What Your CRO Agency Needs From You Before Week One

Five things: platform and analytics access, context your data cannot provide, named decision-makers with a feedback window, a realistic answer on who implements winners, and the constraints nobody thinks to mention until a test is already designed.

Access is the easy one. The other four are what actually determine whether month one produces a hypothesis backlog or a status call about why nothing has started.

What to Prepare, and What Happens Without It

The first row is the one everyone prepares for. The fourth is the one that quietly kills programmes in month four.

1. Access

Request these before the kickoff call, not during it. Permissions in larger organisations routinely take a week.

Two things worth checking rather than assuming:

Who owns the GTM container. On a surprising number of Shopify stores it is a previous agency, and nobody has the login. This is a two-week problem if discovered in week one.

Whether a server-side container exists. If someone set one up and left, it may still be firing events nobody is accounting for.

2. Context Your Data Cannot Provide

Analytics tells us what happened. It does not tell us what matters to you.

Margin structure by product or category. A test that lifts conversion on your lowest-margin line while cannibalising a higher-margin one is a loss that looks like a win. Without margin context, we will optimise for revenue and you will discover the problem at the quarter end.

What you are actually trying to grow. Order volume, average order value, new-customer acquisition, repeat rate and delivered-order rate pull in different directions. A programme optimised for one can work against another.

Operational constraints that shape the funnel. Which pincodes you serve, where COD is available, what your realistic dispatch windows are. A test proposing faster delivery on prepaid is worthless if operations cannot support it.

Upcoming changes. A replatform, a brand refresh, a new product line, a pricing change. Any of these invalidates a test running through them, and we would rather sequence around it than discover it mid-flight.

3. A Named Approver and a Feedback Window

The least glamorous item on this list and the one with the largest effect on output.

A test cycle contains at least two approval points — the hypothesis and the variant design. Most retainers specify a feedback window, commonly around three business days, after which timelines shift.

The arithmetic is unforgiving. Convert.com's guidance on test duration is a minimum of two full weeks per test regardless of when significance appears. A variant waiting a week for sign-off has lost half a test's runtime before it launches. Over six months that is comfortably a full test cycle — at typical D2C volumes, a quarter of your annual testing capacity.

What to establish before week one:

• One named approver per decision type — hypotheses, design, copy, technical implementation

• A committed turnaround, even if it is five days rather than three

• A deputy for when that person is unavailable

• Whether a wider stakeholder review is required, and if so, how it runs in parallel rather than sequentially

An approval process involving four people who have not seen the hypothesis document is not a review. It is a delay with a meeting attached.

4. Who Implements the Winners

The question most engagements leave ambiguous until it becomes a problem.

A test produces a winning variant. Shipping it permanently is usually a development task — the variant was built in a testing platform, not in your theme. Someone has to make it real.

Three honest answers, each with a consequence:

Your engineering team. Fine, if they have genuine capacity. If they are fully committed to product roadmap, winners queue up and the compounding that justifies the whole programme never starts.

The agency. Check whether this is in scope. Variant implementation for a test and permanent implementation in your theme are different line items in most agreements — our post on what's actually in a CRO retainer covers where that boundary usually sits.

A third-party development partner. Workable, but add their turnaround to every cycle.

Decide before week one. A programme with three unshipped winners in month four is not a programme, it is a research project.

5. Constraints and No-Go Areas

Tell us what is off the table before we design around it.

Brand guidelines. Colours, typography, tone, imagery rules, and whether they are genuinely fixed or merely preferred.

Legal and compliance. Claims that cannot be made, disclosures that must appear, regulated language. For brands in regulated categories this can rule out entire classes of test.

Peak trading periods. Festive corridors, sale windows, planned campaign spikes. Tests running through these measure a different audience and the results will not hold afterward. We will plan around them if we know.

Development freeze windows. Common before peak season and genuinely disruptive if discovered mid-cycle.

Pricing. Whether price and discount structure are testable at all, or owned elsewhere in the business.

A designed test rejected at approval for a constraint nobody mentioned costs the same as a designed test that loses — except it teaches you nothing.

6. What Has Already Been Tried

Most brands have tested something. Very few have a record of it.

What helps:

• Previous test results, including the ones that lost. A losing test is an eliminated hypothesis, and re-running it wastes a slot.

• Changes shipped without testing, and roughly when. These show up as unexplained steps in historical data.

• What a previous agency concluded, and whether anyone believed it.

• Known problems you have chosen not to fix, and why. Usually there is a reason, and it is faster to hear it than to rediscover it.

If no record exists, say so. That is a normal starting position and the first month's research accounts for it. What does not work is a vague sense that "we tried something like that before" surfacing after the test is built.

The One Thing You Can Do Before Anyone Starts

Check whether your conversion data is accurate.

Pull your GA4 purchase event count and your platform order count for the same 30-day window and compare them.

This matters more than anything else on the list. Every test is measured against your conversion data. An inflated baseline makes sample size calculations too small, so tests appear conclusive before they are — and the winners do not hold in production.

Finding this in week zero saves a month. Finding it in month four means re-running everything. Our GA4 ecommerce tracking audit guide covers how to run the check, and our post on why CRO results take longer than brands expect covers what it costs when nobody does.

Pre-Kickoff Checklist

• All access granted and tested, not just requested

• GTM container ownership confirmed

• Margin structure and business priorities documented

• Named approver per decision type, plus a deputy

• Feedback turnaround agreed in writing

• Implementation owner for winners confirmed

• Brand, legal and compliance constraints shared

• Peak periods and dev freezes on a shared calendar

• Previous test history shared, or its absence confirmed

• GA4 reconciled against platform order data

Ten items. A brand that arrives with these complete starts testing in week three rather than week six — and at typical test cadence, three weeks is most of a test cycle.

Before all of this, though, comes choosing the right partner. Our guide to choosing a CRO agency covers the questions worth asking first, and how to read a CRO agency's monthly report covers what to expect once the work is running.

Planning a CRO engagement and want to arrive prepared? Talk to FunnelFreaks about a CRO Programme.