Amplitude

Your assistant can pull active and new user counts for any date range, by day, week, or month, and split them by a user property such as platform or country. There is a separate real-time count for today against yesterday when you want to know whether something is happening right now. This is the number people open Amplitude to look at, and asking for it in words is faster than rebuilding the chart.

What you end up with. Active and new user counts for the period you named, with the comparison period next to them and the split you asked for.

Start with it bringing you the information and you deciding.

Never without. Fix the date range and the interval in the question itself. Weekly and monthly counts in Amplitude are unique users over the whole interval, so they do not add up from the daily numbers, and a summary that mixes the two reads as a drop that did not happen.

Built on: Get Active or New Users / Get Real-time Active Users / Get User Composition by Property

Funnel data comes back as a sequence of steps with the count at each one, the conversion rate between them, and how long users took to get through. Because the assistant can also list the event types your project records, it can propose the steps rather than needing you to know the exact event names, which is normally the part that stops this being a five-minute question.

What you end up with. The step where most people leave, what share get through each stage, and how long the ones who convert take.

Start with it bringing you the information and you deciding.

Never without. Have it state the conversion window it used. Amplitude counts a funnel completion only inside a set window, and the same funnel over one day and over seven days gives two different answers with no sign on the result that anything changed.

Built on: Get Funnel Analysis Data / List Amplitude Events / Get Event Segmentation Data

Retention analysis shows how many users return after a starting action, over the days or weeks that follow. Alongside it the assistant can read average session length, the distribution behind that average, and sessions per user, which is the difference between people coming back and people staying once they are there. Those four together answer the question a single retention curve does not.

What you end up with. A retention curve in plain language, plus whether returning users are actually spending time or just bouncing in and out.

Start with it bringing you the information and you deciding.

Never without. Ask for the cohort size behind each point on the curve. Retention percentages from a week with forty signups look exactly like retention percentages from a week with four thousand, and only one of them is worth acting on.

Built on: Get User Retention Analysis / Get Session Average Length / Get Session Length Distribution / Get Sessions Per User from Amplitude

The assistant can find a user by their user ID, device ID, or Amplitude ID, then pull their profile and their event stream. That is the sequence of what they did and when, including the early events that carry where they came from. It is the fastest honest answer to a support ticket that says nothing worked, because you can see the point at which it stopped.

What you end up with. One user's actual event sequence with timestamps, so the support reply describes what happened rather than what should have.

Start with it bringing you the information and you deciding.

Never without. Look up one named user at a time, in response to a ticket you have. This pulls an individual's behavior history into your assistant's context, which is a reasonable thing to do for a person who asked you for help and not something to run across a list.

Built on: Search Amplitude User / Get User Activity from Amplitude / Get User Mappings

The assistant can list every event type in the project, the properties on them, and the categories they have been filed under, then read the individual definitions. Pointed at a taxonomy nobody has looked at in two years, it will tell you which events fire, which ones look like duplicates of each other, and which have names only their author understood.

What you end up with. A written inventory of your event taxonomy with the likely duplicates and the undocumented events called out.

Start with it bringing you the information and you deciding.

Never without. Treat the output as a list to review with whoever ships the tracking, not a list to clean. An event that looks unused is often fired by one paying customer's integration, and the taxonomy is upstream of every chart anyone has saved.

Built on: List Amplitude Events / Get Amplitude Event Types / List Amplitude Event Properties / Get Amplitude Event Property / Get Amplitude Event Categories

Cohorts already defined in Amplitude can be listed, requested for export, polled until the export finishes, and downloaded, optionally with user properties attached. The export runs asynchronously, so the assistant handles the waiting instead of you refreshing a page. What comes back is a file of real users you can hand to the people who will contact them.

What you end up with. The membership of a named cohort as a file, with the properties you asked for, without anyone opening a spreadsheet export screen.

Start with it bringing you the information and you deciding.

Never without. Decide which user properties come with the export before you run it. This takes identifiable customer records out of Amplitude and into wherever your assistant puts files, and the download link Amplitude issues stays valid for days afterward.

Built on: List Amplitude Cohorts / Request Amplitude Cohort / Check Amplitude Cohort Status / Download Amplitude Cohort File

Annotations and releases can both be created through the connector, so the assistant can put a dated marker on your charts when a version goes out and write what changed. Six weeks later, when a metric moved and nobody remembers why, the answer is sitting on the chart. Annotations can be listed and edited afterward, so a wrong date is a correction rather than a problem.

What you end up with. Dated markers on your charts naming what shipped, so a change in a metric can be lined up against a cause.

Start with it doing the work and checking with you.

Never without. Give it the deployment date rather than letting it use today. An annotation is a claim about when something happened, everyone reading the chart will believe it, and a marker on the wrong day sends the next investigation after the wrong release.

Built on: Create Chart Annotation in Amplitude / Create Amplitude Release / List Chart Annotations / Update Amplitude Chart Annotation

Amplitude can alias user identifiers together so one person who signed up on the web and again on mobile stops counting as two. The assistant can search for the identifiers, read the mappings that already exist, and submit the merge. This is worth understanding before you use it: Amplitude's connector flags the mapping action as destructive, and the merge rewrites how that user's history is counted.

What you end up with. One person represented once, with the identifiers that were merged written down so the change can be traced later.

Start with it doing the work and checking with you.

Never without. Read the existing mappings and approve each pair yourself, and do it in a development project first. The action is flagged as destructive, it accepts up to two thousand mappings in one request, and a bad merge changes the user counts underneath every saved chart and cohort you have.

Built on: Search Amplitude User / Get User Mappings / Map Users in Amplitude