Case Study · Agency / B2B SaaS

We dogfooded Compound on our own site. Here's what happened.

Northpeak.io is the agency behind Compound. Before selling the system to clients, we ran it on ourselves — full audit, prioritized roadmap, live experiments. This is the honest account.

2.1×
CVR lift (hero CTA)
3
Experiments shipped
48 hrs
Time to first result

The starting point

What we knew vs. what the data showed

We run CRO for a living. We had strong opinions about what worked on our homepage. We were wrong about most of them.

Compound's audit crawled northpeak.io, extracted brand tokens, ran GA4 funnel analysis, and pulled Clarity heatmaps. The research agent surfaced 32 findings across 8 pages. Three problem spaces emerged: weak value communication, trust deficit, and offer misalignment.

None of that was in our roadmap. We had been optimizing around the wrong assumptions.

The findings

32 findings. 3 root-cause problems.

01

Weak value communication

The hero copy described our process, not the outcome. 'We run conversion optimization' means nothing without 'so you get 2× more paying customers from the same traffic.'

02

Trust deficit

Social proof was buried — a small logo strip below the fold. Clarity heatmaps showed 68% of visitors never scrolled past the hero. Nobody was seeing the testimonials.

03

Offer misalignment

Our primary CTA was 'Book a call' — a high-commitment ask on a cold visit. The free audit as lead magnet wasn't even on the homepage. We were asking for commitment before delivering any value.

The experiments

3 experiments. Prioritized by revenue impact, not gut feel.

Experiment 1 — Hero CTA copy Won · +112% CVR
Control
"Book a strategy call →"
Variant
"Get your free CRO audit →"

The hypothesis: lowering commitment friction increases top-of-funnel CVR. Swapping "book a call" (implicit cost: 30 min of your time) for "free audit" (immediate value, no pitch) doubled click-through on the primary CTA within 14 days.

Experiment 2 — H1 subheadline reframe Won · +34% scroll depth
Control
"We use the power of design and data to inspire users to act"
Variant
"Identify design and content problems — fix the ones that move revenue"

"Inspire users to act" is aspirational and vague. The variant states what we actually do (problem identification + prioritized fixes) in terms a growth team cares about (revenue). Scroll depth to below-fold content increased 34%.

Experiment 3 — Landing page optimization Running

Testing a benefit-led H1 on the LP optimization service page. "Get more conversions from every visitor" vs. the incumbent design-agency-style header. Early signal is directionally positive — results expected in 2 weeks.

"The fastest way to improve conversion is to fix what you'd never notice by looking at your own site."
— Navi Kang, Northpeak

The takeaway

What this means for your site

The biggest wins weren't design changes — they were framing changes. Same page, same offer, different words. That's the pattern we see across clients: conversion problems are usually communication problems.

The audit took 48 hours. The first experiment went live in a week. The results came back in two. That cycle — audit → hypothesis → launch → results — is repeatable. It's the system.

The point of Compound isn't that it's faster than an agency (though it is). It's that it builds up knowledge over time. Every test teaches you something your competitors don't know about your customers.

Early access open

Get your free audit.

Same process we ran on northpeak.io — full AI audit of your site, zero cost, no sales call required.

Start free →