AI recommendations

How Heatwave's AI turns your visitors' behaviour into concrete, evidence-backed changes.

The Recommendations page lists changes to your store, ranked by expected impact. Each one is backed by your own visitors’ behaviour: heatmaps, scroll depth, the first screen, theme sections, speed and problems found in recordings.

How the AI works

1. Collect. With visitor consent, Heatwave records clicks, scrolling, the first screen, speed and anonymous replays. Form inputs and personal data are masked.
2. Detect. Each visit is checked for 29 behaviour patterns, such as failed add to cart, angry repeated clicks, clicks that did nothing and hesitation.
3. Measure. For each page type and device, Heatwave builds a table of facts: heatmaps, fold position, sections seen, speed and how often each behaviour happens.
4. Review. The AI looks at your pages as shoppers see them, using pictures taken by Heatwave’s own browser, together with the facts and short anonymous visit summaries. It then proposes concrete changes.
5. Verify. Before you see anything, automatic checks run on every suggestion. It must cite real facts and real visits from your store, every number it mentions must match one of those facts, and it must not repeat a suggestion you already have. Anything that fails is dropped.

Heatwave uses Anthropic’s Claude models: a fast model summarises visits, and a more capable model writes the recommendations. Your data isn’t used to train AI models.

When recommendations appear

A page type is analysed on a device once it has 200 views there. The Data readiness card shows progress for each page type and device, for example “120 of 200 views”, and marks each one Ready when it gets there. More visits mean sharper recommendations.

Click Analyse my store (later Analyse again) to start an analysis. It covers your ready page types, starting with the ones that matter most for sales, such as product, cart and collection pages. It usually takes 1 to 3 minutes. While it runs, you’ll see each page type and device move from Queued to Running to Done. “Too little data” means that combination didn’t have enough views to produce suggestions.

How often analyses run

Plan	Automatic analyses	Manual analyses per month
Free	None	1
Starter	Weekly	4
Growth	Nightly	Unlimited
Pro	Nightly	Unlimited

The page shows how many manual analyses you have left this month. If nothing new could be analysed, for example because an analysis is already running, it doesn’t count against your allowance. A standard run covers up to 8 page types. On Pro, deep audits cover up to 20 page types per run.

The lanes

Open recommendations are sorted into three lanes:

• Do first: high impact and not much work.
• Plan next: worth doing, but needs more effort or has a smaller payoff.
• Consider: small improvements for later.

At the top, four numbers summarise your list: Open, High impact (likely to move add-to-cart or purchases), Quick wins (meaningful impact, low effort) and Done. You can filter by Page, Device and Category.

Reading a recommendation

Each card shows the title, impact and effort, the page type and device, and the share of visits affected where Heatwave can tell. Click it to open:

• The problem: what’s going wrong.
• The change: what to do about it.
• Evidence from your visitors: the numbers the suggestion is based on.
• See it on the heatmap: opens the right page, device and map type.
• Watch session 1, 2, 3: recordings that show the problem.
• How to do it in Shopify: step-by-step instructions in the theme editor (Online Store > Themes > Customize). When the theme editor can’t make the change, the steps describe the smallest code change instead.

At the bottom you’ll see the category, how strong the evidence is (Strong evidence, Some evidence or Early signal) and when it was suggested.

Done, Snooze and Dismiss

• Mark as done when you’ve made the change. Heatwave tracks the before and after.
• Snooze 14 days to hide it for now. It comes back to the open list automatically after 14 days.
• Dismiss if it doesn’t apply. Dismissed recommendations are kept in case you change your mind.

Use the Open, Done, Snoozed and Dismissed tabs to see each list. Reopen moves a recommendation back to Open.

Measuring impact

When you mark a recommendation as done, Heatwave saves the current value of the metric it was meant to move. Until the next analysis, the card says “Measuring impact” with that starting value. After the next analysis it shows the before and after, labelled Improved, Worse or No change.

Remember that other things change at the same time, such as campaigns, seasons and stock. Treat the result as a strong hint rather than proof.