Practical insights

A useful analysis makes its calculation clear.

A convincing chart can be based on the wrong column. That is why an AI analysis starts with a clear data set and ends with an independent check of the decisive numbers.

LOL.MARKETING / CONNECTING INTELLIGENCE

The essential answer

Use Gemini for analytical suggestions, formulas and drafts. Record metric definitions and control calculations separately. AI-based text classification and reproducible financial or quantity calculations need different checks.

Define columns and metrics first

A monthly report needs a clear period for every row and a fixed meaning for every metric. Decide whether revenue is net or gross, how cancellations are handled and which status represents a won order. Without these rules, even a formally correct formula can answer the wrong question. Keep an unchanged raw-data snapshot as a reference.

Combine model assistance with verifiable spreadsheet logic

Gemini in Sheets can assist with formulas, charts and analysis; the AI function supports tasks such as text summarisation and categorisation. Google’s September 2 release also lists Gemini 3.8 Flash in Sheets for Google AI Pro and Ultra access. This does not establish identical enablement for every Workspace edition. Check the actual surface. Keep model-generated categories in separate columns with review markers, while approved metrics retain traceable calculation logic.

Example: analyse incoming enquiries

A business exports anonymised enquiry descriptions with arrival dates and processing status. AI suggests topic categories. The responsible team checks a sample and every ambiguous case, then stabilises the categories. A regular formula counts enquiries by category and week. The report separates observed changes from possible explanations; three unusual cases do not establish a reliable trend.

Find errors through independent checks

Compare overall totals with the export and calculate at least one selected group independently. Test empty cells, duplicate records, negative values and month boundaries. Also check whether filters hide relevant rows. For AI categories, examine confusions as well as accuracy: two similar-sounding labels can have very different consequences for an operational decision.

Working template: the spreadsheet review brief

Specify the data range, period and metric definition explicitly. An analysis request might say: “Explain the weekly change in enquiry volume. Show the underlying totals and filters. Separate observed values from possible causes. Check whether duplicate IDs or missing dates distort the result.” Then add your independent control total. A suggested explanation remains a hypothesis until further data supports it.

  • Units: do not treat percentages and percentage points as interchangeable.
  • Comparison: use the same scope definition across all periods.

Hand over a report with clear ownership

Document the source, refresh time, definitions, known gaps and accountable person alongside the report. Repeat the control calculation when columns or categories change. The System Check assesses whether a spreadsheet is sufficient or whether data preparation and recurring reporting need a separate workflow. An attractive chart is part of the output, not evidence of quality.

Additional sources change the data review

MCP connectors can expose information from additional business systems. Before analysis, align period, currency, status definitions and refresh time. A fluent report must not treat partial retrieval as a complete dataset. Custom instructions expanded in September help with recurring presentation preferences but do not establish a binding metric definition. Keep that definition alongside the data and control totals.

Keep it verifiable

Primary sources

The next sensible step

Assess your reporting workflow

Eight steps from a general interest in AI to a clearer decision for your business.

Start AI System Check
FreeProvider-neutralNo credentials