Get started

About Kassandra

you want to hear it on tuesday. that is the whole company.

Kassandra exists because a checkout step timed out at 14:20 and the weekly report mentioned it the following Monday. Nobody was careless. The report simply ran on a schedule and the problem did not.

Get started See pricing

Six days and nineteen hours. Kassandra’s job is to shrink that to the length of a Slack notification.

The problem

dashboards are very good at showing you what already happened

A funnel dashboard is a photograph. It is accurate, well lit and out of date by the time somebody opens it. Teams compensate by checking it more often, which turns a growth lead into a person who refreshes a tab.

Checking is a job. Noticing should not be.

Kassandra was built for the gap between those two verbs. It reads your signup-to-paid funnel against the same hour on the same weekday over the last four weeks, so a quiet Sunday is not mistaken for a broken Monday. When the real number leaves the expected range, it says which step moved, offers the likeliest cause and lists the accounts caught in it.

  1. Weekly report, read Monday, questioned Tuesday, fixed Wednesday.
  2. Alert in Slack after 40 minutes under baseline, with the accounts attached.

Principles

four rules we check every feature against

01

say the sentence first

Every alert and every report opens with one plain sentence about what changed. The chart is there to prove the sentence, not to replace it.

Test: could a board member repeat it in a lift?

02

one number, one source

The total on a dashboard tile equals the total on the funnel page, because both read from the same table. If two screens ever disagree, that is a bug and we treat it as one.

03

quiet is a feature

Kassandra learns four weeks of weekday rhythm before it speaks, honours quiet hours, and prefers one useful alert to five anxious ones.

An alert nobody acts on is counted against us.

04

forecasts must explain themselves

A projected retention number comes with its range and its reason. “Fewer members built a second dashboard in week one” is a reason. A confidence coefficient is not.

How it is built and supported

read-only by default, answered by someone who reads the alert

Connectors ask for read-only credentials, and Kassandra never writes back to your Postgres, Stripe, Snowflake or HubSpot. It pulls up to 30 days of history on connection, then syncs on a rolling basis; the app shows “Synced 2m ago” so you can see how fresh a number is.

Support questions about a specific alert get answered with that alert open in front of us. Pantheon workspaces get a named analyst for the harder ones.

  • data accessRead-only credentials on every connector. Revoke them from your own database whenever you like.
  • historyUp to 30 days pulled on connection, so alerts have a baseline from day one.
  • learning periodFour weeks of weekday rhythm. During that time Kassandra only watches, and says so.
  • access controlSSO and role-based access on Pantheon, with a quiet-hours setting on every plan.
  • supportHelp center and community for everyone, a dedicated analyst on Pantheon.

Where it is going

the next things on the list, in the order we’ll get to them

  • Anomaly alerts against weekday baselines, 42 chart types, PDF and Slack digests, and plain-language cohort forecasts.

    Available

  • Alerts that group related dips, so a failing payment step produces one message instead of four.

    In progress

  • More connectors for the tools teams already keep their events in, chosen by what people ask for most.

    Planned

  • Forecasts for revenue cohorts written with the same rule as retention: a range and a reason.

    Planned

No dates on purpose. A date on a roadmap is a promise, and we would rather keep fewer of them.

next tuesday at 14:20, somebody should already know.

Oracle costs $0 and watches three data sources. Prophet costs $69. Pantheon costs $529 and comes with an analyst.

Get started