Definition

The umbrella term for non-data methods of generating candidate behaviors to test in multiplier analysis — marketing-copy reading, onboarding-flow reading, feature-list reading, review reading, team hunches, and similar inputs.

What unifies the methods

The methods are heterogeneous by construction. They cohere as a category through one shared property: each one gathers "what seems to be working" from sources humans interpret rather than from logs they query. There's no shared procedure across them — marketing copy gets read differently than onboarding flows, and team hunches don't resemble either. The category exists primarily through its contrast with retracing, the data-driven counterpart.

What each sub-method surfaces

Each input surfaces a different angle on which behaviors might matter:

• Marketing-copy reading: what the offering claims it does (aspirational framing)
• Onboarding-flow reading: what the offering asks users to do (provider intent)
• Feature-list reading: what the offering can do (capability inventory)
• Review reading: what users are accomplishing with the offering (user-described use)
• Team hunches: what people closest to the offering suspect is working

Pairing with retracing

Hypothesis sourcing and retracing both produce candidate behaviors for the same downstream step, but they cover different blind spots. Hypothesis sourcing can miss patterns nobody articulated — behaviors that drive conversion but aren't named in marketing, onboarding, reviews, or team folklore. Retracing can miss patterns that aren't instrumented — behaviors the offering doesn't track. Running both before multiplier analysis broadens the candidate list more reliably than either alone.

The event catalog's role

The analytics tool's event catalog is a legitimate input for catching candidates that didn't come up through hypothesis sourcing — useful for surfacing what's instrumentable. But it's not a starting point on its own. Events get instrumented for many reasons unrelated to user intent: engineering convenience, debugging, marketing attribution. A candidate list assembled purely from the catalog produces a ranking whose connection to user intent is unknown. Hypothesis sourcing supplies the interpretive grounding; the catalog supplies the technical inventory.

Example

A team sources candidate behaviors before running multiplier analysis against first paid conversion. Marketing copy yields "create your first invoice in under two minutes" and "get paid faster." Onboarding yields "add your business details," "connect a payment processor," "send your first invoice." The feature list yields recurring invoices, payment reminders, multi-currency support. G2 and Capterra reviews yield "I use this to send invoices to my freelance clients," "we use it to track which clients are late paying," "we connect it to QuickBooks so we don't double-enter." Team hunches yield "users who connect a payment processor in week one seem stickier."

Each source contributes candidates with different shapes. Reviews surface outcome-shaped candidates — invoice sent, payment received, books reconciled — that carry the most signal. Onboarding surfaces provider-defined milestones like "connected a payment processor," which test whether the flow's design assumptions match what actually drives conversion. Marketing surfaces aspirational claims like "got paid faster," which test whether the promise corresponds to a behavior the data validates.

The full candidate list goes into multiplier analysis. Which sources produced the strongest-multiplier candidates is itself diagnostic — for instance, marketing-copy candidates underperforming review-derived candidates suggests the marketing is selling something different from what users actually value.

See also

• Tracking gaps