CRO vs SEO: the 3,000-session rule

One of these adds and the other multiplies

SEO adds sessions. CRO multiplies whatever share of them converts. Written as arithmetic the sequencing answers itself: doubling a conversion rate on 800 monthly sessions gets you a handful more enquiries; doubling sessions at a mediocre conversion rate gets you considerably more than that.

The reason the question gets asked at all is that CRO sounds cheaper and faster. It's neither, at low volume. The cost of a test isn't the tool licence — it's the eight weeks you spend waiting for a result that never arrives because the sample never got big enough. Meanwhile the SEO work you deferred has an eight-week head start it will never get back.

The maths: why tests need volume

A/B testing is a statistical procedure, and statistics have a minimum input size. The standard approximation for comparing two conversion rates at 80% power and 95% two-sided confidence is roughly n ≈ 15.7 × p(1−p) / δ² per variant, where p is your current conversion rate and δ is the absolute improvement you want to detect.

Run that with a 2% baseline conversion rate and an even 50/50 traffic split, and you get the table below. The last column is the honest verdict at each level.

• Typical CRO wins are 5% to 20% relative. Compare that against the third column. At 3,000 sessions you can only detect changes roughly four times larger than the wins you're likely to get.
• A higher baseline conversion rate helps a lot. At 5% conversion instead of 2%, detecting a 20% relative lift needs roughly two and a half times fewer sessions per variant. High-intent B2B pages can test earlier than blog traffic can.
• Running the test longer isn't free. Sample accumulates, but so does seasonality, so does the temptation to peek early, and every week the test runs is a week the page can't be improved.
• Peeking is the real killer. Checking daily and stopping when the variant is ahead manufactures winners out of pure noise. Fix the sample size before you start, then don't look.

The two thresholds, stated plainly

Three bands, three different answers. Find yours in Google Analytics 4 — organic sessions, last full month, landing pages that can actually convert. Not total sessions across the whole site.

The conversion wins that need no test at all

None of these require statistical significance because none of them are hypotheses. They're either broken or they're not. Ship them at any traffic level, on any budget, this week.

1. Cut form fields to what you'll actually use in the first call. Name, one contact method, one qualifying question. Every extra field is a reason to leave.
2. Put a tappable phone number and WhatsApp in the mobile header. In India a large share of buyers would rather message than fill a form. Making them choose your preferred channel costs you enquiries.
3. Publish pricing, or publish a range. A price band with an explanation beats "contact us for a quote" for every buyer except the ones you didn't want anyway.
4. Fix mobile page speed. Slow pages lose conversions and rankings simultaneously — it's one of the very few changes that pays into both accounts.
5. Delete the homepage carousel. People see slide one and nothing else. Pick the message you'd have put on slide one and commit to it.
6. Make the primary action obvious above the fold on a phone. Not the only action. The obvious one.
7. Add the trust markers you actually have — real client logos with permission, genuine reviews, real team photos. Never invent one. A fake testimonial is a legal problem, not a conversion tactic.

How a CRO change quietly kills your rankings

This is the part that gets skipped when the two disciplines sit with different teams. Conversion work touches exactly the things search engines read, and the damage takes weeks to surface — long after the test was declared a win.

We've seen more organic declines caused by a well-intentioned redesign than by any algorithm update. The pattern is always the same: conversions up 12% in the test, organic sessions down 30% two months later, and nobody connects the two because the test window closed in between.

• Anti-flicker snippets destroy LCP. Client-side testing tools hide the page until the variant loads. That's a deliberate render delay on your most important Core Web Vitals metric, applied to every visitor including the ones you're trying to rank for.
• "Cleaning up" a page deletes the copy that ranked it. The 600 words a designer wants to cut are frequently the 600 words matching the query. Move them, restructure them, put them below the fold — don't delete them.
• Redirect-based split tests need 302s, not 301s, and the variant needs a canonical pointing at the original. Get this wrong and you hand Google two indexable versions of the same page.
• Nav simplification cuts internal links. Those links were distributing authority to deep pages. Trimming the menu from 30 links to 8 can orphan a lot of URLs at once.
• Personalisation that varies by user agent risks cloaking. If Googlebot systematically sees a different page from your users, you have a policy problem, not a testing problem.
• Gating content behind a form removes indexable text. The gated version ranks for nothing, so the traffic you were converting stops arriving.

Budget split at ₹75,000 and ₹1,75,000 a month

Splits, not principles. These assume search is your primary acquisition channel and that the conversion work is scoped as a real deliverable rather than a line in someone's retainer they'll get to eventually.

• Our SEO starts at ₹75,000/mo, with smaller sites from ₹40,000/mo. If you want SEO, organic social and performance marketing together, the Own Everything Stack is ₹1,75,000/mo. All ex-GST, month-to-month after the first quarter, and you keep every asset.
• Conversion work is scoped separately, because pretending it's free inside an SEO retainer is how it ends up not happening. See conversion rate optimisation and our pricing.
• Whichever split you choose, the number we're accountable for doesn't change: your trailing-90-day qualified organic leads, frozen on day one. Beat it in 90 days or we keep working free until we do. That's how the guarantee works, and it's why sequencing matters to us as much as it does to you.