← Case Studies
Case № 05 · Practicum by Yandex

A/B Testing the Externship USP

"How to validate a custdev hypothesis at scale — and shorten the sales cycle without running 11,000 interviews."
+30–40% conversion lift in 2 weeksSales cycle shortenedHypothesis confirmed at scale
Slide 01The Hypothesis

The page sold choice. Customers were buying employment.

Control — old page
"Choose your career"
4 tracks · broad selection · generic promise
vs
Variant B — externship USP
"Apply your skills with partner companies. Add experience to your resume."
Real business problems · Portfolio projects · Live presentation to employer representatives
Where the insight came from Custdev interviews with 30% of active and potential users. The test confirmed what those conversations revealed.
Slide 02The Test Design

Parameters locked before launch — not adjusted mid-run.

Test parameters
Baseline
3.0–3.5%
Visitor → qualified lead on control page
Target MDE
+30% rel.
Minimum meaningful lift worth detecting
Sample needed
6–11k
Total visitors across both variants
Duration
≥ 2wks
Full weekly cycles · no early stops
Traffic allocation — every visitor sees one version, same parameters apply to both
A · 50%
B · 50%
Single URL · client-side swap · user assigned via cookie · 95% confidence · 80% power · Tools: Google Optimize → VWO / Optimizely · GA4
Slide 03The Numbers

A relative lift of 30–40% — with an honest caveat.

3.5%
Control · old page
visitor → qualified lead
5.0%
Variant B · externship USP
↑ 30–40% relative lift
Shorter sales cycle · More qualified conversations · Faster enrollment decisions
Note: a PR push was running in parallel and was factored in as an additional variable during analysis. Results reflect the combined effect, with the page change as the primary tested factor.
Slide 04The Real Insight

9 months later, the data confirmed what the test suggested.

2 weeks
A/B test ran
Conversion lift confirmed
Immediate
Sales cycle effect
Shorter time from visit to enrollment decision
9+ months
Completion rate
Students who enrolled via externship USP graduated at higher rates
The test was designed to measure conversion. The completion rate finding came later — as a bonus. In EdTech, Completion Rate is one of the most important product metrics. The externship page didn't just convert more leads. It attracted students who were more committed to finishing.

The setup.

A/B testing is one of the most overrated and simultaneously underused tools in product marketing. Overrated — because most teams expect it to generate discoveries. Underused — because most teams test the wrong thing: button color instead of the offer, CTA wording instead of the value proposition, presentation instead of substance.

A test doesn't discover anything. It confirms and monetizes what you already learned through qualitative research. The strongest A/B results always follow deep customer interviews — when the hypothesis is sharp enough to produce a noticeable effect on real traffic.

For companies trying to improve conversion, the right question isn't "what should we test" — it's "what do we already know well enough to bet money on." Without an answer to that question, a test becomes an expensive way to confirm that you don't know your product or your customer well enough.

What had to be true.

Practicum by Yandex USA tripleten.com ↗
EdTech · 1.9M monthly web visits

Best Online Bootcamp 2020–2022.

By the time the A/B test launched, the externship program had already been validated through customer discovery and an MVP. Interviews with 30% of active and potential users had given a clear answer: people were buying employment, not education. The old landing page with its "Choose your career" headline and four tracks on the first screen wasn't selling that — it was selling optionality.

The test had one job: to check whether the new offer — "Apply your skills with partner companies. Add experience to your resume." — with real business problems, portfolio projects, and live presentations to employer representatives as proof — produced a measurable lift in visitor → qualified call lead conversion. And to do that in a way where the decision came from statistics, not gut feel.

A PR push around the program launch was running in parallel and was factored in as an additional variable during analysis — standard practice when you can't control all external factors in a real business environment.

You've done the interviews. You think you know what your customer needs. How do you confirm it at scale — without interviewing another 11,000 people?

How it moved.

00CustDev interviews with 30% of active and potential users — hypothesis formation
01Baseline conversion rate recorded before test launch
02MDE defined — minimum effect size meaningful to the business MDE (Minimum Detectable Effect) — the smallest improvement worth detecting. The smaller the effect you want to catch, the more traffic you need.
03Required traffic volume and test duration calculated
04Single primary metric selected, tied directly to revenue
05Downstream tracking configured — lead progression to payment, not just form submission
06Test deployed: 50/50 split, variant assignment via cookies, single URL with client-side swap
07Internal traffic and bots excluded; PR push flagged as confounding factor
08Results analyzed against lead quality, not just volume

What I actually did.

The process started before any test was designed — with customer discovery interviews across 30% of active and potential users. Those conversations revealed the core insight: students weren't evaluating curriculum breadth. They were asking one question — will this get me a real job? The externship program was the most concrete answer the product had. The hypothesis was that leading with it, rather than burying it, would change conversion.

Before the test launched, three things were locked in: the baseline conversion rate of the control page (3.0–3.5% visitor → qualified lead), the minimum effect size considered meaningful (MDE of +30% relative, targeting ~4.5%), and standard reliability thresholds — 95% confidence and 80% test power.

From those parameters, the required traffic was calculated — roughly 6,000–11,000 people across both variants. At a daily traffic of ~800 people on the page, that meant approximately two weeks. Two full weeks was a hard floor — not a guideline — to capture complete weekly cycles. Weekday and weekend behavior differ, and stopping early inflates false positives.

One thing was tested: the value proposition on the first screen. Control: the old generic "Choose your career." Variant B: "Apply your skills with partner companies. Add experience to your resume." — with real business problems, portfolio projects, and live presentations to employer representatives on the first screen. 50/50 split, single URL with client-side rendering — no redirect. Redirects add latency and break attribution. Each user was assigned to a variant via cookies — the same person always sees the same version.

Tools: Google Optimize initially, then VWO and Optimizely after Google Optimize was discontinued. Measurement layer: GA4.

Primary metric: visitor → qualified call lead. Guardrail metrics: what happened to the lead downstream — did they show up to the call, enroll, complete the program, reach payment. This turned out to be the key insight: the externship variant lifted not just submission volume but lead quality and downstream conversion to payment. A page that "lost" on raw form submits could win on students who actually paid and completed.

One honest caveat: the parallel PR push brought in a different audience and raised brand awareness simultaneously with the test. That's a confounding factor we couldn't isolate. The 5% conversion figure is a before/after with several simultaneous changes — not a clean A/B. Inside the test itself, the control ran at 3.0–3.5% and variant B at 4.5–5.0%, a relative lift of 30–40%.

The most important finding wasn't in the numbers. The team believed the selling point was breadth of choice — four tracks, a strong curriculum, flexibility. That didn't move conversion. What moved it was one specific proof of employment. Customer interviews predicted this. The test confirmed it.

Beyond conversion, the sales team reported shorter conversations: leads arrived already understanding the externship offer and asking fewer foundational questions. The time from first contact to enrollment decision shortened. Nine months later, an additional finding surfaced: students who enrolled through the externship-led page completed the program at higher rates. In EdTech, completion rate is one of the most important product metrics — and this was a bonus the test wasn't designed to measure.

Measurable outcomes.

  • Variant B showed 4.5–5.0% conversion vs. 3.0–3.5% for control — a 30–40% relative lift in 2 weeks
  • Sales cycle shortened: leads arrived pre-informed, enrollment decisions came faster
  • CustDev hypothesis confirmed at scale
  • 9+ months later: higher completion rate among students enrolled via externship USP page

The stack behind it.

Skills
A/B TestingGrowth HackingCustDevProduct PositioningPerformance Tracking & ReportingKPI DevelopmentROI Analysis
Frameworks
Hypothesis-First TestingMDE FrameworkStatistical Significance TestingDownstream Funnel AnalysisConfounding Factor Identification
Tools
Google OptimizeVWOOptimizelyGA4Notion
Metrics
Visitor → Qualified Lead Conversion RateLead Quality ScoreDownstream Conversion RateMDEStatistical Significance (95%)Test Power (80%)
Resume
Confirmed a 30–40% conversion lift and a shorter sales cycle in two weeks — by running an A/B test built on custdev interviews with 30% of real users. Nine months later, the same page change showed up in completion rates.