Micro conversions are the small actions visitors take on the way to your main goal, such as adding to cart, clicking a product, starting a trial or joining an email list.
Macro conversions are the main goal itself: a purchase, a paid subscription, a qualified lead.
Track both, but judge A/B tests on the macro conversion, because our database holds many tests where the micro conversion rose and sales did not.
The terms were popularized by analyst Avinash Kaushik in a 2008 post, "Measure Macro AND Micro Conversions".
Kaushik's point was that most visitors never buy, so a site that only counts orders misses most of what its visitors do.
Macro conversions: the outcome the business is paid for
A macro conversion is the action that directly produces revenue or a sales opportunity.
Nielsen Norman Group defines it as "a desired user action that directly contributes to the primary goals of your business," with completed purchases, submitted lead forms, event registrations and account creation as examples.
Which action counts depends on the business model:
| Business | Typical macro conversion | Typical micro conversions |
|---|---|---|
| Ecommerce store | Completed order (judged as revenue per visitor) | Product view, add to cart, checkout start, email signup |
| SaaS | Paid plan or trial that converts to paid | Pricing page view, trial start, demo request, onboarding step |
| Lead generation | Qualified lead or booked call | Form start, calculator use, phone number click |
| Mobile app | Paid subscription | Install, onboarding completion, paywall view, trial start |
For a store, the order count alone is not the full macro picture.
A change can raise orders while shrinking order value, so most practitioners in our database judge store tests on revenue per visitor, which combines conversion rate and average order value.
For subscription businesses, a trial start is often treated as the macro conversion, but it is really a step.
The outcome is the trial that turns into a paying customer.
A test that lifts trial starts but not paid conversions has not moved the macro outcome.
Why macro conversions are hard to test on
Macro conversions are rare. NN/g cites an average macro conversion rate of 2.9% across industries.
Rare events need large samples: to detect the same relative lift, a 3% purchase rate needs far more visitors than a 20% add-to-cart rate.
That is the whole appeal of micro conversions.
They happen more often, so tests on them reach a result faster.
A rough rule of thumb (Lehr's rule, for 80% power at a 5% two-sided significance level) puts the visitors needed per variant at about:
n = 16 × p × (1 − p) / (p × relative lift)²
Worked example: to detect a 10% relative lift on a 3% purchase rate, n = 16 × 0.03 × 0.97 / (0.003)² ≈ 51,700 visitors per variant.
On a 20% add-to-cart rate, n = 16 × 0.20 × 0.80 / (0.02)² = 6,400. That is about eight times fewer visitors.
The sample size calculator gives exact figures for your own numbers.
Micro vs macro conversions: what to track and what to test
NN/g splits micro conversions into two kinds. Process milestones are steps on the path to the macro conversion, like add to cart or entering payment details.
Secondary actions are not required for the sale but suggest future value, like a newsletter signup or an app download.
Both kinds belong in your analytics. In GA4, any event you collect can be marked as a key event, Google's current name for what GA4 used to call a conversion.
Marking add-to-cart, checkout start and email signup as key events lets you see where a conversion funnel leaks.
Testing is a different question. A micro conversion is only useful as a test metric if moving it reliably moves the macro conversion.
Our tests show that often does not happen.
When the micro conversion rose and sales did not
Clean Commit added one-click size buttons to desktop collection cards so shoppers could add a size without opening the product page.
Add-to-cart rose 29.7% across 13,432 desktop visitors. Conversion rate changed by 0.0%, orders fell 3.5% and revenue per visitor fell 10.4%.
The adds were easier to make, but they did not become orders.
GrowthRock saw the same pattern with quick shop on an apparel store's listing pages.
With hundreds of thousands of visitors, the test showed more add-to-carts but no more orders, and no change in conversion rate.
Clicks are an even weaker signal. Clean Commit's wider desktop collection page lifted click-through 4.0%, a statistically significant gain, while conversion rate fell 5.9%, orders fell 8.0% and revenue per visitor fell 6.1%.
A category link on product pages raised click-through 8.7% and cut bounce 5.8% over 66,172 visitors, yet conversion rate moved -1.1% and revenue per visitor -0.3%.
The same thing happens in SaaS. Hubstaff's homepage redesign raised visitor-to-trial conversion 49% and hero form submissions 34% over 259,000 visitors in a three-month split URL test.
Trial-to-paid fell 7%, which the team attributed to factors outside the homepage.
A trial-start metric alone would have called it a clean win.
Clean Commit
GrowthRockClean CommitClean CommitWhy micro conversions mislead
- The change makes the micro action cheaper. A one-click add or a bigger button lowers the effort of the step itself. People who were never going to buy now take the step, so the rate rises without any new buyers.
- The step moves earlier in the journey. Quick-add buttons let shoppers add before they have seen the product page, where they would have learned the details that settle the purchase.
- Clicks measure attention, not intent. A layout that pulls more clicks can also pull visitors away from the path to checkout.
- The micro conversion can fall while the macro rises. Britannica's article page CTA and email capture test, run by MarketingExperiments (MECLABS), got 30% fewer clicks but 13% more free trials and 24% more paid subscriptions per unique visitor. A click-based test would have rejected the winner.
When micro conversions do track the macro
Micro conversions are not useless. In GrowthRock's sticky desktop buy box test for a supplement store, add-to-carts rose 8.6% and orders rose 7.9%, with about 2,000 orders per variation and 99% significance.
Decathlon UK's product-view count line raised add-to-cart clicks 1.5%, transactions 2.4% and revenue 1.3% across 400 product pages.
Apps show the same thing. Slopes, a ski tracking app, tested a one-tap trial quickstart card that skipped the paywall screen.
Trial starts rose 25% and trial-to-paid stayed about the same, so the extra trials were worth roughly as much as the old ones.
Checking trial-to-paid is what made the trial-start lift trustworthy.
The difference is in the mechanism. In those tests the change made the product more convincing or easier to buy, so the extra adds came from real buyers.
In the failed tests, the change mainly made the add itself easier.
Research on proxy metrics backs this up. A 2023 paper on Pareto optimal proxy metrics describes a large-scale recommendation system where the authors built proxy metrics "eight times more sensitive than the north star" that "consistently moved in the same direction."
The proxies were chosen by checking them against the outcome across many past experiments, not assumed to work.
A practical rule set
- Pick one macro metric as the primary metric before the test starts: revenue per visitor for stores, paid conversions for SaaS and apps, qualified leads for lead gen. See choosing A/B test metrics.
- Log micro conversions as secondary metrics. They explain why a test won or lost. Clean Commit's quick-add test only makes sense once you see add-to-cart up and orders down side by side.
- Use a micro conversion as the primary metric only when you have checked it. If past tests on your site show that add-to-cart and orders moved together, it may be a fair proxy for changes of the same type. If you have no such history, don't assume it.
- Be most careful when the change acts on the micro action directly. Buttons, quick-add, sticky bars and click-heavy layouts are where micro and macro split most often.
- Treat a micro win with a flat or falling macro as a loss or a draw, not a partial win. If revenue is protected by a guardrail metric, a falling macro should stop the rollout.
- On low-traffic sites, accept longer tests rather than switching to clicks. A fast answer on the wrong metric is worse than a slow answer on the right one.
When you analyze results, read the funnel from top to bottom.
A micro win that holds through checkout is a real win.
A micro win that disappears by the order step is a tell that the change moved effort around without adding buyers.
You can browse more examples of both patterns in the full test database.