What does a CRO agency do?
It studies why visitors leave without acting, fixes what it finds and, when you have the volume, proves each fix with a test. CRO means The share of visitors or clicks that take the step you want, such as calling, booking or sending a form. More optimization: raising the share of visitors who call, book or buy.
| Service | What you get | Needs lots of leads? |
|---|---|---|
| Tracking setup | Calls, forms and bookings counted as conversions | No |
| Research and audit | Session recordings, heatmaps (maps of where people tap and scroll), surveys, a ranked list of fixes | No |
| Page and form fixes | New headline, offer, layout, shorter form | No |
| A/B tests | Two versions shown to split traffic, and proof which wins | Yes |
| Personalization | Different content for different visitor groups | Yes |
| Offer and price tests | Which package or offer sells | Yes, unless the change is big |
The top half needs no traffic threshold; the bottom half does.
For a service business, the job isn't done when the form is sent. If nobody calls back fast, page fixes buy more unanswered leads. Ask whether the agency's work goes past the form; if not, check speed to lead and missed calls first.
CRO agency pricing: what does it cost?
No regulator or platform publishes a rate; the figures online come from the industry itself. Testing-software maker Convert, drawing on agency owners and one agency's own figures, reported these prices (Convert, updated April 2026):
| Pricing model | Reported range | What to check |
|---|---|---|
| Monthly retainer | $2,000 low, $16,000 average, $30,000 premium | Tests per month; developer hours included; minimum term |
| One-time conversion audit | $2,500 to $4,500 | Who makes the fixes, and at what price |
| Hourly or per test | No clear current figure | A monthly cap |
| Pay tied to results | A base fee plus a share of the lift | What counts as lift, against which baseline, over which months |
Tools come on top. Some are free: Microsoft says "Clarity is a free service forever" with no traffic limits (Microsoft Learn: Clarity FAQ), and Google Analytics is "free of charge" (Google Marketing Platform). Paid testing platforms can charge by traffic: Wingify (VWO and AB Tasty, now merged) quotes custom prices "based on your Monthly Active Users" (Wingify: Pricing).
To judge a quote: monthly fee ÷ profit per job = the extra jobs a month you need to break even.
Do you have enough traffic for A/B testing?
Count leads, not visitors. Experimentation researchers Kohavi and Longbotham recommend "the number of users available for a test be at least in the thousands" (2023), but how many of them convert sets how long a test runs. The standard sample-size formula (Kohavi et al., 2009, at 95% confidence and 80% power) works out to about 400 leads in each version, roughly 800 in all, to prove a 20% lift (say, 6% of visitors converting instead of 5%). A 10% lift needs about 1,500 per version; a 50% lift, about 60.
| Leads a month from the tested page | One 20%-lift test takes | What fits |
|---|---|---|
| 15 | About 4 years | No testing. Fix what recordings show |
| 50 | About 16 months | Test only big changes: offer, price shown, form length |
| 200 | About 4 months | Big changes, plus a one-time audit |
| 800 or more | About a month | A steady test program, from an agency or a tool |
Run tests in whole weeks: "at least 1 week (to capture a full weekly cycle)", then longer (Kohavi and Longbotham, 2023). Small sites "can run A/B tests to detect moderate or large changes" (same source), so test bold changes, not button colors.
The top rows are real. In October 2026, at an accounting firm we run marketing for, paid ads brought 2,689 visitors and 13 bookings (0.5%). At that pace, one 20%-lift test on its ad pages would take about five years. For ad landing pages, Google Ads itself can split ad traffic between two pages: see landing page best practices.
What results can you expect from a CRO agency?
Fewer wins than case studies suggest. Across 173,000 experiments it analyzed, Optimizely found:
"Only 12% of experiments produce a statistically significant improvement on the primary metric." (Optimizely: Top 10 takeaways from running 173,000 experiments)
At Microsoft, only one third of ideas tested improved the metric they were built to improve (Kohavi and Longbotham, 2023). Optimizely adds: "If nine out of ten of your tests are winning, something is off." A fair agency promises a process and a written log of what it learned, not a lift.
Without volume, you can still find what's broken. The same firm logs every step of its booking calendar. What stopped people was the page around it: a cookie notice covering half a phone's first screen, a "Book" button below the fold on short laptop screens after picking a time, ads landing on a bare calendar. In the weeks after the fixes, people who picked a time and finished booking went from 11 of 20 to 18 of 20 (the files can't separate the fixes from more decided visitors). No A/B test was needed to find these problems.
CRO agency or AI tools?
AI tools now do much of the research and test-watching cheaply. They can't make the leads a test needs, and a person still has to check what they conclude.
| Job | Free tools | AI in paid test platforms | Agency |
|---|---|---|---|
| See where people stop | Clarity recordings and heatmaps; its Copilot summarizes heatmaps | Same, built in | Does it and explains why |
| Draft new versions | You, or a chatbot | Suggests new wording | Writes them with you |
| Run the test | Google Ads, for ad pages only | Yes, priced by traffic | Uses one of these platforms |
| Call the winner | You | Can act "within your configured guardrails" | A person reviews |
The vendors name the limits themselves:
- Summaries aren't advice. Microsoft: "takeaways from Copilot in Clarity shouldn't be treated as expert recommendations" (Microsoft Learn: Clarity FAQ).
- Simulated visitors only rank ideas. Wingify's Synthetic AB Test uses "AI digital twins" to rank ideas "before committing live traffic" (Wingify: Pricing).
- "Real-time" optimizing isn't faster proof. Multi-armed bandits shift traffic to whichever version leads, but they "do not generate statistical significance", and "MAB-driven experiments generally take longer to find winners and losers than A/B tests" (Optimizely docs: multi-armed bandit optimizations).
Questions to ask a CRO agency before you sign
| Ask | Good answer | Walk away if |
|---|---|---|
| How many leads does my page need for your tests? | They do the math with your numbers | Weekly tests promised on 30 leads a month |
| What counts as a win? | Booked jobs or qualified leads | Clicks, time on page, form starts |
| How long does a test run? | A sample size set in advance, in whole weeks | Winners called after a few days |
| How many of your tests win? | Most don't; here's our log | "Nearly all of them" |
| Will tests hurt my Google rankings? | They follow Google's testing rules | Google sees different pages than visitors |
| Whose name are the testing, analytics and recording accounts in? | Yours | Theirs |
| Do recordings capture what people type? | No, form fields are masked | "We see everything" |
| Who changes the site? | Named, with developer hours in the fee | A new bill per change |
Google's testing rules: show Googlebot the same pages people see, put rel="canonical" on test URLs pointing to the original, use 302 (temporary) redirects rather than 301, and remove test pages and scripts when done:
"If we discover a site running an experiment for an unnecessarily long time, we may interpret this as an attempt to deceive search engines and take action accordingly." (Google Search Central: Minimize A/B testing impact in Google Search)
Clarity masks input boxes "in all modes" (Microsoft Learn: Clarity masking); ask the same of any other tool. Clinics: tracking on a clinic website. Contracts: marketing agency contract terms.
Example
Illustrative cases, not real businesses. Prices in USD.
A house cleaner with 40 bookings a month
The quote: $4,000 a month for "continuous A/B testing". One 20%-lift test would take about 20 months. She buys a $3,000 audit instead, installs Clarity, fixes the three problems it names, and starts answering web leads within five minutes.
A dental group with 900 booking requests a month on one page
That's enough for about one 20%-lift test a month. At $500 profit per new patient, an $8,000 retainer needs 16 extra patients a month to break even. The group signs for six months, measured by booked first visits, not form starts.
An HVAC company offered "pay only for results"
The agency wants 20% of revenue above a spring baseline. Summer brings more calls with no change at all. The owner asks for the baseline from the same months last year, with a share of visitors still seeing the old page.
FAQ
Is a CRO agency worth it for a small business?
Often as a one-time audit of research and fixes. A testing retainer pays only when the tested page brings hundreds of leads a month.
How long does CRO take to show results?
Fixing something broken can help from day one. Each A/B test needs at least a full week, usually several, before a winner can be called.
Can AI replace a CRO agency?
Partly: AI summarizes recordings and watches tests, some free. It can't create the leads a test needs. Microsoft says Clarity's AI summaries aren't expert recommendations.
Does A/B testing hurt SEO?
Rarely, if it follows Google's rules: Google sees what visitors see, test pages point to the original, redirects are temporary, and the test ends promptly.
Different for you?
- Visitors arrive but don't book: where they drop off
- One ad The page people reach after clicking an ad or link, built to get one action, such as a call or a booking. More to fix: landing page best practices
- Not sure calls and forms are counted: GA4 key events
- Leads arrive but don't become jobs: speed to lead or tracking a lead to a booked job
- Weighing AI tools in general: AI agents for small business marketing
- Want it done for you: Website Agent or Conversion Agent