A team workspace that does its AI work on your device
Meeting notes, filing pages and reading a scan run right here. Nothing is sent.
A team workspace app with about 100 million usersArchetype: team workspace, about 100 million users (reported, rounded)
Saves up to £23.3 million a year
in server work moved onto its users’ own laptops and phones.of per-user AI work moved from servers to the users’ own devices.
At public list prices, with 9 in 10 of its users’ devices doing the work.At public list prices: units a year x list price x device share 0.9, rounded down.
- 8jobs can run on users’ devices
- 6jobs stay on servers
- This visit: nothing yet. Run the jobs below.This visit: nothing measured yet. Run the jobs below.
About 1 million people pay for its AI features (worked out from public figures, rounded down). We assume each records two half-hour meetings a week for 48 weeks, makes 10 quick AI requests a working day, and uploads about 960 scanned pages a year. 9 in 10 of their devices can do these jobs. Each job is priced at the public list price of the outside service it replaces, checked October 2026; filing pages and summarising meetings are priced as the quick AI requests they replace, at a mid-size AI model’s list price. Only work on a person’s own meeting, page or upload, for that person, is counted. A company that runs its own models pays less than list price for this work.
At the lowest public prices and the smallest AI models: £5.69 million. With 3 in 4 devices able to do the work: £19.4 million; with half: £12.9 million.
Storage, sync, shared search, live editing between teammates, long AI writing and agents, and mail and calendar stay on servers and are not counted. Nightly exports also stay on servers, not priced here because their size is not public. The speech job was tested in English, Spanish, French, German and Portuguese.
- People paying for AI features
- 1,000,000: worked out from reported revenue, the published plan’s list price and the reported share of customers paying for AI (1.04 million, rounded down)
- Meetings
- 2 a week x 0.5 hours x 48 weeks = 48 audio hours each a year (our assumption). Priced as live speech to text, because this kind of product shows the notes as the meeting happens
- Quick AI requests
- 10 a working day x 250 days, each reading about 1,100 words and writing about 110 (our assumption). Filing and the meeting summary are priced as the quick AI requests they replace
- Search vectors
- About 1.4 million words of edited pages made searchable each a year (only changed pages, our assumption), counted at list price
- Scans
- 960 pages each a year (our assumption)
- Device share
- 9 in 10 (our assumption)
- Other prices and shares
- Lowest public prices and the smallest model class: £5.69 million; 3 in 4 devices: £19.4 million; half: £12.9 million
- Live speech to text
- £0.33 an audio hour (public list price USD 0.45; public range USD 0.15 to 1.02). At the highest public prices every line would be higher; the headline uses typical list prices because which speech service this kind of product uses is not public
- Mid-size AI model
- £1.50 per million tokens in, £7.54 out (public list prices USD 2 and 10; the smallest class, used for the lowest-price figure, lists at USD 0.10 and 0.50)
- Page reading
- £1.13 per 1,000 pages (public list price USD 1.50; public range USD 0.60 to 1.50)
- Embeddings
- £0.01 per million tokens (public list price USD 0.02)
- Languages
- The speech job was tested in English, Spanish, French, German and Portuguese
- Left out
- Storage and sync, the shared search index, live editing between teammates, long AI writing and agents, and mail and calendar stay on servers; nightly exports also stay on servers, not priced because their size is not public. A company that runs its own models pays less than list price for this work
- Pounds to the dollar
- 0.75491 (reference rate, 15 October 2026); prices checked 15 October 2026, re-checked by 15 January 2027
- Formula
- Units a year x public list price x device share, at 0.75491 pounds to the dollar, rounded down to 3 figures
- This visit
- The work done in this tab x the same list prices x 0.75491, only for jobs that ran on this device and matched their check
- Same jobs elsewhere
- The same jobs fit a legal document assistant, a workplace safety reporting app or a proposal writing tool