How to Analyze Funnels Using AI
Trace a defined account cohort from signup to activation to paid, then compare rates with AI using the metrics in your Cube deployment.
- Define the entry cohort and each conversion event
- Measure step rates with explicit windows and denominators
- Compare channel counts and rates without averaging them
What is conversion funnel analysis with AI?
Conversion funnel analysis measures how many records in a defined starting population reach each later step within fixed time windows. Product, growth, marketing, and revenue teams use it to see where progression changes. Keep one entity grain across stages, or document how records such as leads, accounts, and opportunities map to one another.
An AI agent can query available governed metrics, compare step and end-to-end conversion by segment, and summarize observed differences. This is a historical diagnostic, not a causal explanation or forecast; a funnel only shows the events and definitions represented in the model.
Define the funnel before comparing it
Choose one entity, the entry event, the ordered stages, and the full observation window. The examples below use fictional account data.
| Metric | Illustrative example |
|---|---|
| Entry cohort | 1,000 unique accounts with a first signup from Jan 1 through Jan 31, 2026 UTC. |
| Activation | 620 accounts had a first qualifying activation within 14 days of signup: 620 / 1,000 = 62.0%. |
| Paid conversion | 186 activated accounts had a first successful payment within 30 days of activation: 186 / 620 = 30.0%; 186 / 1,000 = 18.6% end to end. |
| Signup channel | Organic: 400 / 280 / 98 signups, activated accounts, and paid accounts; paid search: 350 / 210 / 53; partner: 250 / 130 / 35. |
| Complete follow-up | Event data is complete through 2026-03-17 00:00 UTC, exclusive, so the latest eligible account has the full activation and payment windows. |
Copy. Paste. Run your funnel review.
Connect to Cube via MCP or our CLI, then run the prompt below in your preferred agent.
Start with a Cube deployment connected to your data and a model containing the metrics you want to analyze.
Connect over MCP
Connect your agent to your existing Cube deployment and sign in through OAuth.
https://cubecloud.dev/mcp
Complete OAuth in your agent to connect your Cube account.
Use your default deployment or select another one available to your account.
Connect with CLI
Install the Cube CLI and sign in from a terminal-capable agent.
curl -fsSL https://raw.githubusercontent.com/cube-js/cube/master/install-cli.sh | sh
cube login --url https://TENANT.cubecloud.dev
cube spec --json
02Run this prompt in your agent
Choose your agent and copy the prompt below.
Use the Cube MCP server connected in Codex and select the intended deployment. Run read-only analytics queries. Review the historical signup-to-activation-to-paid funnel in my connected Cube deployment.Use read-only requests. Do not expose credentials or customer identifiers. DEFINITIONSInspect the model for the approved account grain, signup event, activation event,first successful payment event, signup-channel definition, time zone, and exclusions.Report each event's metric name and numerator/denominator rules. Do not infer abusiness definition from a field name. If the model lacks a required definition,explain the gap and ask before substituting another measure. COHORT AND WINDOWS1. Include one unique account whose earliest signup timestamp is at or after 2026-01-01T00:00:00Z and before 2026-02-01T00:00:00Z.2. Count an account as activated when its first qualifying activation timestamp is at or after signup and before signup + 14 x 24 hours.3. Count an activated account as paid when its first qualifying successful payment is at or after activation and before activation + 30 x 24 hours.4. Confirm source data is complete through 2026-03-17T00:00:00Z exclusive. If an account lacks full follow-up, do not treat it as a non-conversion; withhold the affected rate or report a fully observed sub-cohort with its exact dates. ANALYSIS1. Count each account once at each stage, using its first qualifying event. Show the entry count, activated count, paid count, and exact time windows.2. Calculate signup-to-activation as activated / signups, activation-to-paid as paid / activated, and signup-to-paid as paid / signups. Show each numerator and denominator. Recalculate overall rates from total counts; never average segment percentages.3. Compare those counts and rates by signup channel as recorded at first signup. Keep an Unknown channel bucket if the value is missing.4. Report invalid account IDs, missing event timestamps, duplicate mappings, excluded records, and unknown channels. Count an absent stage event as a non-conversion only when its source is complete for that account's full observation window. OUTPUTReturn the model definitions, cohort dates, source-completeness check, anoverall table and a channel table with numerator, denominator, count, and rate.Include the filters and time zone used. Separate observed differences fromhypotheses about causes. Do not claim causal impact or predict future conversion.
How it works
Lock the entity and stage definitions
Confirm the account grain, first signup, activation event, successful payment event, channel attribution, and exclusions in the governed model. If a sales funnel uses leads or opportunities instead, define that grain separately and do not mix it with account counts.
Calculate each step with its own window
Use the same eligible signup cohort, a 14-day activation window, and a 30-day payment window after activation. Show numerator and denominator for every step and the end-to-end rate; recompute totals from counts rather than averaging channel rates.
Review differences, then investigate
Compare both counts and rates by signup channel. A lower step rate identifies where to inspect further, but does not explain why it changed; check event quality, attribution rules, and supporting product or sales data before proposing a cause.
A conversion funnel, by stage
The image and worked example use fictional January 2026 values, not executed queries, customer results, benchmarks, or predictions.
| Signup channel | Signups | Activated (of signups) | Paid (of activated) | Paid (of signups) |
|---|---|---|---|---|
| Organic | 400 | 280 (70.0%) | 98 (35.0%) | 24.5% |
| Paid search | 350 | 210 (60.0%) | 53 (25.2%) | 15.1% |
| Partner | 250 | 130 (52.0%) | 35 (26.9%) | 14.0% |
| Total | 1,000 | 620 (62.0%) | 186 (30.0%) | 18.6% |
- Overall signup-to-activation is 620 / 1,000 = 62.0%; activation-to-paid is 186 / 620 = 30.0%; and signup-to-paid is 186 / 1,000 = 18.6%. The total rates use summed counts, not an average of channel rates.
- Organic has the highest activation rate (280 / 400 = 70.0%) and activation-to-paid rate (98 / 280 = 35.0%). It also has 98 paid accounts, compared with 53 for paid search and 35 for partner.
- Partner's activation rate is 52.0%, while 35 / 130 = 26.9% of its activated accounts paid within the window. These descriptive differences do not identify a cause or establish that a channel performs better in a different cohort.
This hand-calculated example and its image use fictional values, not executed queries, customer results, benchmarks, or predictions. Actual rates depend on event definitions, account identity mapping, attribution, and complete source data; a funnel does not show why a step changed.