How Store Decode chooses an observation

Store Decode chooses an observation with fixed rules that run each night over your own numbers. It keeps the cards whose minimum sample is met, ranks them, and shows at most three on Home. Each observation describes what was measured and offers an option; none predicts an outcome.

The method in plain words

An observation is a short card about one pattern in your numbers, such as a slide few visitors see or an image people click that is not a link. Each night, once yesterday’s numbers are final, twelve rules look for their pattern in the last 7 or 14 days.

A rule that finds its pattern still waits for enough data. Each rule has a minimum, counted in its own unit: sessions, views, clicks or impressions. Below it, Home shows the progress as “have of need” and the card is not ranked.

The cards that pass are ranked by how many sessions a week the pattern touches and how strong it is. Then Store Decode removes repeats, holds back what you already acted on, and fills at most three places on Home. The full list is on the Observations page.

Separately, Explore and the page map mark sections with signals, such as Dead clicks or Working well. A signal is a plain-words label from the section’s own numbers, with its own simpler rules.

What it means for your numbers

• Few cards, on purpose. A quiet Home usually means no rule found its pattern with enough data, or you have already acted on what it found. Explore still has every number.
• No flip-flopping. Once you Mark as done a card, no card about that section comes back for 60 days. Its before and after report is the only follow-up.
• A card read as an option. The sentence states what was measured. The option comes after “You could:” and ends with what to compare afterwards. Some cards carry numbers and no option, when no option fits.
• You decide. Store Decode cannot see your design, your brand or your campaigns, so a card describes and suggests, and you decide.
• Before and after. A card you mark done starts a report that compares the same numbers before and after the day you marked it. This is not a test. Promotions, seasonality and other changes move it too.

In the sample store

Tidewell Home is an invented sample store, and every number here is made up. It marks one example of each of three signals (the full rules would mark a few more): Dead clicks, Seen, not clicked and Working well.

Sample store, Mobile · 7 days · n=4,100	Signal	The numbers that give it
Why Tidewell	Dead clicks	231 dead clicks (50 or more needed)
Shop by category	Seen, not clicked	Views 91% (3,731 of 4,100), 0.2 clicks per 100 views
Bestsellers	Working well	31.8 clicks per 100 views and 44 orders after a click, seen by 1,642 sessions

Its three cards are the app’s own observation types, each with the rule’s own wording and option. One is about dead clicks on Why Tidewell. One is about Shop by category, seen by many and clicked by few. The third is about slide 2 on the Hero slideshow.

The exact rules

The twelve rules

A card shows only when its minimum is met. Each rule also has a second, larger sample size that raises the card’s confidence, and where a rule uses a statistical check, that check has to pass for the higher level. Windows end yesterday, in your store’s time zone. How Store Decode measures a Shopify store explains the units.

The pattern, in merchant words	Looks at	Fires when	Needs at least
Visitors never get past the top of the homepage	Homepage, each device, last 7 days	At least 35% of mobile (25% of desktop) homepage sessions never see anything below the second section	300 homepage sessions on that device
Slide 2 is seen by few of the slider’s viewers	Homepage slider sections, all devices, last 14 days	The second most viewed slide is seen by under 25% of the slider’s views. Slides are numbered by views, not by their order on the page	300 slider views
A section that is clicked more than most but seen by few	Homepage sections with a link, button or field, all devices, last 14 days	Views under 30%, clicks per 100 views at least twice the median of sections with 100 or more views, and at least 3 orders after a click	1,000 sessions that had the section on the page
Seen by many, clicked by few	Homepage product grids, tiles, slideshows and banners, all devices, last 14 days	Views of 50% or more, under 0.3 clicks per 100 views and no orders after a click	1,000 views of the section
Dead clicks on one element	Homepage, all devices, last 7 days	The element with the most dead clicks in a section has dead clicks equal to at least 2% of the sessions that viewed the section	30 dead clicks on that element
Two slides or banners in one place compare differently	The two most viewed creatives of a homepage section, all devices, last 14 days	Their click rates differ significantly, or their sales after a click per 1,000 views differ at least twofold with at least 3 orders on each	500 views on the smaller one
Products low in a collection grid are rarely seen	Grid positions in each collection, each device, last 14 days	At least 3 positions are seen by under 15% of mobile (25% of desktop) collection page views	500 collection sessions on that device
A product card seen often and clicked rarely	Product cards in each collection, last 14 days	At least 500 impressions, and a click rate at or below the collection’s 10th percentile and under a quarter of its median	500 impressions
Filters take the first screen on mobile and few use them	Collection pages on mobile, last 14 days	Of the sessions that had filters on the first screen, under 3% used them	500 such sessions
Visitors see only the first product image	Product pages, each device, last 14 days	Under 30% of product page sessions reach the second image	300 product page sessions
A block further down a product page goes with more add to carts	Product pages, each device, last 14 days	At least 20 sessions added to cart, under 40% of sessions reach a block below the first screen, and those that do add to cart at least 1.5 times as often as those that do not	300 sessions on the smaller side
Before and after a change	A change Store Decode detected or you marked done, at 7 and 14 days	Always, once both windows are closed	300 sessions on the smaller side

The card about a block further down a product page compares two groups of visitors. It describes who added to cart and carries a caveat that it does not show why.

How up to three are picked

1. Each candidate gets a weight: the sessions it affects in a week, times how strong the pattern is. A small factor that grows slowly with the sales after a click in that section multiplies it.
2. Only cards that reached their minimum are considered.
3. 60-day cooldown. If you marked a card done, no card of any rule for that section is picked for 60 days.
4. 14-day re-show rule. A card that was on Home in the last 14 days comes back only when its main number is at least 20% worse than when it was last shown, on the same device. A card still on Home stays until you act on it or a higher-scoring card replaces it. Before and after reports are exempt.
5. Duplicates for the same section are removed and the higher score kept. If two cards would suggest changing the same section, the lower one keeps its numbers and loses its option.
6. The places fill by score, at most three. Each card is in one category: move, fix or remove. When the next card’s category is already on Home, a card from a category not yet shown takes the place if it scores within 5% of it.

In Explore, a product-page behaviour that under 20% of sessions do carries the signal Worth a look when those sessions added to cart at least 3 points more often than the rest. It shows its numbers and no option, and it is a comparison, not a cause.

Section signals

A section gets one signal at most, in this order of priority: Dead clicks, then Seen, not clicked, then Phones skip it, then Working well, then Rarely seen. No section gets a signal until 300 sessions have had it on the page.

• Dead clicks. 50 or more clicks in the section that did not land on a link, button or other control.
• Seen, not clicked. Views over 90% with under 1.5 clicks per 100 views, both bounds strict. It applies to a section that has a link, button or form field. The header, footer and announcement bar are left out.
• Phones skip it. Views on desktop at least 35 points above Views on mobile, with 300 or more sessions of the section’s own on each device.
• Working well. Among the top 3 sections for clicks per 100 views or for orders after a click, seen by at least 100 sessions. The clicks ranking needs at least 2 clicks per 100 views and counts only sections with a link, button or form field. The orders ranking needs at least 3 orders. The header, footer and announcement bar are not ranked.
• Rarely seen. Views under 35%.

How a card reads

• The observation is at most 25 words, holds measured numbers, and never starts with an instruction. Any comparison in it is with your store’s own numbers, never with another store’s.
• The explanation is at most two sentences and is descriptive.
• The option reads after “You could:”, starts with a lowercase verb, and ends with what to compare afterwards, as in “, then compare clicks per 100 views for the two weeks after”.
• The wording comes from a fixed template for each rule. A rewrite is kept only if every number in it matches and it passes the same claims check. Otherwise the template stays.

Before and after reports

When you mark a card done, Store Decode records the change and reports on it after 7 days and again after 14, on all devices together. No report comes after 28 days. A report is not an observation rule: it reads Higher after, About the same or Lower after on one number, never “worked”. Compare your Shopify store before and after a change gives the periods, the four numbers in order and the sessions each side needs.

See the sample store’s numbers and an observation card