Experiments are only useful when they improve a model you understand.
Too many companies run growth experiments before they can describe how growth is supposed to happen. They test channels, landing pages, offers, emails, onboarding steps, referral prompts, and sales sequences without a clear model of the business. The result is not learning. It is scattered evidence.
Before experiments, build the growth model map.
A growth model is the operating logic of growth
A growth model explains how a company turns market demand into durable revenue. It is not a spreadsheet forecast. It is a causal map.
It should show:
- who the right customer is;
- where demand originates;
- how buyers discover the company;
- what creates trust;
- what converts interest into usage or purchase;
- what drives activation;
- what keeps customers;
- what expands accounts;
- what creates new distribution inputs;
- what constraints limit the system.
This sounds basic. It is not. Many teams can recite their funnel metrics but cannot explain their growth model.
Different models require different operating choices
A sales-led enterprise company needs account selection, trigger-based outreach, executive trust, implementation capacity, and expansion paths. Its bottlenecks may be sales capacity, proof quality, security review, champion enablement, or deployment success.
A PLG company needs fast activation, product-delivered value, usage expansion, invitation or collaboration mechanics, and clear upgrade moments. Its bottlenecks may be activation quality, self-serve comprehension, team adoption, or monetization friction.
A marketplace needs supply and demand liquidity. Its bottlenecks may be geographic density, trust, matching quality, take-rate sensitivity, or chicken-and-egg sequencing.
A content-led company needs durable buyer problems that people search for or share, strong editorial judgment, conversion paths, and authority that compounds. Its bottlenecks may be topic quality, domain trust, distribution, conversion intent, or content decay.
A partner-led company needs partners with access, incentive alignment, enablement, and repeatable joint value. Its bottlenecks may be partner motivation, deal registration, conflict with direct sales, or customer ownership.
A paid acquisition model needs economics that survive scale. Its bottlenecks may be payback period, saturation, creative fatigue, conversion quality, or retention.
A community-led model needs repeated member value, identity, contribution, and trust. Its bottlenecks may be moderation, content quality, member density, or the gap between engagement and revenue.
An enterprise expansion model needs retained value, executive visibility, multi-threading, and adjacent use cases. Its bottlenecks may be adoption depth, account planning, customer success capacity, or product readiness.
An ecosystem-led model needs integrations, platform dependency, co-selling, developer adoption, or workflow embedding. Its bottlenecks may be integration quality, marketplace ranking, platform risk, or partner economics.
These are not interchangeable. Running the same experiment in two different models can produce opposite lessons.
The growth model map
Use a simple map. Do not over-engineer it.
`
Target segment -> Demand trigger -> Discovery path -> Trust mechanism -> Conversion path -> Activation event -> Retention driver -> Expansion path -> Distribution loop -> Main constraint
`
Example for a sales-led enterprise workflow product:
`
VP Operations at regulated companies -> audit deadline or operational failure -> peer referral / targeted outbound / industry event -> proof from similar company -> discovery + pilot -> first workflow live with real team -> weekly operational dependency -> adjacent departments adopt -> customer story + partner referral -> implementation capacity and champion enablement
`
Example for a self-serve collaboration product:
`
Team lead with coordination pain -> project kickoff -> search / shared template / peer mention -> useful free artifact -> signup -> first shared workspace with invited teammate -> repeated weekly planning -> more seats and premium permissions -> invited collaborators create new workspaces -> activation quality and team invitation rate
`
Notice what this does. It turns growth from "try more channels" into a chain of assumptions.
Experiments should target assumptions
Once the map exists, experiments have a job.
They test a specific assumption in the growth model:
- Are we targeting a segment with urgent pain?
- Is the demand trigger observable?
- Can this channel reach the buyer in the right context?
- Does the trust mechanism reduce risk?
- Does activation predict retention?
- Does expansion follow usage depth?
- Does the distribution loop create more qualified inputs?
- Is the constraint actually where we think it is?
This is very different from testing button colors, subject lines, or random campaign ideas because the team needs a quarterly experiment count.
Sequence matters
The right growth move depends on stage.
At $1M ARR, a founder-led sales motion may be the best way to learn the market. At $20M ARR, the same motion may be a bottleneck if the company has not built repeatable segmentation, enablement, and expansion infrastructure.
Early PLG teams may need activation clarity before monetization optimization. Later PLG teams may need enterprise expansion, admin controls, and sales-assist motion.
Early content-led teams may need editorial wedge and distribution proof. Later teams may need refresh systems, conversion architecture, and authority defense.
The question is not "is this a good growth idea?" The question is "is this the right move for this model at this stage, given the current constraint?"
The output
A good growth model map should produce three decisions:
- What to focus on now. The constraint that most limits healthy growth.
- What not to do yet. Moves that are attractive but premature.
- What to measure. Signals that prove the model is getting stronger, not just busier.
If your experiments do not connect to those decisions, they are probably theater.
Start with the model. Then experiment.
