Answers

How can I tell whether our growth problem is acquisition, conversion, activation, or retention?

Why the order beats the analysis#

Every dollar spent fixing acquisition while retention is broken is a dollar wasted 60 days later when those users churn. Every hour on new-user acquisition while activation is broken produces trials that never convert.

Fix the bottom first. The diagnosis itself is straightforward once you're running it in the right sequence, and the sequence is what most teams get wrong.

The four layers with their thresholds#

Numeric thresholds matter here. Without them, every layer looks like it's "a bit down" and the team argues about interpretation instead of acting.

Layer 1: retention#

Pull four quarters of monthly logo churn, revenue churn, and cohort curves by signup quarter. If churn is up 20% or more, stop here. Nothing above this layer matters until it's fixed.

The two breaks you'll usually find: one customer segment started churning, often your smallest deal size or a vertical your product no longer serves well, or a competitor moved on your ICP with a better offer.

For catching the competitor version early, see How do I tell authentic third-party advocacy from paid or fabricated mentions?.

Layer 2: activation#

Customers who churn early almost always never activated. Pull time to first meaningful action, week-two retention on new signups, and onboarding completion rate, then compare the last cohort to the trailing three.

Drops usually mean the product changed and new users can't find the core value, onboarding changed without testing, or sales is closing poor-fit accounts.

This layer is as much qualitative as quantitative. Watch 5 to 10 recorded new-user sessions before you trust the dashboard, because the numbers tell you something broke and the recordings tell you what.

Layer 3: conversion#

Pull lead-to-opportunity rate, opportunity-to-closed-won rate, and sales cycle length. Compare last quarter to the trailing three.

Common causes: sales turnover, ICP drift bringing in wrong-fit leads, a competitor updating their comparison content, or pricing perception shifting quietly.

For the content-side version of this diagnosis, see Should I cut content spending when traffic is growing but revenue is flat?.

Layer 4: acquisition#

Only now. And split by channel and quality rather than total volume: organic, AI-referral, paid, and branded direct, each with its own conversion rate.

The channel-level split matters because AI surfaces behave differently from each other. Our own state of AI search research found Perplexity mentions the tracked brand in 41.7% of matched B2B answers versus 32.3% on ChatGPT, with Google AI Mode citing in 97.4%. A drop concentrated on one surface is a targeted problem with a targeted fix, not a category-wide crisis.

See How do I track ChatGPT traffic in GA4? for the tracking.

What the diagnosis usually finds#

For B2B SaaS between $2M and $50M ARR, in rough order of frequency:

1. Retention drift in one segment. Fix: focused onboarding, ICP tightening, or deliberately sunsetting the segment.
2. Activation degraded after a product change. Fix: onboarding rework.
3. Conversion drop from ICP drift. Fix: tighten targeting on organic and paid together.
4. Acquisition drop from AI-answer absorption. Fix: redirect content investment toward bottom-funnel and AI-search visibility.

That fourth cause is genuinely new. Two years ago acquisition problems were traditional SEO or paid channel shifts. "AI answers absorbing our top-funnel clicks" is now common and most dashboards still don't measure it. Pew Research found that when an AI summary appears, users click a source link in only about 1% of visits, which is the mechanism behind an acquisition drop with no ranking loss to explain it.

Run it in five steps#

1. Build one spreadsheet, one tab per layer. Last 12 months of monthly data.

2. Compute each layer's delta. Last quarter vs. trailing four-quarter average, as a percentage.

3. Find the biggest negative delta. That's the bottleneck.

4. Fix nothing above it until it's resolved. This is the discipline most teams break and the reason growth investments feel wasted six months later.

5. Re-run monthly. Bottlenecks move, and last quarter's problem often isn't this quarter's.

Automate the monthly version#

The discipline this framework depends on is diagnosing in reverse order, and that's exactly the discipline that collapses under pressure when acquisition is the number leadership is asking about.

Encoding the order in a Branch node removes the judgement call. Each layer computes on its own path, and the agent reports the deepest broken layer rather than the loudest one.

The monthly agent:

Start (schedule, 1st of month) → HubSpot Get CRM Objects (churn, activation events, opportunities, 4 quarters) → Branch (four paths, one per funnel layer, each computing its own delta against the trailing four-quarter average) → GA4 AI Traffic Overview → share-of-voice recipe → visibility-losers recipe → model-blind-spots recipe → Code (rank the four layers by negative delta and apply the numeric bottleneck thresholds) → Conditional (if the bottleneck is acquisition, continue to the channel split, otherwise stop and report) → Prompt LLM (name the bottleneck and state explicitly which layers not to work on) → Send Email to CMO and CEO.

The model-blind-spots recipe earns its place on the acquisition path specifically. It surfaces which AI surfaces know least about your brand rather than averaging them into one reassuring number, so an acquisition drop concentrated on a single surface reads as a targeted problem rather than a category-wide crisis.

The instruction to name which layers not to work on is the part that makes this actionable. Most growth diagnoses fail at the point where three teams all start fixing their own layer simultaneously.

FAQ#

Related answers#

• Should I cut content spending when traffic is growing but revenue is flat?
• Why is my SEO traffic growing while revenue stays flat?
• How do I prove SEO is generating revenue and not just traffic?