All agents

AI Data & Excel Analysis Agent

An autonomous data analyst that turns a raw CSV, Excel file, or your CRM into a deterministic executive report — trends, segments, anomalies, forecasts, and reconciliation — then answers questions about it in plain English.

View in marketplace Open dashboard

The AI Data & Excel Analysis Agent is a BI team in a single agent. You upload a CSV/Excel file or connect your CRM, and it auto-detects the schema, profiles and cleans the data, runs trend/segmentation/funnel/anomaly analysis, forecasts what's next, reconciles multiple files, and writes an executive report with charts and prioritized recommendations. Every number is computed by a deterministic engine — the AI reasons over and explains those figures but never invents them — and you can then ask questions about your data in plain English.

What it does

For founders, operators, finance and RevOps teams who live in spreadsheets but don't have a data team, this agent delivers BI-grade insight without dashboards, SQL, or Python. It ingests raw spreadsheets or CRM data, auto-detects column types, profiles data quality, and then runs the full analysis loop — trends over time, breakdowns by segment, funnels by stage, anomaly detection, forecasting with confidence bands, and cross-file reconciliation — before synthesizing an executive report with narrative insight and prioritized, action-worthy recommendations.

It operates as a senior data scientist plus business strategist in one autonomous agent (maxTurns 14). The defining guarantee is determinism: a separate analysis engine computes every figure, and the AI is instructed never to invent, estimate, or round numbers — it only reasons over and explains tool results, with report charts built verbatim from those results. After producing the report it can push the top action items into the CRM as tasks, and it always records a concise run summary. Day to day, users work in the Data workspace at /dashboard/analysis to upload files, trigger Analyze/Question/Reconcile, read and download reports, and chat with their data in plain English.

How it works

Your Business
1
Ingest
Upload a CSV/Excel file (≤10MB, ≤50k rows; larger files truncated to 50k) or call analysis_load_crm to pull CRM pipeline/contacts/outreach into a dataset. analysis_list_datasets finds the target (the configured default_dataset, or the most recent upload).
2
Profile & clean
analysis_profile inspects every column — type (number/date/category), nulls, distinct values, min/max/mean/median/stddev, top values — and identifies the date column and the numeric measures to analyze.
3
Analyze
analysis_timeseries computes the trend on the main measure; analysis_query breaks it down by segment/category and by stage (funnels, top-N, mix); analysis_anomalies flags unusual spikes/drops via z-score. All numbers come from the deterministic engine.
4
Forecast & reconcile
analysis_forecast projects the measure forward (default 3 periods) with a confidence band via linear (least-squares), moving-average, or seasonal-naive methods. For multi-file checks, analysis_reconcile matches two datasets on key column(s) + an optional amount column (matched / only-in-A / only-in-B / mismatches).
5
Synthesize & report
The agent reasons like a strategist (what / why / risks / recommendations) and calls analysis_make_report to save a titled executive report — one-line summary, markdown sections (Executive summary, Key findings, Drivers, Risks & opportunities, Recommendations) and trend/segment/forecast/funnel charts built verbatim from tool results.
6
Push & record
If autonomy is push_to_crm, it sends the top 1–2 action items to the CRM via crm_create_task. It always finishes by calling record_output exactly once with a concise summary and where to find the report (within its maxTurns of 14).
Outcomes delivered

Setting it up — owner / admin

  1. 1
    Subscribe & deploy
    Subscribe to the AI Data & Excel Analysis Agent, then deploy it from the dashboard. Its deployment/config page lives at /dashboard/agents/[id].
  2. 2
    Connect tools (optional)
    No connection is required to analyze uploaded files. Optionally connect the built-in CRM so the agent can pull pipeline/contacts/outreach into a dataset (analysis_load_crm) and create tasks (crm_create_task). Nothing else needs wiring.
  3. 3
    Fill the config fields
    Set Business context (textarea — what the business does and the metrics that matter; sharpens analysis and recommendations). Optionally set Default dataset id (leave blank to analyze the most recent upload). Choose Autonomy: 'Report only' or 'Also push key insights to the CRM as tasks'.
  4. 4
    Set the schedule (optional)
    It's on-demand by default. Use the Run frequency field (daily/weekly/monthly) to run recurring auto-reports on connected CRM data; otherwise trigger runs yourself from the workspace.
  5. 5
    Do the first run
    Open the Data workspace at /dashboard/analysis, upload a CSV/Excel file (or load CRM data), select the dataset, and hit Analyze for a full report — or type a question. To reconcile, select dataset A, pick dataset B and the key column (and optional amount column), and run Reconcile.

Using it day to day — your team

  1. 1
    Open the Data workspace
    Go to /dashboard/analysis. You'll see your datasets, a list of saved reports, and a chat panel.
  2. 2
    Add data
    Upload a CSV/Excel file (it confirms row count and warns if truncated to 50k), or load CRM data (pipeline/contacts/outreach) into a dataset, then Refresh.
  3. 3
    Run an analysis
    Select a dataset and click Analyze for a full executive report, or ask a one-off question in the Question/chat input ('why did revenue dip in March?'). To compare two files, select dataset A, pick dataset B and the key column, then Reconcile.
  4. 4
    Read the results
    Open the generated report to see the one-line summary, sections (Executive summary, Key findings, Drivers, Risks & opportunities, Recommendations) and charts; download it for sharing. A one-line summary also appears on the dashboard.
  5. 5
    Act on recommendations
    Work the prioritized recommendations. If the agent runs with push_to_crm autonomy, the top action items already appear as tasks in your CRM ready to assign.
  6. 6
    Keep asking
    Continue the conversation in the chat panel to drill into segments, top-N, anomalies, or forecasts — every answer is computed live from your data by the same read-only analysis tools, not guessed.

Use cases

Revenue & sales analysis
Upload sales data and get a monthly trend, revenue by segment, sales-rep and stage breakdowns, anomaly flags, and a next-quarter forecast with a confidence band — packaged as an executive report with charts.
Multi-file reconciliation
Drop in two files (e.g. bank statement vs ledger, CRM vs billing), pick the shared key column (and an amount column), and get matched / only-in-A / only-in-B / amount-mismatch results with totals and what to fix.
Ask your data anything
Chat with a dataset in plain English — 'top 10 customers by spend', 'which month was the outlier and why', 'forecast pipeline next quarter' — and get a direct answer backed by computed numbers and a short rationale.
Scheduled BI on CRM data
Connect the CRM and schedule recurring reports (daily/weekly/monthly) on pipeline/contacts/outreach, optionally pushing the top action items back into the CRM as tasks.

What to expect

  • Executive reports (one-line summary + markdown sections + trend/segment/forecast/funnel charts) saved to the Data workspace and viewable/downloadable
  • Plain-English Q&A over your data answered with computed evidence and a short 'why'
  • Multi-file reconciliation: matched / only-in-A / only-in-B / amount-mismatch breakdowns with totals (aSum/bSum/sumDiff)
  • Forecasts of any date + measure (linear, moving-average, or seasonal-naive) each with a confidence band
  • Top action items optionally pushed into the CRM as tasks (push_to_crm autonomy)
  • A one-line summary on the dashboard and a metered, auditable record_output for every run

Metrics to watch

  • Time-to-insight: a full executive report generated from a raw spreadsheet in minutes per run
  • Grounding: 100% of stated figures traceable to a tool result (no fabricated numbers) — the core guarantee
  • Forecast quality: actuals landing inside the projected confidence band over successive periods
  • Reconciliation hygiene: shrinking only-in-A / only-in-B / amount-mismatch counts and the sum difference across runs
  • Action follow-through: share of pushed CRM tasks (action items) actually worked and closed
  • Coverage: datasets profiled cleanly without hitting the 50k-row / 10MB truncation limit

FAQ

Does the AI make up numbers?
No. Every figure comes from the deterministic analysis engine behind tools like analysis_query, analysis_timeseries, analysis_forecast, analysis_anomalies and analysis_reconcile. The system prompt enforces an absolute rule: the agent may only reason over and explain tool results, never invent, estimate, or round figures from memory. Report charts pull their data arrays verbatim from those tool results.
What do I need to connect, and what file types work?
Nothing is required to start — upload a CSV or .xlsx file (up to 10MB / 50k rows; larger files are truncated to 50k). Optionally connect the built-in CRM so the agent can pull pipeline/contacts/outreach into a dataset with no file, and optionally enable push-to-CRM so it can create tasks from action items.
Can it forecast and reconcile multiple files?
Yes. analysis_forecast projects a measure forward (default 3 periods) using linear least-squares, moving-average, or seasonal-naive methods, each with a confidence band (lower/upper). analysis_reconcile cross-checks two datasets on shared key column(s) and an optional amount column, returning matched, only-in-A, only-in-B, and amount mismatches plus totals (aSum/bSum/sumDiff).
How do I ask questions about my data versus run a full report?
In the Data workspace, select a dataset and choose Analyze for a full executive report, or use the Question input / chat panel to ask a one-off question in plain English (e.g. 'why did revenue drop in March?'). Question mode answers directly with supporting numbers and a short 'why', and saves a brief report if a chart helps. Both run the same deterministic tools.
Is my data private?
Yes. Datasets and reports are org/tenant-scoped to your workspace (all tool access is org-scoped via ctx.orgId) and never shared across tenants.