What I learned building a deterministic lead scoring API

Over the last months, I’ve been working on a lead scoring API.

Not a CRM.
Not a marketing tool.
Just a system that receives leads and answers one question consistently:

How good is this lead, based on predefined criteria?

At first, I assumed this would naturally involve AI or machine learning.
But the more I talked with sales ops teams and agencies, the clearer one thing became:

They didn’t want intelligence.
They wanted reliability.

The real problem wasn’t lead quality

It was inconsistency.

Same lead coming from different sources.
Different formats.
Different rules applied by different people.
Different results depending on timing or tooling.

That breaks reporting, automations, and trust in the system.

So I made a deliberate decision early on:
deterministic scoring only.

Same input → same score → same flags → recommended action.
Every time.

No model drift.
No hidden behavior.
No surprises.

What the API does

• Accepts any JSON structure (no fixed schema)
• Maps fields automatically
• Applies predefined quality rules
• Returns:
  • a score (0–100)
  • flags explaining deductions
  • a clear recommended action

It’s designed to sit in front of CRMs or sales pipelines, not replace them.

The main takeaway

Not everything needs AI.

Some systems are more valuable when they are:

• predictable
• explainable
• boring (in a good way)

Building something boring but trustworthy turned out to be harder — and more interesting — than adding a model and calling it “smart”.

If you’re building internal tools, scoring systems, or anything operations depend on, this tradeoff is worth thinking about.

If anyone wants to see the docs or how the API works in practice:
https://www.postman.com/leadflags

Comments

[Billie Jean]

Hi, I am working on something similar, would love to connect and talk more!
Followed.