Context
A Series-A B2B SaaS company — collaboration software for marketing teams — had great top-of-funnel numbers and a flat activation curve. The CEO's framing: "We're pouring water into a leaky bucket and we don't know where the leak is." Activation, defined as "signed up → performed three core actions within 7 days," had been stuck at around 24% for two quarters. Paid conversion was predictably anemic.
Challenge
The product team had strong opinions about what to fix (onboarding copy, a fifth-step modal, an empty-state illustration). The growth team had strong opinions about what to fix (pricing page, signup form length, free-trial duration). No one had a quantified picture of the funnel, and no one was running controlled experiments. Every release was half-feature, half-intuition.
My brief: give the team a shared source of truth for the funnel, prove or disprove the loudest hypotheses with experiments, and leave behind a playbook they could run without me.
Approach
1. Instrument the funnel
The product already tracked events, but naming was inconsistent and three steps were missing. I audited the tracking plan, added five events, and wrote a SQL model that collapsed the mess into seven clean funnel steps. That model became the single source of truth — the product team, growth team, and CEO looked at the same numbers for the first time.
2. Cohort analysis
A flat funnel number hides the fact that different user segments behave differently. I split signups by (a) company size, (b) acquisition channel, and (c) day-of-week signup. Two findings jumped out:
- Signups from paid search converted at 16% activation. Signups from referrals converted at 52%. The product worked; the traffic was the issue for a third of spend.
- Signups who invited a teammate on day 1 activated at 71%. Those who didn't activated at 12%. The "invite a teammate" step was the single largest leverage point in the entire funnel.
3. A/B test design
I worked with product and growth to pick four high-confidence experiments and wrote the test docs — hypothesis, primary metric, sample size, MDE, guardrails, and a clear stop rule. We ran them sequentially to avoid interaction effects.
- Test 1: Make teammate invite the first step of onboarding (vs. last). Primary metric: day-1 invite rate.
- Test 2: Shorten signup form from 6 fields to 2. Primary metric: signup completion.
- Test 3: Add a "seed workspace" with 3 sample projects for new accounts. Primary metric: first project created.
- Test 4: Extend free trial from 14 to 21 days. Primary metric: trial-to-paid.
Solution
Two tests shipped as permanent product changes; two did not.
- Test 1 (teammate invite first): shipped. Day-1 invite rate went from 44% to 68%. Activation for the variant cohort: 39%.
- Test 2 (shorter signup form): shipped. Signup completion +11%. No measurable quality drop in downstream conversion.
- Test 3 (seed workspace): rolled back. First-project rate went up, but activation was flat — users engaged with the seed content instead of creating their own. Good lesson about vanity metrics.
- Test 4 (21-day trial): inconclusive. Trial-to-paid moved +0.8 pts, within noise. We did not extend the trial globally.
"The biggest shift wasn't the numbers — it was that we finally stopped arguing about the funnel and started testing it. That's the muscle Jayant Sharma built with us." — Head of Growth
Impact
The annualized revenue impact, using the company's own LTV numbers, landed at roughly $840k in incremental ARR in the first twelve months — with a paid search cohort that went from unprofitable to break-even.
Lessons
- A shared funnel is half the engagement. The moment product, growth, and leadership all looked at the same numbers, debates shrank from weeks to minutes.
- One leverage point is usually bigger than all the others combined. In this case: the teammate invite. Finding it took one cohort split; all four tests pointed back to it in some way.
- Roll-backs are part of the process. The seed-workspace test looked like a win on its direct metric and a loss on the real one. Writing the guardrail metric into the test doc up front was what made the roll-back easy and unemotional.
- Leave the playbook behind. I spent the last two weeks writing a 12-page test-design template the growth team now uses on every new experiment. That's the work that survives the engagement.