How is agentic analytics different from traditional BI?

Traditional BI starts with a human building a report and interpreting it. Agentic analytics gives an agent controlled access to the analytics workflow, while people keep approval over credentials and product changes.

Can a coding agent create a product analytics funnel?

Yes. With an authorized OnRamp CLI session, a coding agent can create a project, configure a funnel from tracked milestones, create segments, and read analytics. Deleting analytics configuration stays in the dashboard.

← OnRamp Analytics

Agentic analytics for product teams

Agentic analytics for product teams and coding agents

Q: What is agentic analytics?

Agentic analytics lets an AI agent do bounded analytics work: inspect evidence, make recommendations, and take approved configuration actions. For product teams, that can mean instrumenting milestones, defining funnels, and checking conversion results from a terminal.

OnRamp is an agentic analytics platform for instrumenting meaningful product milestones, defining conversion funnels, and investigating the evidence behind acquisition and conversion drop-off. Your team keeps control of authorization, customer data, spend and destructive changes.

Start free Read the CLI guide

1. Instrument

An agent reads your codebase and adds explicit step() calls at real product moments, such as account creation or first value.

2. Configure

With an organization-approved CLI session, it can create a project, turn those steps into a funnel, and add useful segments.

3. Investigate

It can read conversion, retention, breakdown, and exit-path data to make a grounded next-step recommendation.

What is agentic analytics?

Agentic analytics is analytics work performed by an AI agent with clear boundaries. Instead of only answering a question about a dashboard, the agent can use approved tools to complete parts of the workflow: understand the product, add the events needed to measure it, configure the report, and inspect the result.

For product analytics, the useful boundary is not broad access to every data system. It is a narrow, auditable set of actions around the customer journey. OnRamp's CLI is built for that workflow: it can create projects, manage funnels and segments, manage ingestion keys, and read analytics. It has no delete command; destructive changes remain in the dashboard behind a typed confirmation.

Agentic analytics vs. traditional BI tools

Traditional BI tools are designed around an analyst creating reports over broad company data. They are useful for that job. Product teams often have a tighter question: where do people abandon onboarding, what changed by app version or acquisition source, and what should we fix next?

What an agentic product analytics workflow should do

Agentic product analytics is not a chatbot that invents a dashboard from a vague prompt. It is a bounded workflow where an agent can connect product context, instrumentation, funnel configuration, and evidence-based investigation. The agent should be able to explain what it observed, which cohort or step supports its recommendation, and which action still needs a person's approval.

Instrumentation

Read the product flow and add stable events at real moments of progress, not every tap or screen view.

Investigation

Compare funnel conversion by app version, platform, source, and retention to find the highest-impact problem.

Guardrails

Use scoped tokens, auditable commands, and human approval for product or destructive changes.

OnRamp focuses an agent on those product questions. The agent works from named milestones and funnels rather than guessing from raw logs, and the results stay tied to real users, versions, platforms, referrers, retention, and tracked navigation paths. That makes the recommendation inspectable instead of a generic AI summary.

A practical agentic analytics workflow

Ask the agent to find three to five meaningful onboarding moments in your codebase.
Have it add OnRamp SDK calls at those moments and propose the funnel before it changes code.
Authorize the CLI in your browser, then let the agent create or update the funnel for the new step names.
After traffic arrives, have it inspect conversion and breakdowns, then recommend one measurable improvement.

An acquisition-to-revenue investigation

Ask: “Compare Apple Search Ads keywords and Google Ads campaigns by onboarding completion and paid conversion. Find cheap installs with weak activation, identify the largest drop-off step, and propose one experiment. Do not change spend.”

OnRamp keeps campaign context attached to the visitor journey while the agent inspects funnel, retention and revenue evidence. See mobile ad attribution for the underlying measurement flow.

The product onboarding analytics workflow remains useful even when an agent performs the setup: event names need to represent real user progress, and a human should decide which product change to ship.

For the full implementation pattern, see our guide to agentic product analytics. It covers a concrete event plan, the questions an agent should investigate, and the permissions it should never receive.

Use a coding agent without giving up control

Browser login grants the CLI access to your organization rather than exposing a password to the agent. You can also mint a scoped token from Team, with project access and read/write permissions set at creation. The CLI returns JSON by default, which lets an agent use the result safely in the next command. Access can be revoked from Team at any time.

Try agentic analytics on your onboarding

Create an OnRamp project, copy the setup prompt into your coding agent, and turn your first real milestones into a funnel.

Add OnRamp analytics to this app.

1. Install the right @onramp-sdk package for our stack
   (see https://getonramp.dev/docs/getting-started) and
   initialize it once at startup.
2. Find 3-5 real onboarding milestones in the codebase -
   signup, first key action, activation - not every screen
   or tap. Propose the list before you add any calls.
3. Add OnRamp.step() calls at those moments.
4. Install the OnRamp CLI (npx @onramp-sdk/cli), run
   `onramp login` to authorize it against my account,
   create a new app, and create a funnel matching the
   milestones you just instrumented.
5. Show me the funnel URL when you're done.

Create a project CLI commands