Check the numbers, commitments and records

1. Ask across the data, then reconcile the total

Google’s October 19 Connected Sheets announcement lets eligible users ask Gemini questions across a full BigQuery dataset, rather than only the rows visible in a spreadsheet. BigQuery is Google’s cloud data warehouse. Users need permission to query the connected data; rollout can take up to 15 days. Business Standard and Plus are among the eligible plans.

Illustrative scenario: a Canadian wholesaler already keeps order data in BigQuery. An operator wants to compare provincial sales without asking an analyst for another export. A familiar spreadsheet could make that exploration easier. The tradeoff is that a plausible answer can still use the wrong definition of sales.

This week, choose one completed month and one question whose answer you already know. Specify Canadian dollars, whether tax is included, how returns are treated and which date determines the month. Compare the result with your approved report before asking follow-up questions.

This is most relevant to firms already using the connected tools. It is not a reason to move a small spreadsheet into a warehouse. The announcement establishes a capability and rollout, not accuracy on your records or a proven saving. Ask your data owner about query costs and access before testing.

2. Translate the conversation and confirm the commitment

On October 22, Google announced speech translation across 63 languages in Meet, with automatic language detection and individual listening controls. Access depends on an eligible organiser or plan, with regional and age restrictions. Rapid Release started October 22; Scheduled Release begins November 2. Check your account rather than assuming access this week.

Illustrative scenario: a Canadian service business meets a supplier whose preferred language differs from the buyer’s. Translated speech could let more people ask questions directly. The tradeoff is that a fluent conversation can make a mistaken quantity, name or deadline harder to notice.

This week, if your account has access, try a short internal call using fictional order details in the languages your team actually uses. Ask each person to write down the quantity, currency and due date. Compare notes, then agree how real commitments will be confirmed in writing.

Google lists important limits: translation is unavailable in recordings and client-side encrypted meetings. Room hardware can receive translated speech but cannot translate speech originating in the room for others. That makes the meeting setup part of the test. Treat wider language coverage as a chance to improve participation; it does not establish reliable interpretation of your specialist vocabulary.

3. Decide which AI conversations become business records

Google’s October 22 Vault update adds filtered exports of Gemini conversations inside Workspace apps. Eligible administrators can select users and dates, with exports containing prompts, answers, attachments and source links. The scope covers conversations created with history turned on. Google says retention rules and litigation holds for this new source will follow later.

Illustrative scenario: an advisory firm wants to understand how a draft recommendation was prepared. An export could help reconstruct what the assistant saw and produced. It also creates another copy of potentially sensitive material that someone must control.

For a Canadian firm discussing customer information, the practical question is who may retrieve that history and where the export will go. Availability includes Business Plus and specified enterprise editions or the Vault add-on; do not assume every paid account includes it.

This week, ask your administrator to map one fictional drafting exercise: where history appears, who can export it and who owns deletion of downloaded copies. Review the process before using client material. Exportability alone does not prove a complete record, an effective retention policy or compliance with your obligations.

The bigger pattern

These changes bring AI closer to the work people already recognise: a report, a conversation and a record. Start with one known total, one written commitment or one traceable draft. Give someone responsibility for checking it. A useful trial should leave the next person better able to understand the work.

Which recurring task in your business would improve most if someone could check its numbers, commitments or source record in five minutes?