Conversion optimization agency: the four kinds, the twelve questions, and how to judge the first ninety days

A conversion optimization agency is hired to raise the share of your visitors who do the thing you want: buy, enquire, book, subscribe. The good ones work as a testing programme rather than a redesign: they research where visitors drop out, form a hypothesis about why, run a controlled test, and keep the version that wins. Before you shortlist anybody, know the three numbers that decide whether an agency can help you at all: your monthly traffic on the page you want to improve, your current conversion rate, and the value of one conversion. Those three set the size of test you can run and therefore what kind of agency is worth paying.

This page covers buying and scoping the engagement. For running the work yourself, including the test sizing arithmetic, see conversion rate optimisation.

The four kinds of provider

Kind	What they sell	Suits
Testing programme agency	Research, hypotheses, build, run and analyse a continuous series of tests	Sites with enough traffic for statistically readable tests
Research and UX consultancy	User research, session analysis, usability testing, a prioritised list of changes	Lower traffic sites, and anyone who needs to know why before testing
Design and build studio	Redesigned pages and funnels, implemented	Teams who know what to change and need it built well
Performance or growth agency	Traffic and conversion together, usually alongside paid media	Teams buying acquisition and landing pages as one job

Being clear which one you are buying prevents the common failure: hiring a testing agency for a site without the traffic to read a test. On lower traffic sites, research and build work delivers more, and that is a genuine strength of the consultancy model rather than a consolation.

Bring your three numbers to the first call

• Monthly visitors to the target page or funnel step. Not site wide traffic. The page you want improved.
• The current conversion rate at that step, measured the same way every month.
• The value of one conversion, as average order value or as a qualified enquiry's worth to your sales team.

With those, any competent agency can tell you in the first call what size of improvement a test could detect in a reasonable window, and what a month of their work would need to produce to pay for itself. An agency that will not do that arithmetic with you is selling activity.

The twelve questions

• Show me a test that failed and what you learned. A programme with no losing tests is not a testing programme. This is the single most revealing question.
• What is your research process before a hypothesis? Look for analytics, session recordings, form analytics, customer interviews and support tickets. Hypotheses from opinion are redecoration.
• How do you decide what to test first? Ask for the scoring method, and ask them to apply it to your funnel on the call.
• How do you size a test? They should talk about baseline rate, the smallest effect worth detecting, and the run time that follows. If the answer is a fixed two weeks for every test, that is a calendar rather than a method.
• When do you call a test? Agree the rule in advance: a predetermined sample and duration, full business cycles, no stopping early because it looks good.
• Who builds the variants? Their developers, or yours? Build capacity is the usual bottleneck in a stalled programme.
• Which testing platform, and who pays for it? Get the licence cost separated from the fee, and confirm the account is in your name.
• How do you handle tests that affect page speed or SEO? Client side testing tools can add load time, and variants need handling that keeps one canonical version indexable.
• What do we get each month? A specific count: tests launched, tests concluded, research sessions, a written readout per test.
• Who owns the data, the test archive and the research? You should, in writing, with an export at the end.
• What do you need from us? Developer hours, design review, legal sign off, analytics access. An agency that claims to need nothing has not run this before.
• What would make you tell us to stop? Honest answers include insufficient traffic, a product problem rather than a page problem, or a pricing issue no test can fix.

Fee models, and what each one rewards

• Monthly retainer. The standard. Specify the minimum number of tests launched and concluded per month, and what happens to unused capacity.
• Project fee. Right for a defined piece: a research audit, one funnel rebuilt, a platform implementation.
• Retainer plus performance bonus. Workable when the baseline and the measurement method are agreed in writing before work starts. Specify who calculates it and from which system.
• Pure performance. Rare and heavily qualified. Read the measurement clause closely, because the definition of an incremental conversion is where the disagreement will be. For how these arrangements are structured elsewhere, see pay per lead agencies.

Warnings worth taking seriously

Google's own guidance on hiring search and marketing help is blunt about promised outcomes: it states that no one can guarantee a number one ranking on Google, and advises caution with anyone who claims to guarantee rankings, alleges a special relationship with Google, or advertises a priority submit. Treat an uplift percentage promised before any research the same way. An agency that commits to a specific percentage improvement before it has seen your funnel is pricing a sales line.

Four more things to check:

• Case studies with a percentage and no base. A 300 per cent lift on 40 conversions a month is 120 conversions. Ask for the absolute numbers and the test duration.
• Tests called in three days. Ask for the sample size and the run time on every case study they show you.
• A redesign sold as optimisation. Changing everything at once means you learn nothing about which change worked, and you carry the risk of all of it at once.
• Nobody asking about your margins. An agency optimising checkout completions without knowing which products make you money can raise the rate and lower the revenue.

How to judge the first ninety days

• Days one to thirty: analytics and tracking verified, a research readout naming the biggest drop off points with evidence, and a prioritised hypothesis backlog. One test live by day thirty is a good sign.
• Days thirty to sixty: tests running and being concluded with written readouts, including the losers. A shared dashboard showing the baseline metric and every test's status.
• Days sixty to ninety: at least one implemented winner in production, a measured effect on the baseline, and a backlog reordered using what the first tests taught.

Agree the scorecard before you sign: tests launched, tests concluded, win rate, the cumulative effect on the primary metric, and the revenue or enquiry value attributable to implemented winners. For a full proposal comparison framework across any agency type, see before you hire a marketing agency, and for metric choice, the KPIs worth reporting.

What people ask about conversion optimization agency

How much does a CRO agency cost?

Fees are set by the number of tests and the build capacity included, so the figure differs widely by scope and market. Judge it against your own arithmetic instead: the fee divided by the value of one conversion gives you the number of extra conversions a month the engagement needs to produce, and that number tells you immediately whether the scope is sensible.

How much traffic do I need?

Enough for the test to detect the size of improvement you care about in a reasonable window, which depends on your baseline rate and the effect size. Below that level, buy research, usability testing and build work, which improve the page using evidence rather than split traffic.

How long before we see results?

Expect research in the first month, tests from the first or second month, and implemented winners affecting the baseline from around the third. A programme is judged over two or three quarters, because a single test is one data point.

Agency, freelancer or in house?

A freelance specialist suits a single funnel. An agency suits a continuous programme needing research, design, build and analysis together. In house wins once the backlog is permanently full, and most teams get there by hiring the person who ran their agency programme.

Should the agency have our analytics access?

Yes, read access to analytics and the testing platform, under accounts you own, with their users added rather than a shared login. Keep the admin rights and the billing in your name so the test archive stays with you.

What should be in the contract?

Minimum tests launched and concluded per month, the test calling rule, named people on the account, ownership of data and the test archive, licence costs separated from the fee, a notice period you can live with, and an exit handover that includes every readout.

The visitor who arrives already persuaded

The numbers you bring to the first call describe people who reached your site and then did or did not convert. Some of your best buyers have already been through a conversation with an assistant before they land, and they arrive with a name in mind and fewer questions.

That changes what a conversion test can find, because the page is confirming a decision rather than making one. Ask the assistants which company they would recommend for your service and note whether yours is named. If it is not, no landing page test will fix it. The platform at AI Knows Us reports which companies each assistant names, which gives the agency a fourth number to be held to.