Skip to content

Ask

Ask in plain English, get an answer that shows its work

A dashboard makes you assemble the answer yourself. Ask does the assembly, then shows you the parts so you can check it.

What is Ask?

Ask is Bastle's assistant for subscription revenue: you ask a question the way you would ask a colleague — why MRR moved, which cohort churns hardest, what a price change did to expansion — and it answers in a sentence, then links the customers, subscriptions and invoices it used to get there. It reads the same calculation engine the dashboards use, so its answers and your charts cannot disagree.

The question

The words you would actually use

No query language, no filter builder, no learning which of eleven charts holds the answer. Ask handles the questions people genuinely have about revenue — comparative, causal and open-ended — and asks a clarifying question when yours is ambiguous rather than guessing and sounding confident.

  • “Why did MRR drop last month?” — decomposed into movements, not restated as a chart
  • “Which plan has the worst 6-month retention?” — computed across cohorts
  • “Which annual renewals in the next 30 days are at risk?” — a list you can act on

Why did MRR drop last month?

MRR fell $4,180 (−3.1%) in March. Almost all of it is contraction, not cancellations: 31 Scale customers downgraded to Growth after the annual renewal window opened.

47 customers3 plansMRR movement ledger

The receipts

Answers you can audit

Every answer names the metric definitions it applied, the period it covered and the records it read, and each of those is a link. That matters more than it sounds: an unsourced answer from a language model is a guess with good grammar, and nobody should put one in a board deck. Ask is built so you can check it in about ten seconds.

  • Customers, subscriptions and invoices cited inline
  • The metric definitions used, stated explicitly
  • Follow-up questions keep the context, so you can drill without repeating yourself
bastle.co/demo/assistant
Open

This is the real product running on sample data. Click to use it.

Agents

Your tools can ask, too

Ask is not only a chat box. Bastle ships an MCP server that exposes the same capabilities as typed tools, so an agent in your editor, terminal or workflow can query revenue directly instead of scraping a dashboard. The tools are deliberately narrow and well-described, which is what makes them reliable in an agent loop.

  • Metric summaries, single-metric queries and metric explanations
  • Customer search, cohort retention and anomaly detection
  • Write tools for annotations, goals and data import

MRR

+5.2%

$162,400

Net revenue churn

−0.3pp

2.1%

Active customers

+38

1,284

ARPU

+1.4%

$126.48

Models

Model choice, with the bill in the open

Ask routes through OpenRouter, so you pick the model rather than inheriting whichever one we standardised on — Anthropic, OpenAI, Google, Moonshot and others are all available. Every call is written to a usage ledger with its token counts and cost, and checked against workspace budgets before it runs.

  • Choose the model per workspace; the registry syncs from OpenRouter rather than being hardcoded
  • Per-call cost recorded to an auditable usage ledger
  • Workspace budgets enforced before a call is made, not after the invoice

Churn definition

Editable
Delinquency window
21 days
Trials counted
Never
Refunds
Deducted in period
Downgrades
Contraction, not churn

Change any rule and every historical figure recalculates against it.

Capabilities

What Ask includes

All of it sits in the same workspace. Bastle is free while in beta, so nothing on this list is an add-on, an upgrade or a separate subscription.

Grounded in one engine

Ask reads the same metric engine the dashboards render, so an answer and a chart cannot quietly disagree with each other.

Segment-aware

Ask a question about a saved segment and the answer respects it, including custom attributes you imported yourself.

Says when it does not know

Where the data cannot support an answer — too few trials, too short a history — Ask says so instead of producing a confident number.

Available over the API

The same question-answering endpoint is callable from your own code, so an internal tool can embed it without a browser.

Related

Where this connects

Ask reads the same normalised billing history as the rest of Bastle, so a change in one place shows up in the others without an export.

Metrics

Most analytics tools hand you a number and ask you to trust it.

Learn more

API & MCP

Analytics tools that treat the API as an afterthought make you re-implement their metric definitions in your own code.

Learn more

Segmentation

A single company-wide churn rate is an average of several different businesses.

Learn more

Alerts

Finding out about a bad week when you open the dashboard on Monday is finding out four days late.

Learn more

Keep going: see everything Bastle does, look up the maths behind any number in the metric glossary, or run a figure through the free calculators.

Practical questions are usually answered by the connector list, an honest comparison with the alternatives, or the same product framed for founders and for finance teams.

Frequently asked questions

How is Ask different from asking ChatGPT about my metrics?

A general assistant has no access to your billing data, so it can only tell you how a metric is defined in general. Ask is connected to your workspace and computes against your actual subscriptions using your configured definitions, then cites the specific customers and invoices behind the answer. The difference is between an explanation of what churn means and a list of who churned and what it cost.

Does the model see my customer data?

Ask sends the model the data needed to answer the question you asked — typically aggregates and the specific records relevant to that question, scoped to your workspace. It is not a bulk upload of your customer list, and data from one workspace is never used to answer another workspace's questions. Full detail on retention and processing is on the security page.

Can Ask be wrong?

Any language model can be wrong, which is exactly why answers are built to be checked. The arithmetic is done by the metrics engine rather than by the model, so the numbers themselves come from the same code that renders your charts. What the model contributes is the interpretation and the phrasing, and every claim it makes links to the records it drew on, so a wrong reading is visible rather than buried.

Which models can I use?

Models are routed through OpenRouter, which covers Anthropic, OpenAI, Google, Moonshot and other providers, and the list is synced from the live registry rather than hardcoded. You choose the model per workspace and can change it whenever you like. Every call is recorded to a usage ledger with its cost.

What is the MCP server for?

MCP is the protocol that lets an AI agent call external tools. Bastle's MCP server exposes subscription analytics as typed tools — metric summaries, single-metric queries, cohort retention, customer search, anomaly detection, plus writes for annotations, goals and imports — so an agent in your editor or workflow can query revenue directly. It is included, not a separate product.

Does Ask cost extra?

No. Bastle is free while in beta and Ask is part of it, with no card required to start. Usage is metered internally against workspace budgets so that a runaway loop cannot quietly generate enormous cost, and you can see exactly what has been spent in the usage ledger.

See your own numbers in about two minutes.

Connect Stripe and Bastle backfills your full history. Free while in beta — no card, no sales call.

Start freeExplore the live demo