Improving Conversion Rate: A Diagnosis-First Playbook
To improve your conversion rate, stop trying tactics from a list and start with one question: where do the most people leave, and why? Define a single conversion, measure it by segment, find the biggest drop-off in the funnel, watch what users actually do there, then test one fix at a time. Tactics such as shorter forms, faster pages and clearer calls to action work only when they remove the friction your own users hit.
This guide walks through that process in order, with the numbers and research worth knowing. It is written for product managers, designers, developers, UX researchers and marketers who share responsibility for the same funnel.
Start with one conversion and a clean baseline
Most stalled optimization programs fail before the first test, because nobody agreed on what a conversion is. Pick one primary conversion: a trial signup, a completed checkout, a booked demo. Then fix its denominator. Is it sessions, unique visitors, or visitors from a specific campaign? Changing the denominator halfway through is the easiest way to “improve” a rate without changing a single user’s experience.
Shopify describes conversion rate optimization as a data-driven, iterative process that raises the percentage of an audience performing a specific action. Semrush frames it similarly, as a six-step process that includes identifying where users drop off. The common thread is that you measure first and change second.
Segment before you compare
A single sitewide conversion rate hides more than it shows. Split it by:
-
device (mobile and desktop behave differently)
-
acquisition source and the intent behind it
-
new versus returning users
-
geography
-
funnel step
Then compare each segment with its own history, not with a benchmark you found in a blog post. A paid-social visitor and a visitor who typed your brand name are not the same audience, and a “good” average can sit on top of one very broken segment.
Pick supporting metrics that guard against false wins
Conversion rate alone can be gamed. Track it next to funnel-step completion, engagement, revenue per visitor, qualified-lead rate, activation and retention. If signups rise but activation falls, you have bought volume, not growth.
Find the largest leak in the funnel
Lay your funnel out step by step and compare completion at each stage. The biggest absolute drop among your most valuable segment is where to look first. A 10% improvement on a step that loses 60% of people is worth far more than polishing a step that loses 5%.
For product-led SaaS, the funnel often continues past the website: signup, first key action, invitation of teammates, first repeated use. Event data makes those steps measurable. Our guide on using event specs to fix product funnel drops shows how to define events so every drop-off is traceable, and this overview of event analytics covers how to tie event data to return on investment.
The numbers tell you where. They rarely tell you why. That is the next step.
Learn why people stall
A funnel chart says that 40% of people abandon the payment step. It does not say whether they hit a validation error, got surprised by a fee, or could not find the coupon field. You need qualitative evidence.
The most useful sources, roughly in order of speed:
-
Session recordings of users who dropped at the step you chose. Watching 15 to 20 of them usually reveals a pattern: rage clicks, dead clicks, scrolling back and forth, repeated form errors.
-
Heatmaps and click maps to see whether people notice your call to action or click on things that are not clickable.
-
Customer conversations and support tickets, which supply the words users use for their own problems.
-
Error logs, because a broken button converts at zero regardless of its color.
We cover the technique in detail in how to analyze session recordings effectively and list practical use cases for session replay in product teams. If you are new to it, our guide to using recordings to grow a SaaS is the gentler starting point.
Turn observations into hypotheses
An observation is not a hypothesis. “Users abandon the form” is an observation. “Mobile users abandon the form at the phone-number field because the format is unclear, so showing an example format and accepting any input style will raise completion” is a hypothesis you can falsify. Write each one with the segment, the evidence, the change and the metric you expect to move.
Fix the friction that research keeps finding
Once you have evidence, a handful of problem areas come up again and again. The research below is specific enough to act on.
Forms
Forms are where intent turns into revenue, and where it most often dies. Nielsen Norman Group’s web form recommendations are a good checklist:
-
Keep forms short and single-column, with labels close to their fields.
-
Do not rely on placeholder text as the only label.
-
Mark required and optional fields clearly.
-
Remove fields you can derive, collect later or omit.
-
Use radio buttons instead of dropdowns when there are only two or three choices.
-
Explain formatting rules before submission, preserve what the user typed after an error, and write specific error messages that do not depend on color alone.
The same NN/G article cites CHI research in which 78% of participants submitted successfully on the first try with forms that followed usability guidelines, against 42% with forms that violated them. That gap is a conversion gap.
Baymard’s checkout research adds detail. It reports that 64% of sites do not use input masking correctly, which matters because people enter phone numbers and card details in many formats. Baymard also recommends saving account creation for the confirmation step and avoiding multicolumn form layouts. Their work is ecommerce-focused, but the logic carries over to any signup flow: ask for the minimum, as late as you can.
Page speed
Slow or jumpy pages lose people before your message has a chance. Use PageSpeed Insights, which combines Lighthouse lab diagnostics with real-user data from the Chrome User Experience Report over the previous 28 days. Google’s current “good” thresholds, measured at the 75th percentile, are Largest Contentful Paint at or below 2,500 ms, Interaction to Next Paint at or below 200 ms and Cumulative Layout Shift at or below 0.1.
Two cautions. Lab and field results can differ, because a lab run simulates one device and network while field data reflects your real visitors. So do not declare a page fast from a single synthetic run. And divide the work sensibly: developers fix the measured bottlenecks, designers protect layout stability and responsiveness, and product and marketing prioritize by real-user segment data. For a broader view of how faster website speed and conversions connect, that is a useful next step after your first audit.
B2B and SaaS pages
If you sell to businesses, remember that the visitor is often not the buyer. NN/G’s B2B website usability research describes prospects who need to understand the problem you solve, the business benefits, technical specifications, pricing scenarios and proof, plus material that helps them justify the purchase internally. One generic “Get started” button rarely serves the engineer, the finance approver and the end user at once. Give each of them a clear path: a self-serve trial, a pricing page that answers cost questions, a technical page for integrations and security.
Pricing pages deserve their own attention, since they sit right before the decision. Our piece on how to A/B test SaaS pricing pages covers what to vary and what to leave alone.
Onboarding
For products with a free trial or free plan, the conversion that matters often happens inside the product. People who never reach their first meaningful result will not pay. Watching new users move through setup is the quickest way to see where they get lost; see our approach to improving onboarding with session replays and the related strategy for raising a product engagement score with replay.
Test one change at a time
A/B testing is how you separate real improvement from noise, but it is a tool for confirming a hypothesis, not for finding one. A few rules keep tests honest:
-
Test one meaningful change per experiment, so you know what caused the result.
-
Decide the success metric and the stopping rule before launch. Peeking and stopping on the first good-looking day inflates false positives.
-
Run the test across full weekly cycles, because weekday and weekend visitors differ.
-
Check downstream quality before shipping. A variant that lifts signups but lowers activation is a loss.
-
Record what you learn, including failures. A losing test with a clear explanation is an asset.
If you have enough traffic to run several experiments, our advanced A/B testing strategies go deeper on prioritization and design. Low-traffic teams can often get further by fixing observed bugs and obvious friction directly, then checking the funnel before and after.
A worked example
Imagine a trial signup funnel with four steps: landing page, signup form, email confirmation, first project created. Your segmented baseline shows mobile visitors completing the form at far lower rates than desktop visitors. Recordings show mobile users repeatedly tapping the phone field and getting a generic error. Your hypothesis: accepting any phone format, showing an example, and making the field optional will raise mobile form completion. You ship it as a test, track form completion as the primary metric and first-project creation as the guardrail. If both rise, ship it. If completion rises but first-project creation does not, the people you gained may be low-intent, and the next question is different.
The numbers here are illustrative, not a benchmark. Your baseline is the only benchmark that counts.
Avoid wins that cost you trust
Pressure tactics can raise a click rate. NN/G calls the worst of them “manipulinks”: guilt-tripping decline buttons and similar patterns. Their argument, in their piece on shaming users for micro conversions, is that A/B results showing more micro conversions are not proof of success, because optimizing the click while pressuring people can damage the relationship you need after it. Prefer clear, reversible choices, and judge success by meaningful outcomes such as activation and retention, not the click alone.
Our perspective
We build a session replay and product analytics tool, so we see the same pattern in team after team: people jump to testing before they have looked at a single real session. A test of button copy on a page where the real problem is a validation error on mobile will teach you nothing, however well it is run.
Our view is that conversion work needs both kinds of evidence. Quantitative data, such as funnels and events, tells you where the leak is and how big it is. Qualitative data, such as recordings and heatmaps, tells you what is actually happening there. Neither is enough alone. Funnels without recordings produce guesses, and recordings without funnels produce anecdotes about edge cases that affect three users.
We also think conversion is a shared job. Developers find the errors and slow pages, designers see the confusing layouts, researchers connect behavior to motives, and marketers know which traffic is arriving with which expectation. When all of them watch the same sessions, arguments about opinion turn into conversations about evidence. Privacy belongs in that process too: collect only what you need and mask sensitive fields, which is part of why we built compliance with GDPR and CCPA into the product.
Put the process to work with LiveSession
If you want the diagnosis steps above in one place, LiveSession combines session replay, heatmaps and click maps, funnels, error tracking and customizable product analytics dashboards, with integrations such as Intercom, Zendesk, Shopify and Segment. You can find the drop-off in a funnel, then jump straight to the sessions behind it. There is a Free plan, and you can compare all plans on the pricing page. For a wider view of the category, see our overview of tools for improving conversion in SaaS.
A 30-day plan to get moving
-
Week 1: define the primary conversion and denominator, build the segmented baseline, and check your Core Web Vitals in PageSpeed Insights.
-
Week 2: map the funnel, pick the largest leak in your most valuable segment, and watch 20 recordings of users who dropped there.
-
Week 3: write two or three falsifiable hypotheses, rank them by expected impact and effort, and fix any clear bugs immediately.
-
Week 4: launch the first controlled test with a guardrail metric, and document what you expect to learn.
Then repeat. The compounding comes from the loop, not from any single change.
FAQ
What is a good conversion rate?
There is no single answer, because rates vary widely by industry, traffic source, device and price point. Compare each segment with its own past performance instead of an industry average. A steady upward trend in a well-defined segment tells you more than any benchmark.
What are the first three things to do to improve conversion rate?
Define one primary conversion and measure it by segment. Find the funnel step with the biggest drop-off. Then watch real sessions at that step to learn why people leave, before you change anything.
How do I improve form conversion rate?
Shorten the form, use a single column, label every field clearly and remove anything you can collect later. Explain formats before submission, keep what users typed after an error, and write specific error messages. NN/G’s research links guideline-following forms to far more first-try submissions.
Do I need A/B testing to improve conversions?
Not always. Obvious bugs, broken flows and confusing forms can be fixed directly, then verified by comparing the funnel before and after. A/B testing pays off when you have enough traffic and want to decide between plausible alternatives.
How long should I run a conversion test?
Set the duration and success metric before launch, and run at least full weekly cycles so weekday and weekend behavior are both covered. Do not stop early because one variant looks ahead. Check a downstream metric such as activation before you ship the winner.
Sources
Related articles
Get Started for Free
Join thousands of product people, building products with a sleek combination of qualitative and quantitative data.