Find where your signup journey loses people.

Write down the journey you want to inspect and what each step means. In this example, a landing-page visitor starts an account flow and later completes it. Distinguish tracked identities from event occurrences and keep the entry period and ordered journey consistent. Counts follow the identities recorded in your project; inconsistent identification can split one person's journey.

PostHog's funnel documentation explains conversion steps and configurable funnel behavior. Use those definitions to agree how your query should work, including the period allowed to complete the journey. A total of all page views divided by all signup events is a different calculation and should not be presented as this cohort funnel. [1]

In Integrations, open Your data, then the PostHog connection under API credentials. Enter the project ID, data region, and a personal API key with the read permissions requested by Taploop. Put credentials in the connection form, not in an agent prompt. In Events Taploop may read, list the real event names needed for the journey.

Choose the Landing event and, where appropriate, the Outcome event, then use Verify and connect. The default page-view event alone cannot establish a multi-step signup funnel. Taploop's supported funnel reads use two to five distinct allowed events and a period of at most 93 days. If tracking is missing, ask the implementation owner to address it before claiming a complete funnel.

Use AI setup → Set up access to my integrations and open Agent access. Select the correct PostHog account and the available traffic or funnel read capabilities. Follow the server address and secure token instructions shown in that panel. Allow results in the agent's client if your workflow requires it, and ask the agent to confirm which read capabilities it can use.

Then request the agreed funnel using a bounded date range. The result should identify the connection, events, period, counting method, and any limits. If the available capability returns event totals without the person-level ordering needed for your definition, ask for that limitation to be stated. Do not silently relabel independent event totals as a verified conversion funnel.

In the fictional cohort, 200 of 1,000 visitors start signup, so the visitor-to-start rate is 20%. Of those 200 starters, 120 finish, making the start-to-completion rate 60%. The overall visitor-to-completion rate is 120 divided by 1,000, or 12%. These three percentages answer different questions.

The first transition loses 800 people, or 80% of visitors. The second loses 80 people, or 40% of starters. Do not add the percentages together: their denominators differ. Ask the agent to show the calculation beside each rate. If your real result has missing data or incompatible counting, leave the rate unresolved instead of making the arithmetic look more certain than the evidence.

The example identifies the largest numerical drop before signup starts. It does not tell you why people leave. A visitor may be researching, the offer may be unclear, the button may be hard to find, traffic may be irrelevant, or an event may be missing. These are alternative explanations to investigate, not conclusions established by the counts.

Ask for a short investigation order: check event recording, inspect the landing-page action, and review relevant customer feedback. Breakdowns by source or campaign can be useful when those properties were actually collected and the groups are large enough to interpret. PostHog also notes that incomplete conversion periods can distort recent trends, so compare periods on the same basis. [1]

For the teaching example, the next action is to verify the landing-to-start measurement and inspect whether the page makes the signup action understandable. If that check identifies a clear issue, your team can propose one focused change. Record the hypothesis, owner, primary measure, and what else must stay consistent for the comparison to be useful.

Save the query definition and evidence with the summary so a teammate can reproduce the baseline. Taploop's Runs and evidence can help you review recorded activity. A later result must be compared using the same definitions; changes in traffic or tracking may explain differences. End with an investigation or experiment decision, rather than a claim that the cause has already been found.