AI Agents
AI lead qualification: scoring accounts with LLMs
How to use LLMs to qualify B2B accounts and leads: inputs, criteria, structured scoring, combining with rule-based scores, and validating accuracy.
Founder, OretonGTM · ·1 min read
Rule-based scoring is great for clean fields like headcount or country. It struggles with questions like “does this company sell to mid-market finance teams?” That’s where an LLM helps.
What AI can judge that rules can’t
- Business model from website copy (B2B or B2C, self-serve or sales-led)
- Target market described in their own words
- Whether job posts suggest the problem you solve
- Product complexity or category fit
The qualification prompt
Provide:
- Your ICP criteria in plain language
- The account’s data: website text, job posts, firmographics
- Output format: score (1 to 5), criteria met, criteria not met, evidence quotes, confidence
- Rules: only use inputs; return low confidence when data is thin
Combine with rules
| Layer | Method | Example |
|---|---|---|
| Hard filters | Rules | Country, headcount range, exclusions |
| Fit points | Rules | Industry, stack |
| Qualitative fit | AI | Business model, target market |
| Signals | Rules + AI | Hiring relevance, news |
Rules filter first (cheap), AI judges the remaining accounts (more expensive). See account scoring model.
My take: Ask the model for evidence with every score. A score without a quote from the inputs is an opinion you can’t audit.
Validate it
- Run the model on 50 to 100 past closed-won and closed-lost accounts
- Compare scores with outcomes
- Read the reasoning on mismatches and refine criteria
- Re-check every quarter
Act on the output
Scores should change routing: tier assignment, sequence choice and whether a rep gets an alert. More on that in lead routing.
Frequently asked questions
Can AI qualify B2B leads?
AI can assess fit from unstructured data like website copy and job posts, which complements rule-based scoring. Final qualification of interest and timing still happens in conversations.
How accurate is AI lead scoring?
It depends on the inputs and criteria. Test it on past closed-won and closed-lost accounts and adjust the prompt until scores match reality.
Should AI replace rule-based scoring?
No. Rules handle clear data like headcount and location reliably. AI adds judgment on messy inputs. Combining both works best.