Earnbound

The problem

Generic lead magnets often become a spray-and-pray strategy: offer everyone the same download and hope it attracts the right people. They rarely reflect what someone needs to solve or what would help them decide. The business gets a contact, but learns little about whether it can help.

The solution

I built Earnbound to turn a business’s expertise into interactive Guides for a specific audience and offer. Public-web research and business knowledge build customer intelligence about needs, buying triggers, objections, alternatives and the proof people need.

When current research is available for the selected audience, it feeds into the Guide alongside the company’s evidence and offer. The buyer answers relevant questions and gets a result and next step that fit their situation.

How it works

The architecture, at a glance.

  1. Customer intelligence

    Business context and research into needs, buying decisions and objections

  2. Guide compiler

    Selected audience, offer, evidence and rules for the result

  3. Buyer experience

    Questions that help clarify the buyer’s situation

  4. Useful result

    A result shaped by their answers and a relevant next step

Each Guide keeps the source information and decision rules used to build it. Current customer intelligence is matched to the selected audience; stale research is flagged for refresh.

Technology

Product
Next.js, React and TypeScript for the creator workspace and published buyer experience.
Customer research
An evidence scout combines internal context with public-web research. Specialist AI analysts examine customer needs, objections, buying criteria, alternatives and product fit, then a synthesis step brings the findings together.
Evidence & freshness
Structured research records retain sources, confidence levels, assumptions and gaps. Guide creation checks the audience match, completion status and freshness before including deep customer intelligence.
Persistence
PostgreSQL and Prisma store company context, customer intelligence, Guide definitions, generation progress and results.
Generation & validation
A staged AI compiler, Zod contracts, source attribution and deterministic checks connect the questions and buyer answers to the result.

Worth mentioning

Getting specific about the customer

I connected Guide creation to research about the selected audience: the problem they’re trying to solve, what triggers a buying decision, what they might choose instead and what would earn their trust. That gives the Guide a clearer purpose than offering everyone the same download.

Making the answers matter

The research helps shape the Guide, and the buyer’s own answers shape their result. I connected the questions to the decision rules so the recommendation responds to their situation and explains why it fits.

Keeping evidence and assumptions separate

The research keeps sources and confidence levels alongside assumptions, contradictions and missing information. Guide creation only includes completed, current intelligence for the selected audience. Claims in the result still need support from the company’s evidence.

Picking up after a failed generation

Building a Guide takes several generation steps. I save progress and validate the output along the way, so a failed step can be recovered without losing completed work. Incomplete or poor-quality results still need to pass review before publishing.