Eleven people, three weeks, one rule: nothing was allowed to be a demo. Every prompt had to touch a real client deliverable, a real invoice chase, or an inbox that needed clearing by Friday. We logged minutes in a shared sheet before and after, which was tedious and occasionally inaccurate, but it beat guessing.
What follows is the unglamorous version of Gemini for Workspace, the one without the polished launch screenshots. It saved us somewhere between six and nine hours a week across the team, and it also cost us two awkward corrections with a client. Both halves matter.
The setup: 11 seats and a fixed budget
We put the whole studio on the Gemini Business plan, which at the time of writing runs about $24 per user each month on an annual commitment, or $30 if you want the freedom to walk away. Eleven seats works out at roughly $264 a month. That is not nothing for a small agency, and it is the number we kept returning to whenever we argued about renewing.
Rollout took an afternoon, mostly because half the team had already been using the standalone Gemini app and had opinions. The admin console is straightforward: assign licences by group, flip the toggle, done. The friction sat elsewhere, in unlearning the habit of copy-pasting into a browser tab and instead working inside the panel that was already open.
If you are weighing this against the enterprise tier of a competing product, the maths gets murkier fast. ChatGPT Enterprise pricing is quoted per company and lands in a very different bracket, as this breakdown of what the custom price actually gets you makes clear. We did not need that bracket. We needed Gmail to stop eating our mornings.
Where it genuinely earned its keep
Gmail is the strongest piece of the whole thing
The side panel inside Gmail is where most of our saved time came from. Not the “help me write” button, which produces the same slightly over-polished email every time, but the summarise-and-reply loop on long threads. One client thread ran to 34 messages across nine days. Gemini condensed it into six bullets and flagged the two open decisions. I read it in 40 seconds and replied without scrolling once.
Small thing, big compounding effect. Three people on the team estimated they got 20 to 30 minutes back per day just from triaging threads they would otherwise have skimmed badly.
Docs handles the blank page, then gets in the way
Asking it to draft a project brief from four bullet points works. The output is generic but structurally sound, and rewriting a mediocre draft is far faster than starting cold. What it does badly is the second pass, where you need a specific voice, an inside reference, or an actual opinion. Every time we asked it to make something punchier, we got conference-brochure copy back.
Use it for scaffolding. Do the personality yourself.
Sheets quietly won the entire test
The least hyped surface turned out to be the most useful. We had a messy export from a project tracker, roughly 900 rows, inconsistent date formats, three spellings of the same client name. Gemini in Sheets suggested a column formula to normalise the dates, then a second one to collapse the near-duplicates. Four minutes instead of the hour I had budgeted for it.
That is unsexy work. It is also exactly the work a small team drowns in.
Where it wobbled
Meeting notes were a coin flip
“Take notes for me” in Meet transcribed fine. The summary was the problem. In a 45-minute call about a website relaunch it captured the timeline correctly and completely missed the thing everyone actually argued about, which was who owned the copy. In another call with heavy cross-talk it attributed a budget figure to the wrong person.
We now treat auto-notes as a rough transcript rather than a record. Somebody still writes the decisions down by hand.
Confident, specific and wrong
The failure mode that matters is not a blank answer. It is a fluent answer with a fabricated detail buried inside it. Over three weeks it invented a job title in a client bio, guessed at a competitor’s pricing and presented the guess as fact, and summarised a policy document using wording from an older version of the same policy.
Every one of those was caught by a human read. None were caught by the tool. Whatever you build around this, budget for the reading.
The admin side nobody puts in the demo
Data governance was the question our clients asked, and the answers are reasonable but need saying out loud. Content you put into Gemini for Workspace stays inside your Workspace tenant and is not used to train the underlying models. Admin controls let you switch the feature off per organisational unit, which matters if you have contractors or a design team handling embargoed material.
Worth remembering that even very well-resourced AI companies have had their security incidents, as this account of OpenAI going from hacker to hacked shows. Your Workspace tenant is one perimeter. Your team’s habits are another, and the second one has no admin toggle.
Running the numbers
- Estimated time saved across 11 people: 6 to 9 hours a week, concentrated in Gmail and Sheets.
- Monthly cost: about $264 on the annual Business plan.
- Time lost to checking, correcting and rewriting outputs: roughly 2 hours a week.
- Net result: modest, real, and heavily dependent on how much email your team handles.
If your people live in Gmail, the maths works. If the work is mostly long-form writing or client calls, the return is thinner and you may do better with a dedicated assistant. Our own honest comparison of the AI chatbots worth your time is a sensible starting point if you are still shopping, and it is fair to say that the best AI chatbot for a given team is often not the one with the loudest launch.
What we changed after three weeks
We kept Gmail and Sheets switched on for everyone. We turned off auto-generated meeting notes for client calls and left them on for internal standups, where the cost of a misattributed sentence is close to zero. We added one line to the onboarding doc: never send an output you have not read end to end.
We also stopped asking it for opinions. Suggestions on phrasing, yes. Judgement calls on strategy, no. The tool is genuinely strong at the mechanical layer of office work, the summarising, reformatting, and pulling structure out of mess, and noticeably weaker the moment a task requires taste.
Three weeks is not a verdict. It is enough to know the licence is worth renewing for the four people who write 40 emails a day, and worth revisiting in six months for everyone else. The renewal date is already on the calendar, along with a note to re-run the same time-log test before we sign anything.