Building Custom Lead Scoring with Custom Attributes

Most lead scoring systems score the same things: company size, industry, job title, and maybe some behavioral signals like email opens. These are reasonable starting points, but they are generic. They score based on what is easy to measure, not what is most predictive of conversion in your specific market with your specific product.

The teams that build truly effective scoring models go deeper. They identify the attributes that are uniquely predictive for their business: signals that generic CRM scoring tools cannot capture because they do not exist as standard fields. A fintech company may score heavily on whether a prospect's company uses a specific legacy banking system. A developer tools company may score based on the tech stack a company has deployed. A recruiting SaaS may score based on whether a prospect has a talent acquisition team over a certain size.

This article explains how to design, build, and maintain custom lead scoring using custom attributes.

The Architecture of Custom Attribute-Based Scoring

What custom attributes are:

Custom attributes are structured data fields you define beyond the standard lead record fields (name, email, company, phone, etc.). Unlike tags, which are flexible labels, custom attributes have types (text, number, boolean, date, select-from-list) and are stored as queryable, filterable values. They represent structured intelligence about a lead that your standard lead record cannot capture.

Examples of custom attributes used for scoring:

• uses_legacy_system (boolean): does the company use the legacy system your product replaces?
• tech_stack_match_score (number 0-10): how many of the specific tools in your ideal customer's tech stack does this company use?
• inbound_intent_signals (number): how many high-intent behaviors has this lead exhibited (pricing page visits, demo request page views, competitor comparison searches)?
• champion_confirmed (boolean): has a champion been identified within the account?
• budget_confirmed (boolean): has budget been verbally confirmed in a discovery call?
• evaluation_timeline (select: immediate, 30-days, 60-days, 90-days, no-timeline): the buying timeline the lead has indicated

How scoring rules work:

A scoring rule is a condition-weight pair. If a lead matches the condition, add (or subtract) the specified weight to their score. The total score is the sum of all matched rule weights, normalized to a 0-100 scale.

Standard scoring rules look like:

• Industry = Financial Services: +15 points
• Company size = 50-500 employees: +10 points
• Job title = VP or above: +20 points

Custom attribute scoring rules look like:

• uses_legacy_system = true: +25 points
• tech_stack_match_score >= 7: +20 points
• champion_confirmed = true: +35 points
• evaluation_timeline = immediate: +30 points
• evaluation_timeline = no-timeline: -20 points
• budget_confirmed = true: +40 points

Note the dramatically higher weights on custom attribute rules. This is intentional. Custom attributes capture signals that are highly specific to your business and product fit. They are almost always more predictive than generic firmographic signals.

Designing Your Custom Attribute Scoring Model

Step 1: Identify your highest-converting leads and work backward.

Before designing any scoring rules, analyze your historical conversion data. Look at leads that became customers in the past 12 months and ask: what did they have in common that your standard fields do not capture? This exercise typically surfaces 3-5 high-signal attributes that are not currently being tracked.

Common discovery questions:

• What did the customer say during discovery that indicated they were a strong fit? (Capture this as a custom attribute.)
• What technology were they using that made them a natural buyer? (Tech stack attributes.)
• What internal situation or trigger event made your product relevant? (Trigger event attributes.)
• What organizational characteristic (team structure, process, budget cycle) correlated with fast closes? (Organizational attributes.)

Step 2: Define attribute types deliberately.

The type you assign to an attribute determines what scoring logic is possible.

• Boolean (true or false): simplest. The rule is either satisfied or not. Use for binary signals: champion identified, budget confirmed, RFP received.
• Number: enables threshold-based rules (score >= X). Use for counts, percentages, and composite scores: number of intent signals, tech stack match score, decision-maker count.
• Select (enum): enables equality and inequality rules across a defined value set. Use for categorical signals: evaluation timeline, competitor currently used, deal stage in prospect's process.
• Text: searchable but not directly numeric. Useful for notes and identifiers but avoid using text attributes as scoring inputs. Convert text-based signals into the appropriate type.
• Date: enables time-based rules (date within last 30 days). Use for signals like last meaningful touchpoint, demo scheduled date, or RFP deadline.

Step 3: Calibrate weights with historical data.

The weights you assign to each rule determine how much each signal contributes to the final score. The common mistake is assigning weights based on intuition rather than data.

For each candidate scoring rule, calculate the conversion rate of leads that satisfy the condition versus leads that do not. The lift in conversion rate (conversion rate with attribute divided by overall conversion rate) becomes the basis for the weight. An attribute that triples the conversion rate of a lead deserves a much higher weight than an attribute that improves it by 20%.

If historical data is insufficient for statistical calibration, start with intuition-based weights and schedule a calibration review after 90 days of scoring. Compare predicted scores against actual conversion outcomes. Adjust weights for rules where the predicted probability does not match the observed probability.

Step 4: Build negative scoring rules.

Scoring models usually add points for positive signals. They are less often designed to subtract points for disqualifying signals. Negative rules are often the highest-leverage part of the model. Common negative custom attribute rules:

• evaluation_timeline = no-timeline: -20 points (explicitly no buying timeline)
• budget_confirmed = false AND evaluation_stage = late: -10 points (late stage but no budget discussion)
• competitor_locked_in = true: -30 points (actively using a committed competitor)
• company_size = 1: -15 points (solo operator, likely outside ICP)

A lead with a score of 40 that includes negative rules is more accurately scored than a lead with a score of 60 that only applied positive rules. Negative rules improve the model's precision.

Practical Application: Launching Custom Attribute Scoring

1. Interview five recent customers. Ask each one: what was your situation before you bought? What triggered the evaluation? What almost stopped you from buying? What made you decide in our favor? These answers are your custom attribute candidates.

2. List your candidate attributes. From the interviews, pull out the 5-10 most frequently mentioned signals. For each one, decide: can this be captured as a boolean, number, select, or date attribute? If not, break it down further.

3. Define the attributes in your lead system. Create each attribute with the appropriate type. Add a description so future team members know what each attribute means and how to populate it.

4. Define who captures each attribute and when. Some attributes are captured automatically (enrichment, behavioral signals). Others require manual input from sales during discovery. Build this into your discovery call checklist.

5. Set initial weights based on your best intuition. Start with 5-10 scoring rules using your new custom attributes. Assign weights based on your interview findings. Do not wait for perfect data to start scoring.

6. Build the score explanation view. For every lead, display which rules contributed to the score and which were not satisfied. This is what makes sales reps trust and use the score rather than ignoring it.

7. Run your first calibration at 90 days. Pull all leads scored in the past 90 days. For each scoring rule, calculate: what percentage of leads that matched this rule converted versus those that did not? Compare against your assumed conversion lift. Adjust weights where the data diverges from your assumptions.

Maintaining the Scoring Model Over Time

The calibration cycle:

A scoring model built on last year's customer data reflects last year's buyers. As your product evolves and your ICP shifts, the predictive signals change too. Schedule a formal scoring model calibration every six months:

1. Pull all leads scored in the past six months.
2. For each scoring rule, calculate what percentage of leads that matched this rule converted versus those that did not.
3. Compare against the assumed conversion lift embedded in the weight.
4. Adjust weights where the predicted lift does not match observed conversion rates.
5. Identify new signals from recent wins. Are there patterns in recent high-converting leads that the model does not capture?

The dead rule audit:

Over time, scoring rules get added but rarely get removed. Deprecated products, changed market dynamics, and evolving ICPs mean that some rules are actively misleading. Include a dead rule audit in your calibration cycle. Any rule where fewer than 5% of leads have satisfied the condition in the past 90 days is a candidate for review. Either the attribute is not being captured (a data collection problem) or the condition is no longer relevant (a model maintenance problem).

Score transparency:

Scores are most useful when the people using them understand what drives them. A sales rep looking at a lead scored 78 should see which rules contributed and which rules were not satisfied. This score explanation, the decomposition of the total score into its contributing rules, is the difference between a score that a rep trusts and uses to prioritize, and a black-box number they ignore.

Common Mistakes in Custom Scoring

Mistake 1: Building custom attributes without a capture process.

A custom attribute that is never populated cannot influence the score. Before creating an attribute, define exactly how it gets populated: automatically from enrichment, manually by a sales rep during a discovery call, or via a product event. If the capture process does not exist, the attribute will remain empty and the scoring rule will never fire.

Mistake 2: Too many scoring rules firing in the same direction.

If you have 15 positive scoring rules and only 2 negative rules, your model skews toward optimistic scores. Leads that look good on paper but have disqualifying signals get high scores because the positive rules collectively outweigh the few negative signals. Build negative rules for every major disqualifying condition you know about.

Mistake 3: Never calibrating against outcomes.

Setting up a scoring model and running it for two years without calibrating it against conversion data is common. The model drifts further from reality every quarter. Schedule calibration before you finish the initial setup. Put it in your calendar now.

Generic scoring models produce generic results. Custom attribute-based scoring, built on signals specific to your product and market, produces the precision that changes rep behavior and deal outcomes. Build the custom attribute infrastructure, identify the right signals from historical conversion data, and maintain a calibration cycle. Your scoring model should improve every six months. If it is not, you are not running the calibration.