Every conversion proposal contains a number, and the number is always someone else's result. What you are actually buying is a team's capacity to form hypotheses, build variants, run them long enough to mean something, and ship the winners. Whether that capacity produces lift on your site depends almost entirely on how much traffic the pages get, which is a fact about your business rather than a fact about the agency, and it is the first thing to settle.
Traffic decides whether testing is even possible
A split test needs enough conversions in each arm to separate a real difference from noise, and a page with a few dozen conversions a month will not produce one in a quarter. Below that floor the honest programme is not testing at all: it is research, usability work, analytics repair and a small number of changes made on judgement and shipped. An agency that proposes a testing roadmap without first asking what your conversion volume is has not looked.
Ask where the hypotheses come from
The difference between a programme that works and a list of button colours is the input. Session recordings, form abandonment, search logs, support tickets and actual customer interviews produce hypotheses about why people leave; a best-practice checklist produces guesses. Ask to see the research artefacts from a recent engagement, with the client's name removed, and check that the tests trace back to something observed rather than to a blog post.
Who writes the code, and who carries the risk
Client-side testing tools inject changes with JavaScript after the page loads, which is fast to deploy and can cause a visible flicker, slow the page and break under a strict content security policy. Server-side testing avoids that and costs engineering time. Decide which you are buying, who implements winners permanently, and whether your developers have capacity to ship them, because an untested winner left in the testing tool is a liability rather than a gain.
Agree how a result is called before the first test runs
Stopping a test the moment it looks good is the most common way these programmes produce numbers that never appear in revenue. Agree in advance the minimum runtime, the primary metric, how many whole business cycles a test must cover, and what happens when a test is flat, which most will be. A supplier who will commit to that in writing is telling you something useful about how they report.
Claims made on the page still have to be true
Optimisation frequently reaches for urgency, scarcity and social proof, and these are the elements most likely to drift into claims you cannot support. Countdown timers that reset, stock counts that are not real and testimonials without a basis are advertising claims like any other. Put a rule in the engagement that anything factual added to a page has to be substantiated by someone in your business before it ships.
Questions people ask about conversion optimization services
How much traffic do we need before conversion testing makes sense?
The usable measure is conversions rather than visits: you need enough of them per variant, per test, to distinguish a modest improvement from ordinary week-to-week variation, which for most sites means hundreds a month on the page being tested. Below that, spend the budget on research and on fixing what is visibly broken instead.
Should conversion work be a retainer or a project?
A research and fix project suits a site that has never been looked at, because the first pass usually finds obvious problems that need no testing to justify. A retainer suits a site with the traffic to run a continuous test programme. Buying a retainer for a low-traffic site is how these engagements end in disappointment on both sides.
Will a conversion agency guarantee a lift?
Some will, and the guarantee is generally structured so that it is met by the definition of the metric rather than by money arriving. If a guarantee matters to you, insist that it is written against a revenue figure your finance team already reports, over a period long enough to include a full sales cycle, and read what happens if it is missed.
Does conversion work affect SEO?
It can, in both directions. Changes that slow the page or push content behind interactions can hurt; changes that make the page clearer and faster usually help. Ask that page speed be measured before and after every shipped winner, so a conversion gain that costs you search traffic is visible rather than discovered a quarter later.