A personalization roadmap fails in a predictable way: the team launches eight segments in the first month, none of them gets enough traffic to produce a readable result, and by month three nobody can say whether personalization works. The fix is not more ambition, it is sequencing. Start with fewer segments than you think, check what your traffic can actually detect before you test anything, and decide in advance what result makes you scale a rule or kill it.
This is a week-by-week, 90-day plan for personalization testing on a B2B website: who owns what, which two segments to start with and why, a sample-size reality check, decision rules, and a roadmap table you can copy. How to turn the results into an ROI figure is covered separately in our guide to measuring personalization ROI; this page is about running the tests.
Who Owns What
Personalization programs without a named owner drift. Five roles cover it, and in a small team one person can hold several:
| Role | Owns | Typical time |
|---|---|---|
| Program owner (marketing) | Roadmap, test backlog, decisions at each checkpoint | 4-6 hours / week |
| Content lead | Headlines, proof blocks and CTA copy per segment | Heavy in weeks 3-4 and 8-10 |
| Marketing ops / analytics | Tracking, holdout assignment, CRM fields, reading results | Heavy in weeks 1-2, then 2 hours / week |
| Sales lead | Segment choice sanity check, lead-quality feedback | 30 minutes at each checkpoint |
| Web developer | Script install, flicker and performance checks | A few hours in week 1 |
The Sample-Size Reality Check (Do This First)
Before choosing segments, find out what your traffic can detect. Most B2B teams skip this and then run tests that could never have produced a result.
A standard approximation (Lehr's rule, 80% power, 5% two-sided significance) is: visitors needed per group = 16 x p x (1 - p) / d², where p is the baseline rate and d is the absolute difference you want to detect. Turned around, the smallest relative lift you can detect is 4 x √(p x (1 - p) / n) / p for n visitors per group.
The table applies that to a test that runs about two months with a 50/50 split, so each group gets roughly one month of the segment's traffic. Two metrics are shown: a form conversion at 2% (demo or contact request) and a CTA click-through at 10%.
| Monthly visitors in the segment (on tested pages) | Smallest detectable lift: form conversion at 2% | Smallest detectable lift: CTA click at 10% | What you can realistically test |
|---|---|---|---|
| 1,000 | ~89% | ~38% | Only large changes on click metrics. Pool segments or use pre/post comparisons for conversion. |
| 5,000 | ~40% | ~17% | Click and engagement tests per segment. Conversion only for bold changes. |
| 20,000 | ~20% | ~8.5% | One conversion test per segment at a time. |
| 50,000 | ~12.5% | ~5.4% | Conversion tests with realistic lifts; a few tests in parallel on different pages. |
Example of the math for the 5,000 row: n = 5,000 per group, p = 0.02. √(0.02 x 0.98 / 5,000) = 0.00198. Multiply by 4: 0.0079, which is a 0.79-point change on a 2% baseline, or about 40% relative.
Three conclusions follow. First, if your segment gets a few thousand visitors a month, use a higher-frequency metric (CTA clicks, pricing-page reach) as the primary test metric and track conversions as a secondary signal. Second, make changes big enough to matter: a reworded button will not produce a 40% lift. Third, if you keep a 30% rather than 50% holdout, you need about 1.2 times the traffic. Run your own numbers in the A/B test calculator, and see our A/B test sample size guide for the reasoning behind the formula.
Which Segments to Start With, and Why
Start with exactly two segments. Pick them on four criteria: enough traffic to pass the reality check, a clear difference in what the segment needs to hear, content you already have (or can write in two weeks), and value to the business.
For most B2B sites, these two pass:
- Your best-fit industry (identified visitors). Pick the vertical with the most closed-won revenue in the last 12 months, not the one leadership is most excited about. Industry is reliably available from visitor identification, buyers in a vertical share vocabulary and objections, and you probably already have a case study or two. Change the hero headline and the customer proof.
- Returning visitors who have viewed pricing or a comparison page. This segment needs no identification at all, so it usually has more traffic than any firmographic segment, and its need is obvious: they are evaluating, so show a direct meeting CTA and evaluation content (security, integrations, implementation) instead of awareness messaging.
If industry does not split your market well, use company size (for example, under 200 employees versus 1,000 and more) instead of segment one. The messaging difference between a team that buys by credit card and a team that runs a procurement process is usually large.
Leave these for later: named-account lists (far too little traffic to test; personalize them, but do not expect a readable test), job-title segments (you rarely know the visitor's role before a form fill), and micro-segments that combine three attributes. Our guide to building an ideal customer profile covers how to pick the fit attributes from your win data.
The 90-Day Personalization Roadmap
Copy this table into your project tool. Each phase has one exit criterion: do not start the next phase until it is met.
| Weeks | Phase | Deliverables | Owner | Exit criterion |
|---|---|---|---|---|
| 1-2 | Foundations | Primary and secondary conversion events defined; tracking verified; personalization script installed; baseline per candidate segment (traffic, conversion, CTA click-through, pricing reach); sample-size check done | Marketing ops, developer | Baseline numbers signed off; two segments chosen that pass the reality check |
| 3-4 | Build | Two variants per segment (headline + proof); holdout assignment by account; analytics event for segment, rule and variant; QA on desktop and mobile; flicker and Core Web Vitals check | Content lead, marketing ops | Both segments live with a holdout; QA checklist passed |
| 5-6 | Run, no peeking | Weekly health check only: identification rate, segment assignment accuracy, holdout balance, page speed. No result decisions | Marketing ops | No technical issues; groups balanced |
| 7-8 | First read | Results on the primary metric against the planned sample; lead-quality feedback from sales; decision per rule using the rules below | Program owner, sales lead | Each rule marked scale, iterate or kill |
| 9-10 | Iterate and expand | Winning rule extended to a second page; one bolder variant for inconclusive rules; a third segment added only if traffic allows | Content lead, program owner | New tests live with planned sample sizes |
| 11-12 | Second read | Results of week 9-10 tests; first pipeline signal (opportunities created by group, directional only) | Program owner, marketing ops | Decision per rule recorded |
| 13 | Review and plan | Playbook updated with winners and losers; quarter-two backlog; results summary for leadership (link to the ROI model) | Program owner | Next quarter's roadmap approved |
Weeks 1-2: foundations
Pick one primary conversion event (usually a demo or contact request) and one secondary (pricing-page reach or a high-intent CTA click), and use them for the full 90 days. Changing the success metric mid-test is how inconclusive results become "wins". Verify that the events fire on every form, including embedded and chat forms. Then pull the baseline for each candidate segment and run the reality check above before committing to segments.
Weeks 3-4: build
Change one or two elements per segment, not the whole page. Headline and proof are the usual pair because they carry the most meaning. Write each test into a one-page A/B test plan with the hypothesis, primary metric, sample size and planned end date, before launch. Assign the holdout at the account level so colleagues from the same company see the same version.
Weeks 5-6: run without peeking
Checking significance every day and stopping the first time it crosses 95% inflates false positives badly. During these weeks, check only the health of the setup: are visitors being identified and assigned correctly, are the two groups similar in size, has page speed changed?
Weeks 7-8: first read
Read each test only when it reaches its planned sample. Ask sales whether the leads from personalized pages look any different. Then apply the decision rules.
Weeks 9-12: iterate and expand
Extend winners to the next page in the journey (a winning industry headline on the homepage usually deserves matching proof on the pricing page). Give inconclusive rules one bolder variant. Add a third segment only if it passes the reality check on its own traffic.
Week 13: review
Record every rule, its result and the decision in your personalization playbook, including the losers. Then plan the next quarter's backlog from what you learned rather than from the original wish list.
Decision Rules: Scale, Iterate or Kill
Write these down before the first test launches, so the results do not get argued into whatever someone hoped for.
| Result at the planned sample size | Decision |
|---|---|
| Primary metric up, lower bound of the 95% confidence interval above zero, lead quality unchanged or better | Scale: roll out to the full segment (keep a small holdout), extend to the next page |
| Primary metric up but not significant; secondary metric up | Iterate once: one bolder variant, same metric, new planned sample. If still inconclusive, kill |
| No difference on primary or secondary metric | Kill: the segment may not need different messaging, or the change was too small. Free the traffic for another test |
| Primary metric significantly down | Kill now, even before the planned end date, and record why you think it failed |
| Clicks up, conversions flat after a full sample | Kill or move: the variant creates curiosity, not intent. Try the message on a higher-intent page |
| Conversions up, lead-to-opportunity rate down | Stop and review: you are attracting the wrong accounts; tighten the segment or the CTA |
Our view: a kill is a result, not a failure. A roadmap where every test "won" is a sign that tests were called early.
A Copyable Test Backlog
Keep one row per test. The two columns most teams leave out are "planned sample per group" and "decision date", and they are the two that stop tests from running forever.
| Test ID | Segment | Page | Element | Hypothesis | Primary metric | Planned sample per group | Start | Decision date | Result / decision |
|---|---|---|---|---|---|---|---|---|---|
| T1 | Best-fit industry | Homepage | Hero headline + proof | Naming the industry's core problem and showing a peer customer increases demo requests | Demo requests / visitor | ||||
| T2 | Returning, viewed pricing | Homepage | Primary CTA | Evaluators respond to a direct meeting offer over a content download | CTA click-through | ||||
| T3 |
Name every rule with a consistent pattern such as segment-page-element-version (for example, "logistics-home-hero-v2"). By day 90 you will have a dozen rules, and unnamed ones become impossible to audit.
Common Mistakes
Choosing segments before checking traffic. The most common reason a 90-day plan produces nothing. Run the reality check in week one.
Comparing personalized visitors with the site average. Identified visitors from target industries convert differently from anonymous traffic anyway. Compare against a holdout from the same segment, or the "lift" is just the segment.
Changing too little. Swapping one word in a headline cannot produce a lift that B2B traffic can detect. Change the message, not the punctuation.
Overlapping rules without priority. An account can be both enterprise and in your target industry. Decide upfront which rule wins (usually the most specific), or two tests will contaminate each other.
Letting the default experience rot. Most of your traffic will still see the default page after 90 days. Keep improving it, ideally through ordinary A/B tests; see A/B testing vs multivariate testing for choosing the method.
Reporting pipeline at day 60. Opportunity and revenue effects need at least one sales cycle. At day 90, present pipeline as directional and point to the measurement plan for the full read.
Markettailor's segmentation features cover the segment types in this plan: industry, company size and on-site behavior such as pricing-page visits.
Frequently Asked Questions
What should a personalization roadmap include?
Named owners, the segments you start with and why, a sample-size check showing what your traffic can detect, a phased plan with deliverables and exit criteria, decision rules for scaling or killing each rule written before launch, and a test backlog with planned sample sizes and decision dates. Without the sample-size check and decision rules, a roadmap is a list of hopes.
How much traffic do you need for personalization testing?
It depends on the metric. At a 2% conversion rate, a segment with 20,000 monthly visitors can detect about a 20% relative lift in roughly two months on a 50/50 split. With 5,000 visitors, only a 40% lift is detectable, so use a higher-frequency metric such as CTA click-through, which at a 10% baseline can show a 17% lift.
Which segments should you personalize first?
Two: your best-fit industry, chosen from closed-won revenue, and returning visitors who have viewed pricing or comparison pages. The first has a clear messaging difference and existing proof; the second needs no identification, so it usually has the most traffic. Leave named-account lists and job-title segments until later because their traffic is too small to test.
How long should a personalization test run?
Until it reaches the sample size you planned before launch, and for at least two full weeks so every weekday is represented. Do not stop the moment a result looks significant: checking daily and stopping on the first significant reading inflates false positives. Pipeline effects need a further full sales cycle beyond that.