
- [ ] Is the impact figure a before and after against a baseline, a correlation between groups, or a survey?
- [ ] Can it show a comparison group that has not adopted yet?
- [ ] What is the smallest group it will show, and can a manager see one person’s data? Microsoft hides groups under a minimum size. Torii opens a person’s own view.
- [ ] What does the extension or agent read: URLs, prompts, files, or none of them?
- [ ] Which features ship today, and which are roadmap?

> **Counting AI use is also counting people** Group-level views protect individuals, and some tools default to them. Others drill down to a named person. Agree in writing who may see what, and check the rules in your country with legal and, if you have one, your worker council. This post is not legal advice.

## Mistakes buyers make

1. **Reading activation as adoption.** A license and a login are cheap to count. The ActivTrak case study shows how far they can sit from real use at one company. Do this today: Define “active” as repeated use over several weeks, and write it down before the first report.
2. **Letting a multiplier stand in for results.** Hours computed from actions rise whenever actions rise. They do not look at the output. Do this today: Time ten real tasks before and after on one team, then compare with the dashboard hours.
3. **Treating a survey as the result.** Self-reports run high. METR says so about its own survey, and the Okta engineers reported fewer hours while output measures held steady. Do this today: Ask what people did with the time saved, then check one output measure.
4. **Comparing early adopters with everyone else.** People who adopt first differ in other ways. At Okta the survey group used Copilot more than all license holders did. Do this today: Roll out in stages and compare teams that have the tool with teams still waiting.
5. **Buying one tool for two buyers.** IT wants spend and risk. The business wants to know if the work improved. Most tools here answer the first well. Do this today: Write down which of the two questions the budget owner asked.

## What this reading cannot tell you

We ran none of these products. Every cell above is a statement about a page, not about the product. A vendor can ship what its page does not mention, and the reverse. We found no public test of how closely any of these dashboards agrees with a timed baseline, so we cannot say which is more accurate. We left out every customer result and market count on vendor pages, such as the average number of AI apps per company, because none gives a method or a sample. The one exception is the ActivTrak activation figures, used once and labelled. The Okta study is a single company in 2024. The arXiv study uses simple tasks. METR’s survey is a convenience sample. None of them is a verdict on a tool in this list.

## Where Nolli fits, and where it does not

Nolli measures adoption and efficiency by team and workflow from the work itself: who uses which AI tools, skills and agents, and how the same workflow performs done by people, by people with AI and by agents, on time, quality and cost per task. It learns from work on company devices, the mail and calendar your company connects, and the logs of AI agents. It does not see personal accounts on personal devices. It has no public price and no published customer results, so there is no number here for you to check. If your question is who uses a Microsoft tool, start with Microsoft’s dashboard. If it is whether the work changed, AI enablement is what Nolli is for, and our guide shows the baseline method on its own. Request a demo to see it on your own work.

## Questions people ask

**What is an AI adoption analytics tool?**

It is software that counts who uses AI tools and how often, usually by team. It reads admin APIs, browser extensions, or signals from email, calendar and work systems. Some tools add spend, surveys or measures of output.

**Can these tools measure AI ROI?**

Most give an estimate: usage times an assumed number (Microsoft uses six minutes per action), a survey, or a correlation between groups. Jellyfish says it joins usage to delivery outcomes for software teams, and Nolli measures time, quality and cost per task on a workflow. Read how the figure is made before you trust the ROI.

**I run Microsoft 365 Copilot. Do I need another tool?**

Start with the Copilot Dashboard. Microsoft says no paid Viva Insights or Copilot license is needed to view it. It counts Microsoft’s own AI, so add another tool only if your people use ChatGPT, Claude or Gemini.

**Which of these tools publish a price?**

Three of the eight. ActivTrak lists $17 per worker per month plus $4 for AI Insights. Nudge Security lists $750 a month up to 150 users, then $5 per user per month to 1,500. Worklytics lists Business from $2,500 a month. The others are quote only or unpublished.

**What is the difference between adoption and impact?**

Adoption is how many people use AI and how often. Impact is the change in the work: time, quality or cost against a baseline taken before the change. Our guide to measuring AI impact covers the baseline.

**Is it legal to track employee AI use?**

It depends on the country, on what is collected and on whether people are told. Some tools aggregate to groups by default and others drill down to a person. Ask your legal team, and your worker council if you have one, before anything goes live.

## Sources

- Fei Liu and Todd McKinnon. "Beyond the Hype: The Efficiency-Throughput Gap with GitHub Copilot." Communications of the ACM, September 17, 2026. https://cacm.acm.org/research/beyond-the-hype-the-efficiency-throughput-gap-with-github-copilot/
- Sunny Yu, Myra Cheng, Ahmad Jabbar, Ilia Sucholutsky, Katherine M. Collins, Dan Jurafsky and Robert D. Hawkins. "The efficiency-gain illusion: People underestimate the rate of AI use and overestimate its benefits on simple tasks." arXiv (2605.22687), May 2026. https://arxiv.org/html/2605.22687
- Joel Becker. "Measuring the Self-Reported Impact of Early-2026 AI on Technical Worker Productivity." METR, May 11, 2026. https://metr.org/blog/2026-05-11-ai-usage-survey/
- "Connect to the Microsoft Copilot Dashboard for Microsoft 365 customers." Microsoft Learn, accessed October 16, 2026. https://learn.microsoft.com/en-us/viva/insights/org-team-insights/copilot-dashboard
- "Copilot Analytics introduction." Microsoft Learn, accessed October 16, 2026. https://learn.microsoft.com/en-us/viva/insights/copilot-analytics-introduction
- "AI Adoption Dashboard for Enterprise Teams." Worklytics, accessed October 16, 2026. https://www.worklytics.co/measureai
- "Worklytics: Workplace Analytics for IT, HR & Leaders." Worklytics, accessed October 16, 2026. https://www.worklytics.co/
- "Worklytics Pricing." Worklytics, accessed October 16, 2026. https://www.worklytics.co/pricing
- "AI Adoption Dashboard: Track AI Usage & Cost by Team & Role." Larridin, accessed October 16, 2026. https://larridin.com/ai-adoption/
- "Scout: Measure AI Usage, Skills, and Impact." Larridin, accessed October 16, 2026. https://larridin.com/scout
- "Frequently Asked Questions." Larridin, accessed October 16, 2026. https://larridin.com/faq
- "AI Impact: Turn AI investments into measurable results." ActivTrak, accessed October 16, 2026. https://www.activtrak.com/solutions/ai-insights/
- "Measuring What Matters: The Missing Link in AI Adoption." ActivTrak, April 29, 2026. https://www.activtrak.com/customers/b2b-tech-ai-adoption-measurement/
- "ActivTrak Pricing." ActivTrak, accessed October 16, 2026. https://www.activtrak.com/pricing
- "Measure AI Coding ROI and AI Impact Dashboard." Jellyfish, accessed October 16, 2026. https://jellyfish.co/platform/jellyfish-ai-impact/
- "Pricing." Jellyfish, accessed October 16, 2026. https://jellyfish.co/pricing
- "The AI Management Platform for credit burn." Torii, accessed October 16, 2026. https://www.toriihq.com/products/ai-management-platform
- Uri Nativ. "AI dashboards." Torii Support, June 11, 2026. https://support.toriihq.com/hc/en-us/articles/51567354723099-AI-dashboards
- Chris Shuptrine. "5 Tools for Tracking AI Adoption Across Your Company in 2026." Torii, June 11, 2026. https://www.toriihq.com/articles/six-tools-for-tracking-ai-adoption-across-your-company
- "Pricing." Torii, accessed October 16, 2026. https://www.toriihq.com/pricing
- "AI Security & Governance Platform." Nudge Security, accessed October 16, 2026. https://www.nudgesecurity.com/use-cases/ai-security
- "Nudge Security Plans & Pricing." Nudge Security, accessed October 16, 2026. https://www.nudgesecurity.com/pricing
- "The system for AI enablement teams." Nolli, accessed October 16, 2026. https://withnolli.com/solutions/ai-enablement
- "Trust and governance." Nolli, accessed October 16, 2026. https://withnolli.com/security