When a CRO agency makes sense — and the threshold that decides it

CRO is the one agency category where 'are we ready' is a math question with a checkable answer. Run these five signals before any discovery call.

• 01
  The volume threshold: roughly 1,000 monthly conversions on the pages you want to optimize. Below it, tests take months to reach significance or never do, and the retainer burns while you wait. The right purchase under the line is research and UX work (heuristic audits, user testing, session analysis), not a testing program. Above it, a real testing cadence becomes possible — and tests-per-month is the single biggest predictor of program ROI.
• 02
  The traffic floor that pairs with it: 10,000+ monthly visitors to optimization-target pages, alongside spend signals like $100K+ monthly ad budgets: the point where a conversion lift pays for the program many times over because it compounds across everything you're already buying traffic for.
• 03
  Working measurement, before anything else. An agency inheriting broken tracking spends its first two months (and your first two invoices) on analytics repair — a pattern the ECM in-house vs agency guide flags on the internal side too. If GA4, server-side events, and revenue attribution aren't trustworthy today, buy that fix first as a bounded project; it's cheaper as a project than as a retainer's warm-up act.
• 04
  An internal owner at five-plus hours a week. CRO programs die of client-side neglect more than agency-side incompetence: hypotheses need business context, tests need approval, and learnings need someone to carry them into the roadmap. No owner, no program — whatever the retainer says.
• 05
  A clear conversion KPI tied to money. Checkout completion, add-to-cart (benchmark: 6.0–7.9%), lead quality — one number the program is accountable to, pulled from your core KPI set and baselined before kickoff. 'Improve the site' is how programs produce activity reports instead of revenue.
• 06
  If checkout is your leak, check the platform before hiring anyone. A meaningful share of 'CRO problems' at $5M–$50M brands are checkout-architecture problems, where the gap between a customized checkout and a platform-native one is larger than a year of testing wins. The ECM Shopify Plus vs Shopify guide covers what Checkout Extensibility now handles natively.

What CRO actually costs in 2026

Pricing clusters into clean tiers, and the tier you belong in is set by your traffic and program scope, not your ambition. The visible retainer is also not the whole bill.

The retainer tiers

• →
  Entry / boutique: $2,000–$5,000/month. Newer teams and niche specialists (Shopify product-page optimization, analytics implementation). Right for first programs at modest traffic, with the vetting bar set accordingly high.
• →
  Mid-tier: $5,000–$8,000/month, the median testing-program band. Established teams with solid track records; typical full-service programs start around $8,500. This is where most $5M–$50M DTC brands with real conversion volume belong.
• →
  Senior / enterprise: $10,000–$35,000+/month, with average top-tier retainers near $15,000 and 6-month minimums standard across the tier. Named enterprise programs price per-test at $2,500–$12,500 including full implementation. Below the retainers: one-time audits at $1,000–$5,000 and hourly consulting at $75–$250+ for bounded questions.
• →
  Performance-based models take 10–30% of incremental revenue or lift, sometimes on a reduced base. Same rule as every performance model: the attribution methodology decides the invoice, so agree on the counting before the testing.

The lines the proposal skips

• →
  Testing tools and platforms: $200–$1,500/month in subscriptions, and enterprise testing platforms run $25K–$60K a year; worth asking whether the agency's stack is included, billed through, or assumed to be yours. Inventory it in a tech stack audit before signing.
• →
  Implementation: 10–40 developer and QA hours per test. Whose developers? If it's yours, the retainer's real cost includes your sprint capacity, and your release calendar caps the testing pace: the detail that quietly halves the value of many mid-tier programs.
• →
  Context for the fee: agency retainers across all marketing services average $3,209/month (Ahrefs, 439 providers polled). CRO sits well above that line because every test consumes design, development, and analysis hours. A specialized CRO program at $3,500–$20,000 versus a full-service bundle at $5,000–$25,000 where CRO is one slice is the real comparison to run; the ECM agency pricing guide covers the full-bundle side.

What good programs actually deliver, and the win-rate reality

The results data is genuinely strong, which is exactly why the inflated version of it needs separating out. Both live in this section.

• 01
  The documented program-level outcomes: structured CRO programs report an average 223% ROI; A/B testing alone averages an 18% conversion lift after six months of structured work; and teams running high-velocity programs (15+ tests monthly) report 20–40% cumulative conversion improvement in year one. The cadence drives the compounding, which is why the readiness thresholds above matter so much.
• 02
  The named cases give the range: SplitBase's optimization work helped Moon Pod generate $3.3M in additional annualized revenue; documented enterprise programs include an 18% conversion uplift for eBay and a 50% improvement for 3M over one year. Individual landing-page tests post 30%+ lifts. These are real, and they came from programs, not single silver-bullet tests.
• 03
  Now the win-rate reality: the industry average is 20–30% of tests producing a winner. Agencies advertising 46%, 52%, or 61% win rates are quoting self-reported figures, sometimes from selective datasets — treat any number above the industry band as a marketing claim until the agency shows the denominator: how many tests, over what period, counted how. A 25% win rate with high velocity and straight analysis beats a claimed 55% you can't verify.
• 04
  Losing tests are a deliverable, not a failure, if the program treats them as one. A serious program documents every loser's learning and feeds it into the next hypothesis; that research asset is a large share of what the retainer buys. An agency that only reports winners is curating, and curated reporting is the first red flag in the next section.
• 05
  Set expectations by phase: 2–5 tests shipped in the first 90 days while research, instrumentation, and QA pipelines come online, with the pace climbing after. An agency promising ten tests in month one is skipping the research that makes tests worth running; one still 'researching' in month four is billing you for a book report.

The questions that expose weak programs

CRO is a methodology business wearing a results costume. These questions check the methodology, which is the part you're actually buying.

• 01
  "Walk me through a recent losing test and what you did with it." The single most revealing question in the category. Strong agencies answer with a hypothesis, a segment analysis, and the follow-up test it produced; weak ones pivot back to their wins reel. You're buying a learning system, and this is how you inspect it.
• 02
  "Show me case studies with before-and-after rates, traffic volumes, and test durations." A percentage lift without a base rate, sample size, and duration is decoration. '125% lift' on 40 conversions a month is noise; 8% on 20,000 is a business result. Insist on the full arithmetic, and discount anything that arrives without it.
• 03
  "What's your statistical standard — and when do you call a test early?" You want a straight answer about significance thresholds, minimum sample sizes, and run-time rules (full business cycles, not first-week peeks). Agencies that call tests at day five are manufacturing win rates, and the revenue those 'wins' report rarely survives the quarter.
• 04
  "How much research precedes your first test?" The right shape: heuristic review, analytics deep-dive, session recordings, and user research feeding a prioritized hypothesis backlog, then testing. An agency whose first-month plan is 'test your button colors' is running 2019's playbook at 2026 prices.
• 05
  "Who implements, who QAs, and what happens to flicker and page speed?" Client-side testing tools can degrade Core Web Vitals and leak revenue while measuring it — the same speed-to-revenue math covered in the ECM replatforming SEO guide. A current agency talks server-side or edge testing for meaningful tests, owns QA end to end, and can tell you exactly how many of your developer hours each month the program assumes.
• 06
  The same engagement disciplines as any agency contract apply: named senior team, no undisclosed subcontracting, you own the data, the test archive, and the learnings at exit. The full framework is in the ECM agency selection guide; CRO adds one clause worth writing in: the hypothesis backlog and results database transfer to you, in a readable format, whenever the engagement ends.

The signals that predict a wasted retainer

Assembled from the failure patterns practitioners describe, not hypotheticals. Any one is a pause; two is a pass.

• 01
  Guaranteed lift numbers before seeing your data. Nobody can promise a percentage without knowing your traffic, conversion volume, and baseline. Published '30% lift guarantees' are marketing structures with fine print, not statistics.
• 02
  A proposal with no research phase, or a testing plan for a store below the volume threshold. The second one is the quiet scandal of the category: selling significance-dependent testing to stores that can't reach significance. If your conversions are under the line and the agency didn't mention it, they either didn't check or didn't care — both disqualifying.
• 03
  Win-rate marketing without denominators, and reporting that only shows winners. Ask for the last twelve months of tests, all of them, with sample sizes. Refusal is an answer.
• 04
  Redesign-shaped proposals sold as CRO. 'We'll rebuild your PDP based on best practices' is a design project, sometimes a fine one, but it's not an experimentation program and shouldn't be priced or measured like one. The tell: no hypotheses, no control, no rollback plan.
• 05
  No mention of your developer hours, your release calendar, or your analytics state. These three constraints determine your real-world testing pace, the biggest ROI predictor in the category, and a proposal that ignores them was written before anyone looked at your business. ECM tracks how the CRO category keeps evolving in weekly Insights.