90-Day Growth Sprint for Shopify: A Week-by-Week Plan

Quick answer: A good 90-day growth plan for a Shopify store spends the first 14 days on measurement, not growth. Do not start a single test until Shopify, Google Analytics and your ad platform agree closely enough to trust: that is the day-14 gate. Then write one hypothesis for each lever you can move (paid, organic search, conversion and retention), write the stop rule before you launch, and run them in two waves from week 5 so no two changes touch the same page, audience and metric at once. Use week 13 to scale, kill or retest, and to plan the next quarter.

Many 90-day plans are lists of phases: foundation, growth, scale. They rarely tell you what to do on a Tuesday in week 3, and they skip the thing that decides whether the other 76 days are worth anything: whether you can trust the numbers you will judge them by.

This plan goes week by week. It is written for an owner-operated Shopify store, roughly $150K to $1M a year in revenue, in Australia or the United States, where one or two people split the marketing. It treats paid, organic search, conversion and retention as one plan, because they push on the same few numbers and interfere with each other when run separately.

Where Prism3 sits: we sell Shopify SEO, GEO and CRO. We do not run ads, and we do not build or manage email or SMS. The paid and retention parts are for you or whoever runs those channels. The tracking check, the organic work and the conversion work are what we do.

This plan will not promise a revenue figure, a ranking or a timeline, because nobody can. Nothing here is a client result: every dollar figure in the worked examples is illustrative arithmetic. Platform rules come from the platforms’ own help pages, which I read on 31 October 2026 and which change.

The 90 days at a glance

Three rules hold it together:

1. One live hypothesis per lever at a time. Four levers, so at most four things in flight.
2. Every change goes in a change log with the date, the lever and the page or audience it touched. Without it, in week 11 you will not know why a number moved.
3. No decision before its pre-agreed point. You can look at a test at any time. You can only stop it for the reason you wrote down.

Before day 1: three decisions

Pick the start date. If your biggest sale of the year falls inside the 90 days, start after it or accept that the weeks around it are not comparable. A sale week distorts conversion rate, order value and new-customer share at once.

Pick the headline number. One number tells you whether the quarter worked, for example revenue after discounts per month, read next to what you spent to win the customers. If you run paid, decide now whether you will judge it on ROAS, MER or blended CAC, because they can disagree; our guide to ROAS, MER and blended CAC explains how. Changing the scoreboard in week 9 is how plans get marked as successes.

Name an owner for each lever. Paid, organic, conversion, retention. One person each, even if it is the same person wearing four hats.

Weeks 1 to 2: get measurement you can trust

You are not after perfect numbers. You want to know how wrong each one is, and to remove the errors that matter.

The Shopify session change to plan around

As at 31 October 2026 there is one dated trap. Shopify changed how Analytics measures sessions in a rollout from 21 to 23 September 2026. Its help page says sessions now end after 30 minutes of inactivity rather than at midnight UTC, some sessions without a pageview are now counted, and identified bot sessions are filtered out of session reports by default. Conversion rate, add-to-cart rate, checkout rates and bounce rate can shift because sessions are the denominator; orders, sales and customer counts are not affected.

Shopify’s advice is to treat post-update data as a new baseline, compare periods that both sit after the update, and use the same “Human or bot session” filter on both. So your baseline window must sit entirely after 23 September 2026. A four-week baseline made up of weeks after that date is fine. One that reaches back into August is not comparable on any session-based number.

Week 1: name the system of record

For each question you will ask, pick the one system that answers it. Shopify is the anchor for orders and sales because it records the order itself; analytics tools estimate around it, so where a tool disagrees with Shopify on orders, the tool is the suspect. Our guide to GA4 against Shopify Analytics covers why they differ even when both work.

Two Shopify definitions help. Conversion rate is the percentage of sessions that resulted in a purchase, and one customer can make several purchases in a session, so orders and “sessions that completed checkout” can differ. Sessions are also counted only where visitors consent to cookies. Neither is a fault, but your conversion rate will not match another tool’s.

Week 2: reconcile, fix, baseline

Do these in order.

1. Reconcile orders. Take the same 28 days, count orders in Shopify admin and purchases in GA4, exclude test orders and confirm both use the same time zone. Illustrative arithmetic: Shopify shows 412 orders, GA4 shows 371 purchases. The gap is 41, about 10% of Shopify’s count. Decide your tolerance before you look. My working rule is within about 5% with a nameable cause for the rest; that is a rule of thumb, not an industry standard. If you cannot name the cause, start with our checklist for checking your GA4 setup and events.

2. Make sure a purchase is counted once. Place a test order and check it appears once in GA4. Google’s ecommerce documentation shows purchase events carrying a transaction ID, and notes that a purchase with a different transaction ID is not deduplicated against the previous one. If you send purchases to Meta from both browser and server, Meta matches the two copies by event name plus event ID, and only if they arrive within 48 hours of the first event with that ID. If the IDs do not match, the same order may be counted twice in Ads Manager.

3. Tag every outbound link. Put UTM parameters on ad, email, SMS, social bio and influencer links. Shopify’s marketing reports attribute traffic with campaign parameters to marketing, and its default model is last non-direct click, which gives all credit to the last channel before purchase, excluding direct visits. A link without campaign parameters carries no campaign detail into your reports, so you cannot judge that channel.

4. Know what your ad platform claims. Meta sets attribution at ad set level (click-through counts events within 1 or 7 days of a link click; view-through counts events within 1 day of an impression). Meta also says results cannot be compared across ad sets using different attribution models, and that if you use external analytics you should evaluate performance there. Record the setting on every active ad set, then compare Meta’s purchases with Shopify orders from the same channel. They will not match. A stable ratio is usable; one that swings weekly is a measurement problem.

5. Save the funnel. Record sessions, add-to-cart, reached-checkout and completed-checkout, by device. Shopify’s behaviour reports include a conversion rate breakdown shown as a funnel with open and closed views. GA4’s funnel exploration lets you define up to 10 steps and save the result as a report, and reports only the first sequence of steps a user completes in the date range. Pick one tool as the funnel of record.

6. Export Search Console. Check the property is verified and the sitemap submitted, then export queries and pages. Google’s traffic-drop guide points to a “Last 16 months” date filter and says to use the Search Analytics API or bulk data exports if you want longer history, so save your baseline now.

7. Write the baseline sheet. One page, four full weeks: the headline number, orders, new customers, returning customer rate, average order value, sessions and conversion rate by device, paid spend by channel, and organic clicks and impressions. Use Shopify’s definitions: returning customer rate is returning customers divided by customers who placed orders, and average order value is gross sales minus discounts, divided by orders.

One timing caution: GA4 says data processing can take 24 to 48 hours and reports may change during that time. Reconcile a window that ended at least two days ago.

The day-14 gate

Each line passes or fails. It is a gate, not a deadline.

If every line passes, move on. If one fails, extend measurement by up to a week and fix it. If it still fails, run the quarter as a measurement-and-fixes quarter: ship the plainly broken things, hold every test and revisit at day 30. Our CRO audit includes a tracking check for the same reason: a test result on top of a broken purchase event is not a result.

Weeks 3 to 4: diagnose, then write one hypothesis per lever

Week 3 is reading. Week 4 is writing. Do not launch in week 3.

Two lanes: fixes and tests

• Fixes are plainly broken things where no experiment is needed: a checkout error, a discount code that will not apply, a sold-out product still advertised. Ship these in weeks 3 and 4 and log the date, because they change your baseline.
• Tests are anything where you are guessing which version is better. These get a hypothesis, a metric and a stop rule.

If unsure which lane something is in, it is a test.

The hypothesis template

Because [evidence from your data], if we [change] for [page or audience], then [one metric] will move [direction] by about [size] over [window], and we will stop early only if [guardrail].

The “because” is the part people skip. If you cannot point to a number, a recording or a customer message, you are shopping for ideas. When you have more candidates than capacity, our guide to what to fix first on a Shopify store helps rank them; score on potential, importance and ease, and pick one per lever.

Lever one: paid (yours, or your ads manager’s)

Look first at the Delivery column in Ads Manager, then spend per purchase against what Shopify can see.

Meta’s help page says ad sets exit the learning phase once they deliver stably, which usually takes about 50 results in the week after the last significant edit. Significant edits include changing the optimisation event, audience or creative, or pausing the ad set; bid or budget changes can count depending on size. Meta recommends waiting to edit until learning ends, and says combining similar ad sets also combines their learnings.

That gives paid a common first hypothesis, and it is arithmetic. Illustrative: if a purchase costs A$40, an ad set needs about A$2,000 a week (50 × A$40) to reach 50 results. At A$100 a day it spends A$700 a week and gets about 17 purchases. So: “Because three similar ad sets get about 17 purchases a week each, if we merge them into one (about 52 a week on the same total spend) and leave it unedited for two weeks, it will leave Learning limited and cost per purchase, read against Shopify orders, will not rise.” The detail is in our guide to Meta’s Learning limited status.

Stop rule: no edits during learning; Meta says editing then resets it.

Prism3: none of this. We do not run ads. We work on what happens after the click, which sets the denominator of every paid number.

Lever two: organic search

Look first at the Search Console performance report, filtered to queries where you appear but not near the top. Our guide to page-two keywords in Search Console shows where a better title, clearer copy and a few internal links are the cheapest move.

Illustrative hypothesis: “Because a core collection page appears for 40 related queries with many impressions and few clicks, if we rewrite its title and introduction to match the search and link to it from three related pages, impressions and clicks to that page will rise over the next 10 weeks.”

Stop rule: organic does not get a week-3 kill. Google’s SEO starter guide says some changes take effect in a few hours and others could take several months, and that you likely want to wait a few weeks to assess the work. Set checkpoints at weeks 8 and 12 on leading indicators (the page is indexed, impressions are moving, the page’s queries are climbing) and make the keep-or-drop call at week 13 on those, not on revenue. Nobody can promise rankings or a date.

Prism3: this is our SEO and GEO work.

Lever three: conversion

Look first at the saved funnel, split by device. Find the biggest drop between two steps on the device carrying most of your traffic.

Illustrative hypothesis: “Because mobile shoppers drop more between product page and add-to-cart than desktop shoppers, and recordings show them scrolling past shipping and returns information, if we show delivery times and the returns promise above the add-to-cart button on mobile, mobile add-to-cart rate will rise, with purchase conversion and average order value as guardrails.” The guardrails matter: a change that lifts add-to-cart and lowers purchases is a loss.

Stop rule: the sample-size rule below. Many stores cannot run a proper conversion A/B test in a sprint; say so and pick another method.

Prism3: this is our CRO service. The six-phase audit includes a tracking check, and fixes are built in your store and approved by you first.

Lever four: retention

Look first at Shopify’s customer cohort analysis, which groups customers by the date of their first order and shows repeat orders by period since that order. Read older cohorts first: when their second orders cluster is your natural repurchase window, and it sets how long a retention test must run. A new post-purchase message cannot be judged in week 6 if customers usually reorder in month 3. Our guide to a retention plan for the first 90 days after purchase covers what to send.

Illustrative hypothesis: “Because older cohorts place their second order mostly in the third month, if we add a replenishment-timed message at that point for customers acquired in weeks 1 to 4, a larger share of that cohort will reorder than the matching cohort last quarter.” Compare cohort with cohort, not month with month.

Stop rule: you will not see the full result inside the sprint. Judge on a leading indicator (clicks, replies, second-order rate so far) and the cohort report at week 13, and carry the full read into next quarter.

Prism3: none of the sends. We do not build or manage email or SMS.

The collision map

Two hypotheses may run together only if they do not touch the same page template, audience or headline metric.

If two of your four collide, delay one by four weeks. That is what wave two is for.

Weeks 5 to 12: run it

The weekly read is 30 minutes: the same handful of numbers from your baseline sheet, the change log and the guardrails. Place a test order every Monday too, flagged so you can exclude it from reporting. The costliest failure of a sprint is a purchase event quietly broken in week 7 by a theme update, a new app or a consent banner change, and noticed in week 12.

The decision rule for stopping a test

Write four things before launch: the metric (one), the minimum run time in whole weeks so every weekday is represented (my practice, not a law), the minimum sample, and the guardrails.

Early stops are for guardrails only. Stop early if checkout errors rise, refund requests spike or margin drops below a floor you set. Do not stop early because a version looks like it is winning. Looking is fine; deciding is not.

How big a sample you need

R. Lehr’s “Sixteen S-squared over D-squared”, published in Statistics in Medicine in 1992, gives an approximate sample size of 16 times the variance divided by the square of the difference you want to detect, for 80% power and a two-tailed 0.05 test (PubMed record, DOI 10.1002/sim.4780110811). The paper’s own title calls it a crude estimate. I apply it here to each version in a two-version test, and for a conversion rate p I take the variance as p × (1 − p).

Illustrative arithmetic for a store converting at 2%, variance 0.02 × 0.98 = 0.0196:

• Detecting 2.0% to 2.4% (a 20% relative lift): the difference is 0.004, so 16 × 0.0196 ÷ 0.004² = 19,600 sessions per version, 39,200 in all.
• Detecting 2.0% to 2.2% (10% relative): the difference is 0.002, so 78,400 per version, 156,800 in all.

If a test covers half of a store’s 20,000 monthly sessions, that is 10,000 test sessions a month, and the 20% test needs about 3.9 months (39,200 ÷ 10,000). It does not fit in a sprint, and it shows why “we ran it for a week and B won” is rarely a finding.

When you cannot reach the sample

Do not run a thin A/B test and call it a result. Choose one of these and say which in the hypothesis:

• Test a bigger change. A change that moves the rate by 50% needs a far smaller sample than one that moves it by 5%.
• Test a nearer metric. Add-to-cart has more events than purchases, so it fills faster. Keep purchase conversion as a guardrail.
• Ship a before-and-after. Compare four weeks before with four weeks after, same weekday mix, traffic mix checked, and the move visible on both mobile and desktop. This is evidence, not proof; write it that way. Our guide to A/B testing on a low-traffic store covers it properly.

Google Optimize was switched off on 30 September 2023, and Google’s help page points to third-party testing tools that integrate with Analytics. On a Shopify store, “experiments” now means a testing tool you chose or a before-and-after read.

Rules for paid tests

Platforms publish their own thresholds, and they differ. As at 31 October 2026:

• Meta A/B tests. Meta recommends at least 7 days, allows a maximum of 30, says shorter tests may be inconclusive, recommends one variable and a hypothesis first, and advises against informal testing by switching ad sets on and off. It says an A/B confidence of 65% or higher represents a winning result, a lower bar than the 90% it uses for lift tests, so treat a 65% win as directional and check it against Shopify orders.
• Google Ads Smart Bidding. Google’s help page says the “Learning” status lasts at most 7 days, performance can take one to two conversion cycles to settle, and results should be judged over two conversion cycles or two weeks, whichever is longer.

The four outcomes

When the stop rule says stop, write one in the log:

• Scale: reached sample, hit the metric, guardrails held.
• Keep running: not yet at sample or run time. Write the date you will look again.
• Kill: reached sample without moving the metric, or a guardrail broke. Record what you learned.
• Retest differently: inconclusive because it was underpowered. Change the design, not the hope.

“Inconclusive” is a valid answer. A quarter of four inconclusive results and a trustworthy baseline beats four false wins.

Week 13: the review is the quarterly plan

1. Rebuild the scoreboard with the same headline number, definitions and window length as the baseline. Mark any weeks where tracking broke.
2. Go through the decision log: scaled, killed, running or retested.
3. Separate what moved from what you did. Seasonality, a sale, a price change or a stock-out can move your headline number without any test doing it. Note them; do not claim them.
4. Pick next quarter’s hypotheses, one per lever, ranked the same way. Carry forward the organic and retention items that need longer.
5. Decide what to stop, including any channel you could not measure honestly.
6. Write the next gate. Anything that broke in measurement becomes a check in the next day-14 gate.

You now have a decision record and a plan, not a deck. For how the four levers fit together, see the DTC growth system across paid, organic, CRO and email.

One last practical point: if you only have capacity for two levers, run two. A plan you finish beats a plan on a wall.

Key takeaways

• Spend the first 14 days on measurement. Treat day 14 as a gate, not a deadline.
• Shopify changed session measurement between 21 and 23 September 2026, so a conversion-rate baseline should sit entirely after that.
• Anchor orders in Shopify, tag every outbound link and check a purchase is counted once.
• Run one live hypothesis per lever, with a dated change log and a collision map.
• Write the stop rule before launch. Stop early only for a broken guardrail.
• Work out sample size first. At 2% conversion, a 20% relative lift needs about 19,600 sessions per version by Lehr’s rule of thumb, which many small stores cannot reach in a sprint. Test bigger changes, nearer metrics or use a before-and-after read.
• Organic and retention need longer than 90 days to read fully: judge them on leading indicators.
• Prism3 does SEO, GEO and CRO. We do not run ads, email or SMS.

FAQs

What should a 90-day growth plan for a Shopify store include?

A baseline you trust, a gate before any testing, one hypothesis per lever (paid, organic, conversion and retention), a stop rule for each, a change log and a week-13 review. It should not include invented benchmarks for a good quarter.

Why wait 14 days before testing?

If Shopify and GA4 disagree on orders, or a purchase is counted twice in Meta, every test inherits that error. Fourteen days is long enough to reconcile and write a baseline. If it takes longer, take longer.

How many tests can I run at once?

One live hypothesis per lever, so up to four, provided the collision map is clean. If two collide, delay one by four weeks.

When should I stop a test early?

Only when a guardrail breaks: checkout errors, refunds or a margin floor. Not because it looks like it is winning or losing. Write the guardrails and minimum run time before launch and decide at the pre-agreed point.

Is 90 days long enough for SEO?

Long enough to start and read early signals, not to promise a result. Google says changes can take a few hours or several months and suggests waiting a few weeks before assessing. Judge organic work at day 90 on indexing, impressions and query positions, and continue into the next quarter. No honest agency can guarantee rankings or traffic.

Does Prism3 run the ads and emails in this plan?

No. We sell Shopify SEO, GEO and CRO. The paid and retention parts are for you or whoever runs those channels. We help with measurement, organic and AI-search visibility, and conversion after the click.

Keep reading

• The DTC growth system: paid, organic, CRO and email
• What to fix first on a Shopify store, in order
• A retention plan for the first 90 days after purchase

Want a second pair of eyes on your numbers before you start?

Not sure which lever deserves your first 90 days? The day-14 gate in a plan like this is a measurement gate, and measurement is where our work starts. Book a free 30-minute diagnostic: we look at where your store loses buyers after the click, using your own numbers, and tell you what to fix first. We do not run ads or email. No pitch. No email gate.

If the diagnostic shows the leak is tracking or conversion, our Shopify CRO service starts with a six-phase audit that includes a tracking check.