CRO agency vs in-house vs tools: what should you hire in 2026, and what does each really cost?

By Zach Wennstedt, founder · Updated November 6, 2026 · 10 minute read

What does each option cost?

CRO pricing is murky because most vendors quote, not publish. These are the figures we could open on November 6, 2026, with the source and its limits next to each.

Two honest notes on those numbers. First, the agency figures come from a single agency’s published guide, so they’re indicative, not an independent survey; we couldn’t find a neutral one. Second, our own programs are cheaper: our CRO services page says most programs run $500 to $3,000 a month, because we mostly work with local and mid-size businesses that don’t have enterprise traffic. Your price should follow your traffic and scope, so treat any quote far outside the range for your size as a question to ask.

The real in-house cost is higher than the salary line. One specialist can research and prioritize, but usually can’t also write the copy, build the variant, do the analytics setup and run the statistics at the same speed. Either you add a developer and designer part-time, or your test count stays low.

How much traffic do you need first?

Traffic decides more than budget does. A test needs enough visitors in each version to tell a real difference from noise. Optimizely’s documentation gives a worked example: with a 15% baseline conversion rate, a 10% relative lift to detect and 95% significance, you need about 8,000 visitors per variation. That is a high baseline. Most service-business pages convert well below 15%, and smaller lifts need far more traffic, so a page with a few hundred visits a month can’t support a clean A/B test at all.

That doesn’t mean CRO is useless for a small site. It means the work shifts from testing to fixing: speed, a visible phone number, a form that works on a phone, clearer headlines, a faster reply to every lead. Those fixes don’t need a statistical test to justify them. ClicksGeek’s guide ties its entry agency tier to about 20,000 or more monthly visitors, which matches our experience that testing programs make sense somewhere above that and that below it you’re better off spending on a fix list than a testing tool.

It also helps to know the odds. In Optimizely’s analysis of about 173,000 experiments, only 12% produced a statistically significant improvement on the primary metric. That is a vendor’s dataset from its own platform, but it tells you something no pricing page will: most tests don’t win, so the value of a program is mostly learning faster and stopping bad ideas cheaply. Whoever you hire, budget for volume and patience.

What are tools good at, and what aren’t they?

Good at: showing you where people stall (heatmaps, recordings), splitting traffic cleanly, doing the math, and being cheap compared with a person. Free analytics plus a $299-a-month testing plan is the lowest-cost way to get real answers on a busy page.

Not good at: deciding what to test. A tool will happily run a button-color test that can’t matter. It doesn’t know your sales process, can’t write the better headline, and won’t notice that your best leads come by phone. Someone has to turn behavior data into a hypothesis.

One more caution: free testing isn’t what it was. Google Optimize, the free tool many small teams used, ended on September 30, 2023, and Google’s help center says experiments still running on that date ended. If an article tells you to use it, it’s out of date. Whatever you choose, keep your data and test records somewhere you control, so a vendor shutting down can’t delete your history.

What does an in-house hire do well?

• Context. They know your product, sales calls and seasonality, and sit next to the people who answer the phone.
• Speed and continuity. No briefing cycle; tests ship when the idea is ready.
• Ownership. The knowledge stays when a contract ends.

Where it goes wrong: one person is a single point of failure, and the skills needed (research, copy, design, analytics, development, statistics) rarely live in one resume. Without access to a developer, the hire spends the year waiting in a queue. Without enough traffic, they run three tests, none conclusive, and the role looks like a mistake. And a salary of about $73,000 is a fixed cost whether tests win or not. If you do hire, hire for research and judgment first, and make sure the role has a named developer, a real analytics setup and a seat in the room where offers get decided.

What does an agency do well?

• Breadth without hiring. Research, copy, design, development and analytics are available from day one.
• Pattern recognition. A team that has seen hundreds of pages spots common leaks fast.
• Flexibility. You can scale the engagement up for a launch and down afterward.

Where it goes wrong: a retainer pays for activity, so make sure you can see outcomes. Onboarding takes weeks. Agencies often have limited access to your CRM, call tracking or sales data, so they optimize form fills instead of booked jobs. Some run “CRO” that is really a redesign. And a few treat your accounts, tracking and test code as theirs. Our guide on how to choose a CRO agency goes through these in more depth.

Which should you choose?

This is our rule of thumb, not a study. It follows from the traffic math above and the costs in the table.

If you aren’t sure where you fall, a free conversion audit will tell you what the site actually needs before you commit to a model.

What should you ask before you sign?

1. Who does the work? The person pitching, or a junior you haven’t met?
2. What do you measure? Calls and booked jobs, or just form submissions?
3. What happens when a test loses? Since most do, a good answer includes what you learned and what you test next. Ask to see a losing test.
4. What traffic do you need? A vendor who will test a 300-visit page for you is selling something other than answers.
5. Who owns the accounts, tracking and test code? It should be you.
6. How do we exit? Month to month, or a long lock-in?
7. What do you need from us? Developer time, CRM access, approval speed. The honest answer is never “nothing.”

For whichever route you pick, start with a baseline: your current conversion rate by device, your cost per lead and your lead-to-customer rate. Without that, you can’t tell whether you got your money’s worth.

What should you do this week?

1. Install a free behavior tool and watch 20 recordings of real visitors on your top landing page.
2. Write down your monthly visitors, conversion rate and value per lead, so you know what a 10% lift is worth.
3. Use the table above to pick a model, then ask two or three providers the seven questions.