More sales from the traffic you already have
No testing tool, no agency, no budget line. Just free instruments, published research and one quarter of disciplined work. This is the exact sequence I would run on any store that has traffic and nothing else.
Every few weeks someone asks me what they should buy to start doing CRO, and my honest answer is: nothing, for the first quarter. The instruments are free, the research that convicts the usual suspects is public, and the scarce resource is not software but sequence. Do things in the wrong order, experiments before measurement, redesigns before leak repair, and the quarter evaporates. So here is the order, as one 90-day roadmap.
You cannot improve a number you have not written down. Two weeks, four instruments, all free.
The scoreboard. Shopify Analytics is the official conversion rate, for reasons I unpacked in the GA4 versus Shopify note. Log four weekly numbers: sessions, conversion rate, checkout completion, average order value. Littledata’s benchmark gives you the context: 1.4% conversion is the Shopify average, 45% checkout completion is normal, above 3.2% conversion is the top fifth.
The funnel. GA4’s funnel exploration, product view to add to cart to checkout to purchase. You want the narrowest step, not the absolute values.
The eyes. Microsoft Clarity, free session recordings and heatmaps. Install it in the first week; by day 30 it quietly becomes the most consulted tool in this entire plan.
The stopwatch. Real-user speed from Google’s CrUX API, which covers any origin with enough traffic, including your competitors. When I pulled July 2026 field data for six well-known stores for the speed benchmark entry, mobile LCP ranged from 1.4s at Allbirds to 3.8s at Brooklinen against Google’s 2.5s threshold. Run the same pull for your store and three rivals; that is your speed baseline, from real Chrome users, for free.
The Chrome UX Report API takes one POST request per origin and returns 28 days of real-user field data. With a free API key from Google Cloud Console, the call is:
curl -s -X POST \
"https://chromeuxreport.googleapis.com/v1/records:queryRecord?key=YOUR_KEY" \
-H "Content-Type: application/json" \
-d '{"origin":"https://www.yourstore.com","formFactor":"PHONE",
"metrics":["largest_contentful_paint",
"interaction_to_next_paint",
"cumulative_layout_shift"]}'
Record the p75 value of each metric, the collection period, and the date you ran it. Repeat for competitors, and repeat monthly on the same day. Thresholds worth writing next to the numbers: LCP good under 2.5s, INP good under 200ms, CLS good under 0.1. If the API returns no record, the origin has too little Chrome traffic; use PageSpeed Insights lab data instead and mark it as lab, not field.
This month needs no experiments because the evidence already exists. Baymard’s abandonment survey convicts the same suspects on every store: extra costs revealed late (39% of abandoners), forced account creation (19%), checkouts that ask too much (18% abandon for length, average US checkout 23.48 form elements against an ideal of 12 to 14). Walk my checkout audit sequence and simply remove what it convicts: show shipping costs early, default to guest checkout, hide every field you cannot justify in one sentence.
Speed belongs in this phase too, because the published studies put real conversion money on it and the fixes are mostly housekeeping: compress the hero images, audit the app list and delete the zombies, kill render-blocking scripts nobody remembers installing. If your CrUX baseline shows the store outside the 2.5s LCP line even after housekeeping, the theme itself is often the floor you are standing on. That is a budget decision rather than a free fix, so I park it on the list with a note: theme-level speed baselines differ measurably between official Theme Store themes, and vendors publishing their catalogs openly, like UTD’s Shopify themes, let you check a candidate theme’s demo speed with the exact same CrUX and PSI method before spending anything.
| Window | What changes | What gets watched |
|---|---|---|
| Days 46 to 59 | One change, for example rewriting the product page above the fold | Conversion rate and add-to-cart rate vs the previous 14 days |
| Days 60 to 73 | Next change, for example shipping threshold messaging | Same metrics, same window length |
| Day 74 to 75 | Write up both windows, keep or revert | The log itself |
Without a testing tool you cannot run parallel variants, so the discipline is sequential: one change per two-week window, everything else frozen, judged against the same metric each time. It is slower and weaker than a real A/B test, and I say so plainly. But with a change log, fixed windows and an eye on traffic mix (a viral post or a sale week invalidates a window; note it and extend), a small store gets steering-quality evidence, which is what it actually needs. Save the formal testing for the day your traffic can feed it.
Where do change ideas come from? Not brainstorms. From Clarity recordings of the narrowest funnel step you found in phase one, and from published test literature, which this diary’s experiments section exists to catalog.
Re-run everything from phase one with the same instruments on the same day of the month: the four scoreboard numbers, the funnel shape, the CrUX pull. Lay the two baselines side by side and write three lists: what moved, what did not, and what you now believe about this store that you did not believe in week one.
A quarter of CRO with no budget buys you the one asset money cannot: a store you have actually measured, and a list of beliefs with dates on them.
Then the next quarter plans itself. If the leaks moved the needle, there are usually more; if speed is the wall, you now have field data to justify the spend; if nothing moved, your problem is upstream of the site, in traffic or offer, and no popup was going to save you anyway. Either outcome, you enter the next 90 days knowing, instead of guessing.
Yes, with reduced certainty. You run sequential tests: fix one thing, hold everything else still for two weeks, compare against the same two weeks' checkout completion and conversion rate, and log traffic mix so you notice when a comparison is contaminated. Baymard-documented leaks like cost surprises and forced accounts do not need a test at all; they are removals, not hypotheses.
Four. Shopify Analytics as the official scoreboard, GA4 for funnel and channel breakdowns, Microsoft Clarity for free session recordings and heatmaps, and Google's CrUX API or PageSpeed Insights for field speed data. All four cost nothing at any traffic level a small store will see.
I will not promise a number, and I distrust anyone who does. The published ceilings are real though: Baymard estimates an average 35.26% conversion upside from checkout design fixes alone for large sites, and Littledata's benchmark shows the gap between an average Shopify store at 1.4% and the top fifth above 3.2%. Where you land inside that range depends on how leaky your store is today.
Spend the first hour counting checkout form elements against Baymard's ideal of 12 to 14, and switching off anything that surprises people with costs late. Those two fixes carry the strongest published evidence per minute of work of anything in this plan.