Sales moved. Shopify can now show what you changed

A sales chart can show that something happened. It cannot tell you what the team changed at the time.

That is a familiar problem in a compact ecommerce team. A new landing page goes live on Monday. An app starts a campaign on Wednesday. The fulfillment setup changes on Friday. When someone reviews the numbers two weeks later, the team vaguely remembers the events but not their order. AI then starts with the same weak material as the humans: plenty of data and too little context.

Shopify's July 24 update is more useful than it first appears. Apps can now add annotations directly to analytics charts. A campaign, launch or operational change can sit on the same timeline as sales, traffic and conversion data.

What Shopify changed in Analytics

A Shopify annotation is a dated marker that adds business context to a chart without changing the report data. An app can mark a specific date or a date range, with its name and icon shown next to the annotation.

Shopify lists product launches, campaigns, discounts, new landing pages, supplier changes, payment methods, and fulfillment changes among the possible events. Teams can compare those markers with metrics such as sales, sessions, conversion rate, average order value, units sold, and fulfillment performance.

An annotation still does not prove why the line moved. It shows what happened in the business at the same time. That gives the analysis something concrete to start from instead of a chart with no memory.

Source: Shopify Changelog: New app-added annotations on your analytics charts.

Why this helps a compact ecommerce team

A larger company may keep separate logs for marketing, inventory, web operations and finance. In a shop run by a handful of people, much of that log lives in the head of whoever made the change. It works until the week gets busy, somebody takes time off or several apps act at once.

Annotations move the memory closer to the data. The immediate gain is practical.

• Your weekly review does not begin with reconstructing what happened.
• You can separate a recorded change from an explanation invented afterward.
• You can give Sidekick or another AI tool a timeline that the team can inspect.

This becomes especially useful when an app works in the background. If a campaign app starts an email, a discount app changes an offer, or an inventory app affects delivery time, the event should follow the work into Analytics. Otherwise, you see the outcome but lose the step that came before it.

Not every app will necessarily add annotations straight away. Start by checking which installed apps use the feature. Keep a simple manual change log for the rest and include it in the weekly review.

AI needs events as well as numbers

Shopify describes Sidekick as an AI commerce assistant that works with the context of your store. It can analyze data, create report queries, and help with recurring tasks. Shopify also presents a weekly performance summary as an example of an instruction that merchants can save and reuse.

That makes Sidekick relevant here, but annotations do not turn analysis into a source of truth. AI can find patterns and draft hypotheses. It can also produce a persuasive explanation that falls apart when someone checks the dates. The task should therefore force the AI to separate three things:

1. What the data shows.
2. Which business event sits nearby on the timeline.
3. What remains a hypothesis.

The distinction sounds basic. It makes a real difference when the team has twenty minutes to decide what happens next.

Source: Shopify Sidekick, Shopify's AI commerce assistant.

Build a 40-minute change log

The aim is not to document everything that happens in the store. Capture the changes that could reasonably affect a metric you use to run the business.

1. Choose six metrics

Use the metrics that guide your weekly decisions. A store might track:

• Net sales
• Sessions
• Conversion rate
• Average order value
• Units sold
• Fulfillment performance

Six is enough to reveal relationships without turning the report into a stack of colored lines. If return rate or advertising cost matters more to your shop, replace one item. The list should follow the business rather than a standard template.

2. Decide which events deserve an annotation

Choose four or five recurring event types. For example:

• Campaign started or stopped
• Price, discount, or offer changed
• Product, collection, or landing page launched
• Inventory, supplier, or delivery method changed
• Payment, checkout, or popup changed

Write a brief rule for each type. A campaign annotation might include the channel, audience, and offer. Keep it factual. "New email campaign to previous customers" is better than "major campaign expected to lift sales".

3. Add an owner and the expected effect

Every meaningful change needs a person who can answer questions. Record the expected effect before anyone knows the result.

For example: "New free-shipping threshold. Owner: Sam. Expectation: higher order value, possibly lower conversion. Review after seven days."

You preserve the original intent and avoid rewriting the story afterward to fit the result.

4. Run the same review each week

Choose a fixed period, such as the last seven days compared with the previous seven. Use the same metrics and the same response structure. Once the routine works, save the prompt as a Sidekick skill or run it in the AI tool your team already uses.

Copy this prompt: weekly Shopify review

Use the prompt in Sidekick with the current report. If the tool cannot read a particular app annotation directly, paste the list under "Events".

Help me review our Shopify store for the week.

Period: [date-date]
Comparison period: [date-date]

Metrics we follow:
- Net sales
- Sessions
- Conversion rate
- Average order value
- Units sold
- Fulfilment performance

Events and annotations during the period:
[paste app annotations and manual changes with date, owner and expected effect]

Do the following:
1. Describe the most important changes in the metrics. Include direction and size when the source contains numbers.
2. Match each clear change with nearby events on the timeline.
3. Label every conclusion as OBSERVATION, POSSIBLE EXPLANATION or MISSING EVIDENCE.
4. Do not claim that an event caused an outcome merely because the dates are close.
5. Suggest no more than two follow-ups that could separate the most plausible explanations.
6. Write a short weekly note with: what changed, what we think, what we do next, owner and review date.

If information is missing, state exactly which report, comparison or annotation is needed. Do not invent numbers or events.

The prompt should produce a working note rather than a polished report. If the answer runs long, ask Sidekick to reduce it to five points and one decision.

Review the answer without getting stuck

Read every line labelled OBSERVATION first. You should be able to check each one in Shopify. Remove any observation that cannot be found in the report.

Move to POSSIBLE EXPLANATION. Look for alternatives. A lower conversion rate after a campaign could come from broader traffic, a different offer, slower delivery messaging or something else entirely. Choose the explanation that you can test with the least friction, not the one that sounds most convincing.

Finish with MISSING EVIDENCE. This list is often more useful than the rest of the AI response. It shows which apps need better annotations, which report is absent and where the team still depends on memory.

A useful weekly meeting ends with a decision, an owner and a review date. Two simultaneous experiments can work. Five will make the next chart as difficult to interpret as the last one.

Three workflows that fit immediately

Campaign and conversion

Let the campaign app mark its start and end. Compare sessions, conversion rate and order value for the same period. Ask AI to propose the next test, then have the team approve the change before another campaign goes live.

Supplier and fulfillment

Mark a change of supplier, warehouse, or delivery setup. Follow fulfillment performance and support questions. If an integration reads order or support data, give it scoped access and keep each run in an audit log.

Landing page and product mix

Mark the publication of a new page or collection. Follow traffic, units sold and average order value. Ask AI to distinguish between attracting more people and attracting people who buy.

When the routine is ready for automation

After three or four weeks, you will see which steps repeat. You can then connect app events, report exports and the weekly note in a lightweight automated workflow.

Use Shopify's existing staff permissions, give integrations access to the reports they need, and keep an approval step before AI changes campaigns, prices, or orders. API keys belong in environment variables or a secret manager, not in the prompt. Those controls are enough to make the workflow useful without turning it into a governance project.

A Tool Forge workflow can carry each app event into the weekly note, so nobody has to reconstruct the week from memory on Friday afternoon.

Start with the last fourteen days. Add five real events, run the prompt and check whether AI can distinguish what you know from what you merely suspect. If it cannot, you have found the gap in the log before it becomes another confident guess.