Calibrating AI lead scoring for different industries

Learn how to create separate AI lead scoring profiles for different industry verticals to avoid false negatives and find high-fit B2B customers.

Get started

Short answer

To set up AI lead scoring for different industries, create a separate scoring profile for each vertical. A good lead in healthcare looks nothing like a good lead in software, so one set of criteria will always miss people. Define the titles, company sizes and vocabulary that signal a buyer in each industry, keep those leads in their own group, then audit the scores before you trust them.

Define your industry signals

A lead is qualified when their profile shows they have the budget, the authority and a problem you solve. Those markers change by industry. A compliance lead at a fintech and a plant manager at a food manufacturer both buy software, and nothing on their profiles looks alike.

Software buyers list tools, growth metrics and funding rounds. Manufacturers list certifications, plant size and shift counts. Agencies list practice areas and client logos. Score all three with one set of words and the AI rewards whichever industry you happened to write the profile for, then quietly drops the rest.

Two people with the same title can score differently depending on the words around it. A Head of Operations whose profile mentions multi-site and P&L is a different buyer from one whose profile mentions office management. The profile text is the evidence, so read it before you write the criteria.

Start with your last ten closed deals. Copy the exact job titles, the phrases in their headlines and the words their companies use to describe themselves. Those are your signals. The words you use in a pitch deck are not.

One profile for every vertical

  • Any manager or director title
  • Company size between 10 and 10,000 employees
  • Keywords: growth, scale, strategy

One profile per vertical

  • SaaS: Head of Growth or VP Product, recent funding mentioned in the profile
  • Manufacturing: Plant Manager or Operations Director, ISO or HACCP in the profile
  • Agency: Managing Partner with a named practice area in the headline

Build a profile for each vertical

A scoring profile is a written description of a buyer plus the words that show up on their profile. Write one per vertical and keep each under a page. If you can't read it back and picture one specific person, it's too vague to score against.

Titles carry more weight than most people give them. A Director at a 12-person agency runs delivery. A Director at a 4,000-person bank runs one function inside a much larger machine. Same word, different authority, different score. Write the company size next to the title so the model can tell them apart.

Write the exclusions too. Students, interns, recruiters, competitors and anyone listing themselves as an advisor are noise in every vertical. Naming them up front is faster than cleaning the list afterwards.

  1. 1

    List the titles that actually buy

    Write down the five to ten job titles that signed your last deals in this vertical. Use the exact wording from their profiles, not the wording from your CRM.

  2. 2

    Name the exclusions

    List student, intern, recruiter, competitor and anyone between roles. Add titles that look senior but never hold budget, like advisor or board member at a small company.

  3. 3

    Set a company size range

    Pick the employee count where this vertical typically has budget for your price. Write it as a range with a reason, so you can defend it when a lead scores lower than you expected.

  4. 4

    Add the industry vocabulary

    Copy the certifications, tools and process words that appear in your best customers' profiles. These are the words the model will look for.

  5. 5

    Write one line describing a 90

    Finish with a single sentence describing your best possible lead in this vertical. That sentence is what the score gets measured against.

Map your scoring logic

Not every signal carries the same weight. A Chief Medical Officer at a hospital group is a direct hit. A nurse practitioner is a user, not a buyer. If both land at 80, the list tells you nothing.

Split your signals into three tiers: title authority, company fit and context. Title authority is strongest because it decides who can sign. Company fit decides whether the problem exists at that size. Context covers a funding round, a job change in the last six months or a post about the problem you solve.

Weight the tiers so a title hit can carry a lead over the line on its own and a context hit cannot. Context is a tiebreaker. If a funding round alone pushes someone to 90, every funded company in the wrong industry floods your list.

The same three tiers land differently by vertical. A certification is a strong signal in manufacturing and meaningless in software. A tool name is the reverse.

FinTech

Industry
Compliance Officer or Head of Risk
Strong signal
Product Manager
Weak signal
Junior Analyst

Retail

Industry
Head of E-commerce
Strong signal
Store Manager
Weak signal
Regional Lead

Healthcare

Industry
Chief Medical Officer
Strong signal
Clinic Director
Weak signal
Nurse Practitioner

Manufacturing

Industry
Plant Manager with ISO or HACCP
Strong signal
Operations Director
Weak signal
Quality Inspector

Keep each vertical in its own lead group

The most common mistake is one big list. Apply a manufacturing profile to a software lead and the AI scores them low because they never mention a plant. You delete a good lead and never find out.

Put leads in a group per vertical and point each campaign at the group that matches it. When you edit a profile, only that group's scores move. You can compare the groups later and see which vertical is actually converting.

Some people sit in two verticals. Pick the one that matches the problem you'd pitch them. A former retail buyer who now works at a logistics firm buys logistics software, whatever their history says.

Grouping also makes the outreach easier to write. A message to a plant manager and a message to a Head of Growth shouldn't share a first line, and separate groups keep them apart.

One group, one profile. If you can't say which vertical a group belongs to in a single sentence, split it.

Write the profile as a prompt

If you score with a model rather than a form, the profile is a prompt. Write it as a description of a person. Models score descriptions better than they score long chains of filters.

Say what a 90 looks like, what a 50 looks like and what a 10 looks like. Anchors do more work than adjectives. Senior means nothing to a model. Owns the operations budget for a single plant means something.

Keep the prompt in a file and version it. When you change a weight, you want the old run to compare against.

Paste this job

You are scoring LinkedIn profiles for a B2B seller.

Target buyer: Head of Operations or Plant Manager at a manufacturer with 100 to 800 employees in food or packaging.
Strong signals: plant, production, operations, shift, ISO 9001, HACCP, continuous improvement, lean.
Weak signals: operations analyst, supply chain coordinator, quality inspector.
Noise: student, intern, recruiter, consultant, retired, open to work.

Score each profile from 0 to 100.
90 to 100: strong title, right company size, at least one strong signal in the headline or about section.
50 to 89: right title or right company size, missing the other.
10 to 49: adjacent role or wrong industry.
0 to 9: noise.

Return the score and one sentence naming the signal that decided it.

Audit the scores before you trust them

Read the top twenty and the middle twenty before you trust any of it. The top should be people you'd email today. If they aren't, the profile is too broad or your exclusions are missing.

The middle is where the money hides. Leads scoring 40 to 60 usually match on title but not on company, or the reverse. If those turn out to be good leads, one of your signals is weighted wrong and you're losing them at the fit line.

Recheck monthly. Job titles move. Head of AI barely existed a few years ago and now appears in half the software profiles you'll see. A profile written last year is scoring last year's market.

  1. 1

    Read the top twenty

    If any of them are wrong, copy the trait that fooled the model into your exclusions list and rescore the group.

  2. 2

    Read the middle twenty

    If they're good leads, find the signal that held them back and lower its weight in the profile.

  3. 3

    Check the fit line

    Count how many leads you kept. If you kept almost everything, the line is too low. If you kept almost nothing, the profile is too narrow.

  4. 4

    Rescore on a schedule

    Update titles and vocabulary once a month, rescore the same group and compare the two runs side by side.

Doing this with Kinlead

Kinlead finds people who match the customer you describe, or your website, and scores each lead from 0 to 100 against that profile. Leads below the fit line aren't kept.

Leads live in lead groups, so you can keep one group per vertical and point each campaign at the group with the right profile. Leads can be exported to CSV or imported from a CSV of LinkedIn profile URLs, which means a list you scored somewhere else can come in and be scored again.

Pro costs $49 a month after a 7-day free trial. A card is needed to start the trial, nothing is charged that day, and you can cancel any time in Settings. If you'd rather see the scoring on your own market first, book a demo.

Frequently asked questions

Related

Get started with Kinlead

Prefer markdown? Calibrating AI lead scoring for different industries.md