Product cue to funnels

[paolodamico]

Changes

Implements a product cue from Trends to Funnels when a user adds more than 3 events/actions to their Trends graph to suggest funnel usage. Goal is to drive more usage to funnels.

• If the user clicks on continue, they're taken to a funnel which is immediately populated with the same events as steps.
• If the user closes the prompt, we'll persist their preference and don't show the suggestion again (done through FF magic).

In addition, this PR:

• Refactors ErrorMessage.tsx to a more generic InlineMessage.tsx
• Refactors insight icons to support variable sizes.

Additional context

• Feature flag is ready to go @ https://app.posthog.com/feature_flags/1185
• Wanted to show the prompt only if the user hasn't used funnels recently but we no longer record each insight interaction in the API and we don't support negative cohorts (to use engagement cohorts + FFs).

How did you test this code?

• With local self-capture. Made sure the prompt showed up when adding 3 or more graph series.
• Made sure proper events were fired when the prompt: shows up, is clicked or dismissed.
• Made sure the prompt didn't showed up again if the opt-out FF was enabled for the user.

[clarkus]

I think this is a good change, but I think there are two improvements we could consider:

• Placement - This feels like it should be higher priority in the page. Placing it adjacent to the thing it describes is good, but it seems like it should be higher in the page hierarchy. Can it be placed above the graph series steps?
• Language in the CTA - "Try this graph as a funnel" is graph the right term here? Should it be insight?

[paolodamico]

Thanks @clarkus! Agreed on language. Unsure about placement, it's indeed not very prominent in absolute terms, but in relative terms, the box appears just below your cursor (bc it shows up as you add a third step), so your eyesight will likely be at that spot. Depending on your scroll position you may not even be seeing the top of the insight box. Thoughts?

[paolodamico]

FYI @mariusandra tagging Michael instead for a review as he has better context on the altered stuff, but happy if you take a look too

[clarkus]

Thanks @clarkus! Agreed on language. Unsure about placement, it's indeed not very prominent in absolute terms, but in relative terms, the box appears just below your cursor (bc it shows up as you add a third step), so your eyesight will likely be at that spot. Depending on your scroll position you may not even be seeing the top of the insight box. Thoughts?

Technically we are prompting a user to change tabs to a different insight type and pre-filling their steps with the graph series they already set. It feels like it should be adjacent to the control that changes types. Placing the prompt adjacent to the control that they'll use to change types next time will help users learn where to change types.

We might also consider a tooltip or some other overlay that points directly at the thing they can use to change types later. This would be an actionable tooltip (if you click on it, it does something) so it'd need to look more like a popover or menu than our typical informational tooltips (black background). I put together a quick example at https://www.figma.com/file/Y9G24U4r04nEjIDGIEGuKI/PostHog-Design-System-One?node-id=1813%3A512. This tooltip would point to the insights tab.

[paolodamico]

Alright, all feedback addressed. Thanks @clarkus, your proposal makes a lot of sense, particularly the point around teaching users how to do this in the future. I've also implemented a kill switch so we can merge safely & test out internally first.

In addition, we won't be able to launch it just now as the simple funnels experiment just launched and we'll contaminate results as we can't have mutually exclusive experiments yet, cough cough @macobo

[paolodamico]

Experiment launched today, tracking results at https://app.posthog.com/dashboard/26377

[marcushyett-ph]

@paolodamico I'm not too sure how to interpret these results?

[neilkakkar]

Q: Why not dogfood this via the experiments feature? (and we can also compare this dashboard to experiment results)

[marcushyett-ph]

I thought the same but this PR is 2 months old, so we'll need to finish the time-machine feature you've been working on @neilkakkar first

[neilkakkar]

oh, duh 🤦 . But since the rollout just started today, we can work with Experiments?

Actually, I'd say we really should do this via experiments, since our FF rollout %s are unreliable :$

For example, here's a FF I created recently, rolled out to 50% of the population - it's been 4 days, but it's already at a 56-44 divide

• and I expect the gap to continue widening

[marcushyett-ph]

Are we doing something about the FF issue - is it tracked somewhere else?

[neilkakkar]

I don't think there's much we can do here (but good idea to track it anyway. I don't think it is, will create an issue for it).

It's unreasonable to always have decide return before loading pages, which means some users will always fall through the crack of Nones.

For running experiments, most definitely use multivariate FFs only!

[paolodamico]

• Re metrics: H1 is more funnel discoveries on users in the experiment group (we measure this in the first two graphs). Secondarily we want to see a solid conversion rate of users seeing the cue prompt and then doing a funnel (second funnel which is still empty).
• Re using experiments, I thought about this, but I wanted to see how Paths performs first for a trends goal, plus keeping in mind the secondary metric.
• Re FF split, this happens a lot (and thanks for documenting it further), that's why I always measured the metric as a ratio to the number of users in each of the experiment/control groups.