If you hire a CRO agency, what does working with a CRO agency actually looks like in the first 90 days?
At Invesp, the work usually starts with understanding the business, the data, and where customers are getting stuck. From there, the team moves into testing, learns from the results, and adjusts what gets tested next.
That process can look different from one company to another depending on things like traffic, tracking, existing research, and how quickly decisions get made. So this isn’t a rigid 90-day formula. It’s a look at how CRO engagements typically unfold based on how we work with real clients at Invesp.
Before the work starts: onboarding and access
Before anyone starts suggesting tests, your CRO agency needs to understand what it is actually trying to improve.
That starts with a kickoff. Expect to discuss how your business makes money, your main products or customer journeys, where you think customers are struggling, what you have already tried, and which numbers matter most to the business.
That last part is important. “Increase conversions” is not specific enough. You need to agree on what success will actually be measured against. It could be revenue per visitor, purchases, qualified leads, subscriptions, average order value, or something else and which metrics should not decline in pursuit of that goal.
There is good reason to settle this upfront. After running thousands of experiments across Microsoft, its experimentation team found that incorrectly interpreting metric movements could lead teams to make decisions that hurt the business by millions of dollars. They describe choosing and interpreting the right metrics as one of the major challenges in experimentation.
Give the agency access to the data it needs
Your agency will usually need access to some combination of:
- your analytics platform, such as GA4 or Adobe Analytics
- your ecommerce or CMS platform
- your A/B testing platform
- heatmaps and session recordings
- previous experiment results
- customer surveys and research
- relevant dashboards and reports
You don’t necessarily need to hand over administrator access to everything. For example, Google Analytics lets you control access at the property level and assign different roles and data restrictions. Give the team the level of access it needs to do the work without sharing permissions it doesn’t need.
And get this done early. If the agency spends its first two weeks chasing analytics access, experiment histories, or the person who owns your testing platform, that is two weeks it cannot spend understanding the problem.
Don’t hide the old work
Share previous test results, including the losing ones, along with old research, customer surveys, support insights, major site changes, and anything your team already knows about customer behavior.
Otherwise, the agency can easily spend time rediscovering something you learned six months ago or proposing an experiment you have already run.
The same applies to your tracking setup. Before an agency bases decisions on the numbers, it needs to know whether those numbers can be trusted. Microsoft calls trustworthiness one of the core requirements of its experimentation platform, which supports more than 10,000 experiments a year. Experimentation platforms themselves flag data-quality problems for the same reason: Optimizely, for example, notes that implementation errors are among the most common causes of abnormal traffic splits in experiments.
Decide who can actually make things happen
Finally, make the approval process clear from day one.
Who approves a test? Who answers analytics questions? Who can make development changes? Who needs to sign off on a new design?
Your CRO agency may find a strong opportunity quickly, but it cannot launch an experiment if a design sits in approval for ten days or nobody knows which developer can implement it.
By the end of onboarding, the agency should have the business context, access, historical data, and internal contacts it needs to start investigating where the real conversion opportunities are. That is what Month 1 is for.
Month 1: Understand the business before changing the website
The first month usually isn’t about coming up with as many A/B test ideas as possible.
Before changing anything, the CRO team needs to understand three things: how your business works, how visitors currently move through the site, and where there is enough evidence of a problem to justify testing a change.
That means the first few weeks are usually heavy on research and analysis.
First, the team needs to understand the business behind the numbers
A 3% conversion rate means very little without context.
The agency needs to understand what you sell, which products or customer segments matter most, where your traffic comes from, how often people come back before buying, what your margins look like, whether the business is seasonal, and which parts of the funnel matter most commercially.
This is also where the team pressure-tests the assumptions you bring into the engagement.
You may come in saying:
“Our checkout is the problem.”
But the data may show that too few visitors are reaching checkout in the first place.
Or you may think a product page needs a redesign when customer research shows that visitors are actually struggling to understand shipping, sizing, pricing, or the difference between products.
That distinction matters because a CRO program should not start by testing whichever idea sounds most plausible. It should start by finding evidence for where the problem actually is.
Then the team looks at what visitors are actually doing
This usually means combining several sources of evidence rather than relying on one analytics report.
Depending on the account, that can include:
- funnel and conversion analysis
- traffic and customer segmentation
- device-level behavior
- landing-page and product-page performance
- session recordings
- heatmaps
- customer surveys or other voice-of-customer research
- previous experiment results
- heuristic or expert reviews of the site
Each source answers a different question.
Analytics might tell you where visitors are dropping off. Session recordings can help show what they are doing before they leave. Customer research can help explain why they are hesitating in the first place.
The useful insights usually appear when several pieces of evidence point toward the same problem.
For example:
Analytics: visitors who reach a particular product page convert poorly.
Session recordings: visitors repeatedly move between the product information and FAQ sections.
Customer research: buyers say they are unsure which version of the product is right for them.
That is much stronger evidence for a hypothesis than simply saying, “We should redesign this page.”
The agency should also check whether the data can be trusted
Before using analytics to decide what to test, the team needs to make sure the underlying measurement is reliable.
That can mean checking whether important events are firing correctly, whether funnels are being measured consistently, whether ecommerce or lead data matches other systems, and whether existing experiments were configured correctly.
Sometimes this is a quick check.
Sometimes it uncovers a bigger problem that has to be addressed before the team can confidently judge what is happening on the site.
If something important is broken, that can change the first-month plan. The team may fix measurement first while continuing qualitative research in parallel rather than immediately launching experiments against unreliable data.
Research then becomes a list of opportunities not a list of random test ideas
By this point, the team will usually have identified multiple potential conversion problems.
But not every observation should become an experiment.
A useful CRO team should be asking:
- How strong is the evidence that this is actually a problem?
- How many visitors does it affect?
- How important is this part of the journey?
- What could the commercial impact be?
- Can we test the hypothesis reliably?
- How difficult will the experiment be to design and build?
That is how a large pool of observations gets narrowed into a smaller set of hypotheses worth pursuing first.
By the end of Month 1, you should know what the team wants to test and why
The most important output of the first month isn’t a prettier website.
It’s a clearer, evidence-backed picture of where the biggest conversion opportunities are and which ones deserve to be tested first.
Depending on the engagement, your Month 1 deliverables may include research findings, analytics analysis, identified conversion problems, experiment hypotheses, and a prioritized testing roadmap.
And each major experiment on that roadmap should have a reason for being there.
Not:
“Best practice says we should add this.”
But:
“We observed this behavior, found this problem across these data sources, and believe changing this part of the experience could improve this specific outcome.”
That is what Month 1 is supposed to establish. Month 2 is where those hypotheses start turning into actual experiments.
Month 2: Turning research into live experiments
By Month 2, the CRO team should have a clearer idea of where the biggest conversion problems are.
Now the job is to turn those findings into experiments.
That does not mean every issue found in Month 1 immediately becomes an A/B test. The team first has to decide which problems are worth testing, what change could address them, and whether the experiment can be run reliably.
A research finding has to become a clear hypothesis
Say the research shows that visitors are hesitating on a product page because they do not understand the difference between two product options.
The test idea should not simply be:
“Let’s redesign the product page.”
The team needs a more specific hypothesis:
If we make the differences between the two options easier to understand, more visitors will feel confident choosing a product and moving toward checkout.
That hypothesis can then become one or more actual test concepts.
This is why Month 1 matters. The strongest experiments should be tied back to something the team observed in the data or customer research rather than being based on a list of CRO “best practices.”
The team decides what gets tested first
You may have dozens of possible experiments by this point. You cannot run all of them at once.
The team therefore has to prioritize.
That usually means looking at questions such as:
- How strong is the evidence behind this problem?
- How many visitors does it affect?
- Where does it sit in the funnel?
- How much business impact could fixing it have?
- Is there enough traffic to test it properly?
- How difficult will the change be to build?
- Are there technical, brand, or business constraints?
A high-impact idea may move down the roadmap if it would take weeks to develop. A smaller change may move up if the evidence is strong and the team can learn from it quickly.
Then comes the part clients often don’t see: getting the experiment live
Even a fairly simple experiment usually passes through several steps before customers see it.
The exact process will vary, but it can include:
Hypothesis → test concept → design → client review → development → QA → analytics check → launch
This is also where timelines can start to vary between clients.
If designs are approved quickly and the experiment is technically straightforward, the test may move fast. If it requires major development work, legal approval, several stakeholders, or changes to the tracking setup, it can take longer.
What you should see during Month 2
At this point, the work should start becoming much more visible.
Depending on the engagement, you may see:
- prioritized experiment briefs
- hypotheses and supporting research
- wireframes or test designs
- experiments entering development
- QA and tracking checks
- the first tests going live
But the important thing is not simply the number of tests launched.
You should be able to see why each test is being run, what customer problem it is supposed to address, and what the team expects to learn from it.
That becomes especially important in Month 3, when the first results start coming back.
Month 3: Results start shaping the next round of work
By Month 3, some of the first experiments may have produced results.
This is where a CRO program should start becoming more iterative.
The roadmap created earlier is not supposed to be a fixed list that the agency works through regardless of what the experiments show. Each result gives the team new information about customers, the site, and the original hypothesis.
And that information should influence what gets tested next.
A winning test is only one possible result
An experiment can win, lose, or come back inconclusive.
A win may show that the original hypothesis was correct and give the team a reason to explore the same problem further.
A losing test may show that the proposed solution did not address the problem — or that the original assumption itself needs to be reconsidered.
An inconclusive test may mean the effect was too small to detect, the test needed more traffic, or the change simply did not matter enough to customer behavior.
What matters is what the team does with the result.
The next test should reflect what the previous one taught you
Imagine research suggested that customers were hesitant because important delivery information was difficult to find.
The team tests a more prominent delivery message, but conversion does not improve.
That does not necessarily mean delivery concerns are irrelevant.
The result may push the team to look deeper. Perhaps customers care more about the delivery date than the shipping price. Perhaps the concern only affects first-time visitors. Perhaps the message appeared too late in the journey.
The next experiment should become more precise because the team now knows more than it did before the first test.
That is how a CRO program should evolve:
Research → hypothesis → experiment → result → learning → next hypothesis
The original roadmap may change
This is also why the testing roadmap from Month 1 should not be treated as permanent.
An experiment may reveal that an opportunity is much larger than expected and deserves several follow-up tests.
Another area may turn out to matter less than the research initially suggested.
The team may also discover a completely new problem while analyzing the results.
For you as the client, this is a good sign. You do not want an agency blindly executing ideas it decided on three months ago. You want it using new evidence to decide what deserves attention next.
What you should see during Month 3
By this stage, depending on traffic and how quickly experiments can run, you may start seeing:
- completed experiment results
- analysis of what those results mean
- recommendations for winning variations
- follow-up hypotheses
- changes to the testing roadmap
- new experiments entering design or development
You should also be able to trace the logic from one round of work to the next.
The question is no longer simply:
“Did this test increase conversion?”
It becomes:
“What did we learn about customers, and what does that tell us to investigate next?”
By the end of Month 3, that learning loop should be taking shape. That is what ultimately separates an ongoing CRO program from a collection of disconnected A/B tests.
