Signals

A signal answers one question: what is true about this store right now that was not obvious before?

It is not a task. It has no owner and no estimate. Some signals never become anything — they exist so that the next finding has context.

Anatomy of a signal

Every signal names its source. If a claim rests on 40 sessions, the signal says so, and HI Engine ranks it below anything better supported.

Where they come from

Agents run on a schedule against each connected source. Each agent looks for a different class of problem:

• Shopify — catalogue gaps, thin product pages, collections that receive traffic and convert poorly, theme issues.
• GA4 — a segment that stopped converting, a channel that changed shape, a funnel step that lost volume.
• Click Context — an element people fight with, a section nobody scrolls to, a form field people abandon, a script that throws errors.
• Klaviyo — a flow or campaign whose performance moved.

Reading them well

Two habits save time.

Split before you conclude. A store-wide conversion rate of 1.3% can be 2.0% on mobile and near zero on desktop. The second fact is the story. Signals that matter usually name a segment.

Check the sample. Two groups of visitors rarely convert at the same rate even when nothing separates them, and the smaller the sample, the more of that everyday variation looks like a real difference. HI Engine says when a difference is not statistically significant, and ranks it as a hypothesis instead of a finding.

From signal to idea

A signal becomes an idea when three things are true:

1. It points at something you can change.
2. The change fits on one surface — one page type, one section, one flow.
3. There is enough evidence to size the change.

The agent then writes the idea, attaches the evidence, and scores it. See Ideas.