Six Questions to Ask Your Agent About Your Product

A dashboard shows every chart and leaves you to find the one that matters. If your agent can read your analytics, you can start from the question you already have. These six come up for games, sites and SaaS products. Each has a prompt to copy and the kind of answer to expect.

Replace my-app with your project name. Every example answer below is from a demo project, with round numbers made up for illustration.

Two things apply to all six:

• Your agent uses the event names your project already sends. Ask Which events does my-app track? first, and swap in your own names.
• Free accounts get core setup and basic reads. Funnels, retention, pages and experiments need Pro, and your agent will tell you when an analysis is outside your plan.

1. Where do people drop off between two steps?

Build a funnel for my-app from page_view to signup_started to signup, for the last 30 days. Show the count at each step and where the biggest loss is. Say whether it is the largest absolute loss, the largest relative loss, or both.

Example from a demo recipe app:

page_view        2,000 users
signup_started     200 users   (10% of previous)
signup              80 users   (40% of previous)
Biggest absolute loss: page_view to signup_started (1,800 users)
Biggest relative loss: page_view to signup_started (90%)

A funnel shows where people leave. It does not show why. The next step is to look at the page the loss starts from, which is question 4.

2. Which release changed activation?

This question needs dates. Agent Analytics project context can hold short annotations for major changes, such as a pricing change, an onboarding change or a release. Ask your agent to add one each time you ship something that could move a number, and see CLI for the format.

Read the project context for my-app. For each annotated change in the last 60 days, compare the activation event in the 14 days before with the 14 days after. Show both counts, say how many users each period had, and tell me which changes line up with a move. Do not call anything causal if two changes fall in the same window.

Example from a demo puzzle game:

Activation event: first_level_complete
Mar 3, new tutorial:   310 users in the 14 days before, 380 after (+23%)
Mar 31, price change:  370 before, 350 after (-5%), too small to call
The two changes are four weeks apart, so their windows do not overlap.

A before and after comparison can show a move. It cannot prove the change caused it. To test whether one change caused a move, run it as an experiment: see AI Agent Experiment Tracking.

3. What do new users do after they sign up?

For people who signed up in the last 30 days, run a weekly retention cohort on signup. Then show which events they send in their first 7 days, ranked by how many of them sent it. Tell me what share reached the activation event.

Example from a demo invoicing app:

Signups, last 30 days: 400
Week 1 return: 45%, week 2: 30%, week 3: 25%
First 7 days, share of signups that sent:
  invoice_created   40%
  client_added      35%
  invoice_sent      15%
Reached activation (invoice_sent): 15%

If most people create something and few reach the event that means value, the gap sits between those two steps.

4. Which page sends signups?

Show session paths for my-app with signup as the goal. Which entry pages lead to a signup in the same session, and which entry pages get traffic but rarely convert? Give the one path most worth improving.

Example from a demo recipe app:

Entry page        Sessions   Reached signup   Rate
/recipes/pasta      1,000         30           3%
/                     800         16           2%
/pricing              200         20          10%
Most worth improving: /recipes/pasta (high traffic, low rate)

Paths are session-local. A goal only counts when it happens in the same session. If people read a page today and sign up next week, or sign up through an agent that never opens a browser page, paths will not link the two. Your agent should say so when the numbers are low.

5. Where does mobile break?

The tracker records a coarse device type and screen size on every page view. After you run a funnel, break down the largest leak by dimensions that exist in the data, including device. Turn on click and error tracking to see broken controls: data-track-clicks="true" records clicks and emits a rage_click when 3 clicks hit the same control within a second, and data-track-errors="true" records JavaScript errors as $error. Both are documented in Tracker.js. The funnel-then-breakdown sequence is in Closed-loop growth analysis.

Compare mobile and desktop on my-app for the last 30 days. Run the signup funnel, then break down the largest leak by device, and also break down the signup event by device. Show where mobile loses more people. Then list the controls with the most rage_click events and the most $error events on mobile, by path.

Example from a demo puzzle game:

Signup funnel, last 30 days
page_view        2,000 users
signup_started     200 users
signup              80 users
Biggest leak: page_view to signup_started (1,800 users, 90%)

Breakdown of page_view by device:
Desktop  1,000
Mobile   1,000

Breakdown of signup by device:
Desktop    120
Mobile      40

Mobile rage_click, top control: /play #start-button, 25 events
Mobile $error, top path: /play, 18 events

If signups are much lower on mobile and one control collects the rage clicks, open that page on a phone and tap that control yourself before changing anything.

6. Which source brings people who activate?

Traffic by source answers who visits. This question answers who ends up using the product.

Which sources and UTM campaigns bring visitors who complete my activation event, not just visitors? Show visitors, activated users and the activation rate for each source, last 30 days. Ignore sources with fewer than 20 visitors.

Example from a demo invoicing app:

Source           Visitors   Activated   Rate
google.com          1,000        50      5%
newsletter            200        30     15%
reddit.com            400         8      2%
Sources under 20 visitors left out.

A source with few visitors and a high rate is a lead, not a result. Check it again with more traffic before you spend on it.

Keep a question running

Grok Bot can run a saved prompt on a recurring schedule. That routine lives in Grok Bot, not in Agent Analytics. See Ask Grok Bot how your site is doing. A weekly version of question 1:

Every Monday at 9:00, run the signup funnel for my-app for the last 7 days and compare it with the 7 days before. Report the step with the biggest change and both counts. If no step moved by more than 20%, say so in one line.

Docs

• Ask Grok Bot how your site is doing: connecting and asking in chat, and recurring scheduled prompts.
• Closed-loop growth analysis: the funnel, paths and breakdown sequence behind these prompts.
• Session paths
• Tracker.js
• CLI: project context and annotations.