Skip to content
Try free

AI

AI for the question. Mammoth for the answer that runs every week.

Plenty of tools will give you a one-off answer from a prompt. The useful version is different: the prompt should produce a pipeline, not a picture. Something that is still right next month without anyone rebuilding it.
Try freeSee the engine

A one-off AI answer

  • Correct on the day you asked
  • Steps taken are not recorded
  • Nothing re-runs when data changes
  • Next month, somebody asks again

An AI-built Mammoth pipeline

  • Correct today, and next month
  • Every step recorded and editable
  • Re-runs on a schedule or a trigger
  • Nobody asks again, it's already there

Scope

What the AI does

Specifically, and only. It writes the pipeline and builds the dashboard; deterministic steps compute the numbers.

Reads the question

Resolves plain language against your real columns, and asks when a term is ambiguous rather than guessing silently.

Profiles the data

Spots mixed date formats, inconsistent categories, duplicate keys and likely join columns on arrival, then proposes fixes.

Proposes the pipeline

Suggests the preparation steps needed, as editable steps in the canvas. Not an opaque transformation you can't inspect.

Builds the dashboard

Chooses the metrics and chart forms that answer the question, applies your theme, and publishes it.

Builds connectors

Point it at a REST API and it builds the connector for you: it reads the first page, works out the shape of the data and proves it can read real records before anything enters a pipeline.

Explains itself

Every action is listed in plain language with the row counts it touched, so you can check the reasoning rather than trust it.

Ask

Plain language, resolved against real columns

You ask for “weekly sales by region against last year.” Mammoth works out which table holds orders, which column is the date, how region is spelled in your data, and what “last year” means given your fiscal calendar. Then it tells you what it decided.

  • No SQL, DAX or modelling layer required
  • Follow-up questions refine the same dashboard
  • Ambiguity surfaces as a question, not a silent guess
  • You can override any decision it made
Ask MammothAI
Show weekly sales by region against last year, and flag any region more than 10% behind.

Built it. Four metrics, three charts, one flag rule.

  • Joined orders to store_master
  • Derived prior-year comparison
  • Flag rule: variance < −10%
  • Scheduled daily at 06:00

Inspect

Nothing is a black box

The output of a prompt is a list of steps, not a finished artefact you have to trust. Each step shows what it did and how many rows it affected, and you can edit or delete any of them.

  • Every transform listed in plain language
  • Row counts at each step, so nothing disappears quietly
  • Swap any AI step for your own SQL
  • Draft mode: check it before anyone else sees it
PipelineRe-runs daily
  1. 1orders.csv
  2. 2Drop duplicates
  3. 3Fill blanks: region
  4. 4Standardise dates
  5. 5Join store_master
  6. 6Weekly Sales

Durability

Then it runs without you

This is the difference that matters. The pipeline the AI wrote goes on a schedule, so next Monday's dashboard is built from next Monday's data. No prompt, no rebuild, no drift between versions.

  • Scheduled or event-driven refresh
  • Alerts when a source changes shape
  • Checkpoints so you can prove what a figure was
  • One shared pipeline instead of three private ones
RefreshAutomated
  • Weekly SalesRan 4m ago
  • Store MarginRan 12m ago
  • Close PackWaiting
  • Cash ForecastRan 1h ago

Data handling

Your data isn't training data.

A fair question to ask any vendor putting AI near your numbers. Your data is used to answer your questions and run your pipelines. Nothing else.

Your data is never used to train models
AI features can be disabled entirely on Enterprise
Everything AI does can be done by hand instead
Encryption in transit and at rest, private cloud available

The sceptical questions

Is my data used to train models?
No. Your data is used to answer your questions and run your pipelines. It is not used as training data.
What stops the AI from quietly getting it wrong?
Two things. First, it never produces an opaque result. Every step it takes is listed, editable and re-runnable, so a wrong assumption is visible rather than buried. Second, the number on the dashboard comes from the pipeline, not from the model. The AI writes the pipeline, then deterministic steps compute the figures.
What are the AI usage limits?
Generations are metered by plan: 5 a month on Free, 25 on Starter, 100 on Pro, and unlimited on Enterprise. There is no overage billing on any plan — no surprise line item for going over, because going over is not a thing you can be charged for. If you need more, you move up a tier.
Can I turn AI features off?
Yes. Everything AI does can be done by hand, with no-code steps or raw SQL, and Enterprise workspaces can disable AI features entirely if policy requires it.
Which model do you use?
We use frontier hosted models and change them as better ones ship. The important architectural point is that the model writes the pipeline; it doesn't compute your numbers.

Keep reading

DashboardsWhat the prompt produces.Data EngineThe deterministic part underneath.vs Excel + AIA brilliant answer once vs every week.

Ask it something hard.

Bring data you'd normally hand to an analyst, and see what the pipeline it writes looks like.

Try freeBook a demo
  • 21-day Pro trial
  • No credit card
  • Viewers always free