CRO. by excusethis

Find the step where your store loses the most sales before you test anything.

The kit is a Claude Code plugin. Your AI agent pulls live data from your store, checks the numbers hold up, and ranks every step of the funnel by the purchases it's losing. Then it points recordings and support tickets at the worst step and hands each finding to the team that owns the fix.

See the demo run Run it on your store

Which funnel cell is the worst?

Example Outdoor Co. is a made-up store with 90 days of synthetic data. Try the other two ranking rules and a different cell comes out worst.

Purchases at stake per quarter

    Norm: the same visitors on the other device, or last periodBelow its normAt or above its norm

    Run it before you decide what to A/B test

    It finds the step where your store loses the most money, works out why, and names the team that should fix it.

    A/B testing tools
    Tell you whether a change worked, once you've picked it. At the demo store's traffic, testing the top fix would take two quarters.
    Analytics
    Reports thousands of numbers. Which one matters, and whether its tracking is right, is left to you.
    Page audits
    Score pages against a checklist whether or not that's where the store loses money.

    One run on the demo store

    Every chart below comes from the kit's own scripts, run on the made-up Example Outdoor Co. and its synthetic data, in the order a run happens.

    1. 1. The data has to pass two gates

      Gate 1 checks the traffic comes from real shoppers, which on the demo store meant cutting every query to the countries it sells to. Gate 2 checks analytics against the orders the platform took. The kit draws no funnel conclusions until both pass.

      Sessions by country, 90 days

        In market, keptOut of market, filtered out

        Analytics against platform orders

        95.0%

        of the 4,410 orders the store took also show up in analytics. The share holds steady month to month, so the rates can be trusted.

        1. 2. Every funnel cell against its norm

          The funnel splits into 18 cells: four steps for new and returning visitors on each device, plus listing click-through on each device. A cell's norm is the same visitors on the other device. New visitors on mobile trail desktop by 1.5 to 3 points at every step until checkout, where the gap is 24 points.

          New visitors, desktop against mobile

          DesktopMobile
          0%25%50%75%100%
          Show the numbers
          StepDesktopMobileGap

          Checkout steps, all devices

            One in six checkouts stops before the shipping step, and another one in six stops between shipping and payment.

          1. 3. Is it the page?

            Some gaps come from the traffic a cell gets, a stock-out or a price change, and redesigning the page won't close those. Before pointing anything at C15, the kit measures how much of its 24-point gap each one explains.

            The 24-point gap at C15, split by cause

              Other causesLeft for the page

              What else could explain it

              • Traffic mix explains 3 points. Mobile gets more paid social (25% of its checkouts against 9% on desktop), and paid social converts less on both devices. Every channel still converts less on mobile, which leaves 21 points for the page.
              • Price and promotion changes explain none of it. Three dated changes fell in the window: a summer sale, its end and a footwear price rise. None moved mobile checkout more than desktop.
              • Stock wasn't checked. The stock test runs on product-page cells, and C15 is a checkout cell.

            1. 4. Every other source looks at the worst cell

              The kit pulls recordings, the page walk, support tickets and competitor pages for that one cell, and three of the four land on the same cause. For the questions only shoppers can answer, it also writes a call guide, a survey and a five-person usability test aimed at the same cell.

              Page walk
              Step 2

              The mobile cart hides the order summary, so the shipping cost first appears at checkout step 2.

              Recordings
              31 of 50

              Watched mobile exits left within 20 seconds of the shipping cost appearing.

              Support
              212

              Tickets about shipping cost, the second-largest reason before purchase.

              Competitors
              Both

              Show the free-shipping threshold on mobile product pages and in the cart.

              All four point at
              C15
              Mobile checkout, new visitors 46.0% finish against 70.0% on desktop

              50 watched mobile checkouts with no purchase

              Left within 20 seconds of seeing the shipping cost: 31 Other reasons: 19 (promo code 9, address error 6, other 4)

              Help desk tickets by reason, 90 days

                Three sources agree on the cause, so the finding can score Confidence 4. A finding from one source is capped at 2, and the rule check fails any run that breaks the cap.

              1. 5. Findings, each with an owner and a lane

                Each finding in the register carries its evidence and receipt, an ICE score, a lane, the team that owns the fix, and the result that would prove it wrong. Before you see any of it, a second agent tries to break each one.

                Impact against ease, sized by confidence

                Implement + MeasureQuick WinBigger dot: higher confidence

                What the lanes mean

                Quick Win: under a day's work, often with no developer, such as a copy or settings change.

                Implement + Measure: build it and watch the cell each week. It's the lane for anything the traffic can't test in about six weeks.

                Test: only when the cell has the traffic to finish an A/B test in that time.

                IDFindingSourcesICELaneOwner
                F01New mobile visitors finish 46% of checkouts against 70% on desktopAnalytics, recordings5 / 4 / 3Implement + MeasureEcommerce
                F02Shipping cost first appears at checkout step 2 on mobileWalk, recordings, support5 / 4 / 3Implement + MeasureEcommerce
                F03Both competitors show the free-shipping threshold on mobileCompetitors, walk3 / 3 / 5Quick WinMerchandising
                F04Mobile product pages add to cart at 7.5% against 9.0%Analytics4 / 2 / 3Implement + MeasureEcommerce
                F05Mobile listing click-through is 34% against 40%, and filters reset on backAnalytics, walk3 / 3 / 3Implement + MeasureEcommerce
                F067.8% of typed searches return nothingSearch2 / 2 / 5Quick WinMerchandising
                F07Overseas bot traffic is kept out of every reportAnalytics2 / 2 / 4Quick WinEcommerce

              2. 6. A one-pager that leads with the recommendation

                $55k
                a quarter at stake in one cell

                Show the shipping cost before checkout on mobile. New visitors on mobile finish 46% of the checkouts they start, against 70% on desktop. That gap is about 466 purchases a quarter. At 1,939 checkout starts a quarter a test would take two quarters, so build it and measure the cell each week.

                Open the full report from this run

              What it covers

              The whole store, in twelve chapters

              UX and trustSearchMerchandising and categoriesRange and pricingAudienceRetention and LTVVoice of customerSizingTestingImplement and measureWriting it upGetting it shipped

              Rules every run follows

              • The funnel waits until both data gates pass
              • A finding from one source can't score above Confidence 2
              • Every number links back to a pulled file and its receipt
              • The agent reads from your tools and never writes to them
              • A second agent tries to break each finding before you see it

              174 page checks

              They cover 14 areas of a store, from the home page and navigation to checkout, accounts, accessibility and trade buying. A browser probe measures 34 of them on the live page and the agent judges the rest, starting at the worst cell. Each check spells out what a pass looks like and cites where the idea comes from, such as WCAG 2.2 or web.dev.

              Your stack, over MCP

              The kit lists MCP connectors for about 70 tools in 12 areas of an ecommerce stack, including:

              GA4ClaritycommercetoolsShopifyWooCommerceBigCommerceAlgoliaHotjarKlaviyoZendeskand more

              Run it

              Install onceIn Claude Code.
              /plugin marketplace add christiansbaker/CRO
              /plugin install cro@cro
              See this demoBuilds this run and its report on your machine in about a minute, with no logins.
              /cro:demo
              Your store, no loginsWalks your pages and checks trust signals, speed and one or two competitors. The findings stay hypotheses until your own data backs them up.
              /cro:audit https://yourstore.com
              Your store, live dataConnects read-only to your analytics and tools, then runs the full diagnosis.
              /cro:setup
              /cro:diagnose
              Close the gapsTakes what the last run couldn't check, worst cell first. For each gap it tells you what it needs from you and waits before it runs.
              /cro:next