How it works

The find, size, ship and test loop in detail — what runs on a schedule, what waits for your decision, and where each step is visible in the app.

HI Engine is a loop, not a report. This page describes one full turn of it.

1. Find

Agents read your connected sources on a schedule and write insights — short, dated statements about what changed or what is not working.

Source	What the agent reads
Shopify	Products, collections, orders, inventory, theme files, storefront pages
Google Analytics 4	Sessions, device and channel splits, funnel steps, conversion rate
Click Context	Clicks, rage clicks, scroll and section visibility, form behaviour, JavaScript errors
Klaviyo	Campaign and flow performance

An insight is an observation, not yet a job. See Insights.

2. Size

Insights that point at a fixable problem become ideas. An idea names one change, on one surface, with the evidence attached.

Every idea gets two numbers:

Estimated impact, in your store currency, per month.
A TRICE score, which ranks the idea against every other open idea.

TRICE is five dimensions, each scored 1 to 5:

Letter	Dimension	Weight
T	Traffic affected	0.20
R	Revenue potential	0.30
I	Implementation ease	0.15
C	Confidence in the data	0.20
E	Expected time to a result	0.15

The weighted composite decides queue rank. How impact is estimated shows the arithmetic behind the euro figure.

One change, one idea

The agent never bundles several changes into one "overhaul". Each idea is a single change, so you can approve one thing without accepting four others.

3. Ship

You approve an idea. From there the agent:

Reads your live theme and finds the right template, section or snippet.
Writes the change as a typed theme section or block, with theme editor settings for any customer-facing text.
Runs Shopify Theme Check and fixes what it reports.
Publishes a draft theme named after the task.
Sends you a preview link and waits.

Why a draft theme

A draft theme is a full copy of your live theme with the change applied. Your customers never see it. Publishing it to the storefront is a separate action that only you can take, inside Shopify.

Two rules the agent always follows:

No custom_liquid. Every change ships as a real section or block with a {% schema %}, so it stays visible and editable in the theme editor.
Copy is yours to edit. Any text a customer reads becomes a theme editor setting, not a hardcoded string.

See Reviewing a build.

Testing is the default path for an approved build, not an extra. Half of your visitors get the current experience, half get the new one. The test runs until the difference is statistically distinguishable from noise, or until it is clear that it is not.

You choose who runs the split:

Tool	Who runs it
HI Engine	HI Engine runs and measures it. The recommended option
Intelligems	You run it. HI Engine prepares the launch instructions
Shoplift	You run it. HI Engine records the hypothesis and the result
Shopify Rollouts	You run it. HI Engine records the hypothesis and the result

A winner can be published. A loser is rolled back. Either result feeds the next round: the agent recalibrates how it scores that category of idea for your store.

See A/B tests.

What runs without you

Job	Cadence
Agent runs over Shopify, GA4 and behavioural data	Scheduled, continuous
New insights and ideas written to your board	As they are found
Weekly queue scoring and task selection	Weekly
Build of an approved task	Starts on approval
A/B test measurement	Continuous while a test runs

What always waits for you

Approving an idea for build.
Approving, commenting on, or rejecting a finished build.
Publishing a theme to your live storefront.
Starting and stopping an A/B test.