What is a "good" conversion rate?

It depends enormously on industry, traffic source and what you call a conversion. Compare yourself with yourself: same page, same source, last month versus this month. Industry benchmarks are indicative, not targets.

Do I need an A/B testing tool?

If you have fewer than a few hundred conversions a month, no. You need session recordings, correctly configured GA4 and the discipline to measure before and after every large change.

With the definition of a conversion and correct tracking. Without them, any optimisation is guessing. Then: speed, clarity, trust, friction — in that order.

CRO fundamentals: from traffic to customers, without guessing

The situation

You get 2,000 visitors a month and 12 quote requests. Someone tells you "you need more traffic". Maybe. But if only 12 out of 2,000 people take the next step, you have a website that loses 1,988 people a month. More traffic into a leaking site is more leaking.

CRO — conversion rate optimisation — means turning a larger share of the visitors you already have into customers. It is usually the cheapest growth available, because you do not pay for every additional unit of result.

What a conversion is for you

The first mistake: measuring what is easy instead of what matters. "Button click" is not a conversion. A conversion is the moment a visitor becomes something valuable for the business:

Clinic — a confirmed appointment (not a submitted form; confirmed).
Car repair shop — a phone call or a booking request with valid contact details.
B2B firm — a qualified quote request, not a newsletter signup.
Online store — an order placed and paid.

Define one primary conversion and, optionally, one or two micro-conversions (pricing page view, WhatsApp click) that show intent. Everything else is noise.

The funnel: from landing to customer

Every visitor goes through, consciously or not, a few steps:

Lands on a page — from an ad, from Google, from a link.
Understands within seconds whether they are in the right place.
Believes you can help them (proof).
Acts — calls, fills in, buys.
Completes — the payment goes through, the appointment is confirmed.

You lose people at every step. CRO means finding out where you lose the most and why.

The 5 usual leaks

In practice, almost every site leaks in the same places:

Unclear offer — the headline says "Complete solutions for your business". The visitor does not know what you do, for whom, or what happens next. They leave in 5 seconds.
Slow page — a 5-second LCP on mobile. Some people leave before they see anything. (The Core Web Vitals lesson covers exactly this.)
Form friction — 9 fields, a mandatory company registration number, a captcha that does not work on phones, no confirmation message.
No trust signals — zero reviews, zero real names, zero photos of the team or the place, prices hidden for no reason.
Wrong traffic — the ad promises "free repairs", the page sells a subscription. Or broad keywords bringing people who are not looking for what you offer. No page optimisation fixes wrong traffic.

Qualitative before quantitative

GA4 data tells you where people leave. It does not tell you why. For the "why" you have two cheap and extremely effective tools:

Session recordings and heatmaps — free tools exist. You watch 20 real sessions on the page that leaks. You see people scrolling in search of the price, trying to tap something that is not a button, abandoning the form at the same field.
5 customer conversations — you call 5 recent customers and ask: "What made you choose us? What made you hesitate? What did you look for on the site and not find?". Five 10-minute calls give you more good ideas than a month of reports.

Only once you know why do you measure how much.

When an A/B test makes sense

Honestly: most small and medium businesses cannot run statistically valid A/B tests. A test needs enough conversions in each variant for the difference not to be noise — we are talking hundreds of conversions per variant, within a reasonable period. With 12 quote requests a month, a test would take a year and you still would not know for sure.

What to do instead:

Large sequential changes — fix the 5 leaks, compare the conversion rate on the same traffic source, before and after, over at least 4 weeks. It is not a controlled test, but for large changes the signal is clear.
Test only when you have volume — online stores with hundreds of orders a month can test the checkout or the product page. Services with dozens of leads cannot.

Prioritising: impact × ease

You always have more ideas than time. Put each idea in a simple table: how much impact it would have if it works (1–5) and how easy it is to do (1–5). Multiply. Start with the highest scores. A clearer offer at the top of the page: impact 5, ease 4 = 20. A full redesign: impact 4, ease 1 = 4. The order is obvious.