Lift

Agent > Lift compares users ARCY helped against users it never touched, across every flow in your product. There is nothing to author and nothing to switch on: every flow you build derives its funnel automatically from its own steps.

Headline

The page opens with one pooled comparison across all your flows, on four numbers:

Finish rate: how many users complete, with ARCY and without.
Time to finish: how long the two groups take from the first step to the last.
Revenue carried: the monthly plan value behind the users who finished in each group.
Furthest step reached: how far each group gets on average, which is the number that still says something on a flow almost nobody completes either way.

The lead figure is the relative lift ("+38% more users finish"), with the raw percentage point difference and both underlying rates underneath so you can check it.

Flows with too little data are left out of the pooled number entirely rather than counted as zero, so a quiet flow cannot drag the headline. The page says how many of your flows the number covers, and stamps when it was worked out. Refresh recomputes on demand.

How this number is measured

Both groups are counted as finished the same way: they reached the last step of the flow in your own product, however they got there. Running ARCY's flow is not what counts as finishing, so the comparison is a real one rather than a restatement of who used ARCY.

With ARCY: users who ran the flow.
Without ARCY: users who reached the same starting point and never used ARCY at all.

Users choose whether to use ARCY, and that choice is not independent of whether they were going to finish anyway. So this is a comparison between two groups of people, not a controlled test. A note saying so appears next to the number and inside the Impact Report, because anyone technical who asks should find an answer rather than a gap.

If your product does not send plan value through identify(), the revenue figure says so rather than showing a zero. The same is true of every other number here: an absent measurement never reads as a bad result.

Funnel cards

Below the headline, one card per flow, most-reached first.

Each card draws the same funnel twice, side by side: Without ARCY on the left, With ARCY on the right, one row per step with the step's own label. Bar length is the number of users who reached that step, and both columns share one scale, so a step is comparable with its counterpart at a glance. The drop-off percentage is annotated between each pair of rows, which is how you find your worst step. Hover a bar for the exact count. On a phone the two columns stack.

Each card ends with both finish rates and that flow's own lift.

A card with too little data still draws both funnels, because the counts are true at any size. Only the lift percentage is withheld, and the card tells you how many more finished users each group needs. A flow nobody has reached yet says so rather than drawing empty bars.

Users who reached the flow and used ARCY somewhere else without running this flow are in neither column. Their count is shown on the card, so the two columns are never mistaken for everyone.

Each card also carries its own timeline: when that flow was published, withdrawn, or had its steps or content changed.

When a flow has no funnel

A flow appears on Lift as soon as its completion step is something ARCY can observe. Two kinds of flow cannot be measured:

the last step is a fill, because nothing in the browser reports a field being filled.
the last step navigates away from your product, where ARCY records nothing.

Such a flow saves and runs normally. The flow builder tells you which of the two it is, so you know exactly what to change if you want the funnel. Adding a click or a navigate step after the unmeasurable one is usually all it takes.

Impact Report

Write the report asks ARCY to describe one flow's comparison in plain language, using only the numbers already on the card. You get the summary, a recommended next step, a stat grid, and a Copy report button, meant for forwarding to your own leadership. The copied text carries the method note with it.

The report is written fresh each time you ask for it, so it always describes the numbers as they stand rather than a week that has since passed.

The Demo environment

Lift renders a full, populated page in the Demo environment, with a banner saying the numbers are sample data. Nothing there comes from your own traffic.