Explainable AI for Lead Qualification: An Analysis of QuickOn

The key challenge in AI-powered lead scoring isn’t speed—it’s trust. Speed can be handled by any webhook, but explaining why a particular lead received a score of 92 out of 100 is much more difficult. QuickOn has built its product around this challenge.

The architecture consists of five stages: acquiring a lead, normalizing and removing duplicates, enriching the data, scoring, and syncing with the CRM. During the scoring stage, a 10-criteria rubric linked to your Ideal Customer Profile is applied. Objective criteria—company size, industry, and geography—are evaluated deterministically in the code. Subjective criteria—purchase intent and urgency—are handled by a selected AI model.

This separation is the most interesting engineering solution in the product. Instead of a “black box” that spits out a number, you get a transparent breakdown for each criterion: ICP match, decision-maker authority, purchase intent, and so on. For each qualified lead, the system generates not only a 0–100 score and an A–D grade, but also a rationale, risks, a recommended next action, and a draft outreach email.

A provider-agnostic approach is consistently implemented here. The free plan uses open-source models via Groq (Llama 3.3, GPT-OSS, Qwen) or a local Ollama instance. Paid options—Claude, GPT, and Gemini—are integrated on a per-organization basis without any code. Data enrichment is handled via People Data Labs, Clearbit, or Apollo, with caching to avoid duplicate calls.

The pipeline is built with a focus on reliability: each step is idempotent and retries independently, ensuring that leads are not lost. Security features include Postgres row-level security, encryption of secrets, hashing of API keys, rate limiting, and verification of webhook signatures.

There’s a trade-off when it comes to CRM integrations. Only HubSpot works out of the box. Pipedrive, Salesforce, Zoho, and Freshsales require custom configuration. For teams already using HubSpot, this isn’t a problem. For ecosystems using other CRMs, it means extra work.

Pricing is transparent: $0 for 100 leads per month using open-source AI, then $49 for 1,000 leads, and so on. Data enrichment is billed separately so there are no surprises on your bill.

The main question with such systems is to what extent the rubric and AI scoring actually correlate with closed deals. QuickOn provides tools to audit every decision, which already sets it apart from the “trust the model” approach. But the only real test is an A/B test on your own data: comparing conversion rates to closed deals between manual qualification and the platform.

Source: Hacker News - Newest: ""AI" "LLM""