Lead Score Thresholds and Routing Bands: How to Set Them

A lead score does nothing until a threshold turns it into an action. Most teams copy the cutoffs from a template, launch, and never check whether the leads above the line actually convert better than the ones below it. This guide covers the cutoffs alone: how to set them, test them and keep them honest. For the model that produces the score, start with the lead scoring guide.

What is a lead score threshold?

A threshold is the score at which a lead changes band, and a band is a promise: who owns the lead, what happens next and how fast. If two bands get the same treatment, one of the cutoffs is not doing any work.

These are the illustrative defaults from the fit plus intent model. Your CRM may ship its own: HubSpot's lead scoring tool (knowledge base, updated September 2026) labels engagement and fit scores High, Medium and Low with an example of 70 to 100 as High and 40 to 69 as Medium, and combines the two into a nine-cell A1 to C3 grid. Either set is a starting point, not an answer.

How do you set lead score thresholds from your own data?

Put the cutoff where conversion actually changes. Bucket past leads by score, compare outcome rates per bucket, and draw the Hot line at the lowest bucket where the rate clearly and consistently beats the average.

1. Export scores as they were at routing time

Pull the last two to four quarters of leads with the score each one had when it was routed, plus three outcomes: meeting held, opportunity created and closed-won. Today's score is the wrong input, because decay and new activity have moved it since. If your CRM does not keep score history, start stamping a score_at_routing field now and calibrate next quarter.

2. Compute conversion per score bucket

Group leads into 10-point buckets and compute the meeting rate and the opportunity rate for each. Use meetings as the primary signal (there are more of them, so the read is steadier) and opportunities as the check. Closed-won is the slowest outcome; use it to confirm the order of the bands, not to place the line.

3. Place the line where the rate steps up

In a working model the rate rises with the score and usually jumps at one point. That jump is your Hot cutoff. If the rates are flat across buckets, no cutoff will fix it; the weights are wrong, and the fix is in validating the ICP against closed-won data before you touch the thresholds.

Every cutoff trades two errors. The Google Machine Learning Crash Course (updated January 2026) puts it plainly for any classifier: raising the threshold generally cuts false positives but adds false negatives. For a lead score, a false positive is a rep chasing a lead that will never buy, and a false negative is a buyer left in nurture. Decide which costs you more before you move the line.

How do you check a threshold against sales capacity?

A Hot band your reps cannot work inside the response target is not a Hot band. Multiply the expected Hot volume by the time a first touch takes and compare it with the hours reps actually have for inbound.

Illustrative example: if the cutoff from step 3 would put 60 leads a day in Hot, each first touch takes 15 minutes, and three reps each have two hours a day for inbound, the band needs 15 hours against 6 available. Either the cutoff rises until Hot volume fits, or the Warm band takes the overflow with a slower target. Do not keep the low cutoff and let the SLA quietly fail.

Capacity runs the other way too. If the Hot band is nearly empty and reps have idle inbound hours, lower the cutoff one bucket at a time and watch whether the newly promoted bucket still converts above the average.

Should fit and intent have separate thresholds?

Yes. A single combined cutoff lets heavy engagement from a poor-fit contact, or a perfect-fit account that never engaged, land in the same band. Add a fit floor that must be met before any lead routes as Hot.

• Fit floor: for example, at least 30 of the 50 fit points. A student, a job seeker or a competitor can rack up intent points; the floor keeps them out of the sales queue. The fit criteria themselves come from your Pain-First ICP scoring.
• Hand-raise override: a demo or contact-sales request from a lead above the fit floor routes as Hot whatever the total, because the buyer has told you what they want.
• High fit, low intent: these are target accounts, not inbound leads. Send them to account-based outreach rather than waiting for their intent score to climb.

The split between fit and intent signals is covered in behavioral vs demographic scoring.

How do you stop leads bouncing between bands?

Give each band an exit point below its entry point. A lead enters Hot at 80 and leaves only when it drops below 70, so a single unsubscribe or a week of decay does not flip ownership back and forth.

• Entry and exit cutoffs: a gap of about 10 points between them is a reasonable start; widen it if reassignments stay frequent.
• Route on the band, not the raw score: store the band as its own field and trigger workflows on a band change only. The assignment and SLA ladder that follows lives in BDR queue routing from lead scores.
• Owned leads stay owned: once a rep has made contact, a score drop changes the band label for reporting but does not take the lead away. Decay is for unworked leads; see lead score decay.

How often should you recalibrate lead score thresholds?

Monthly for the first quarter after launch, then quarterly, plus an off-cycle review whenever a trigger fires. Change one cutoff at a time so you can tell what moved the numbers.

• Hot share moves sharply: a new form, a campaign or a scoring rule change has inflated or starved the band.
• Hot-to-meeting rate drops: the cutoff is too low or one action is over-weighted.
• Warm leads keep closing: sales is finding buyers below the line, so the cutoff is too high or the fit floor is too strict.
• The ICP changes: new segment, new product, new price point. Re-run the buckets before trusting the old line.

Keep a threshold changelog: the date, the old and new cutoffs, the reason and the next review date. Agree it with sales as part of the MQL-to-SQL handoff, because a threshold sales does not trust is a threshold sales ignores.

What are the most common threshold mistakes?

• Copying template cutoffs and never comparing them with conversion.
• Calibrating on today's scores instead of the score at routing time.
• One combined cutoff with no fit floor, so engaged non-buyers reach sales.
• No exit point, so leads flip owners on every small score change.
• A Hot band larger than capacity, which turns a 5-minute target into a same-week one.
• Changing weights and cutoffs at once, so nobody can say what improved or broke.

How do you put thresholds into your CRM?

Store three fields on the lead: the total score, the band, and the score at routing time. Workflows set the band using the entry and exit cutoffs, fire only on a band change, and stamp score_at_routing when the lead is assigned. That last field is what makes next quarter's calibration possible. Ready-made property sets are in the HubSpot and Salesforce lead scoring templates.

The Lead Scoring agent builds the fit and intent model, thresholds, decay and CRM workflow inside your own Claude; use the steps on this page to tune its cutoffs once outcomes come in. For what leaky thresholds are costing you now, see your leaks priced in dollars.

Frequently asked questions

Sources and references

• Google for Developers, Machine Learning Crash Course, "Thresholds and the confusion matrix" (updated January 2026): raising a threshold reduces false positives and increases false negatives.
• HubSpot Knowledge Base, "Overview of the lead scoring tool" (updated September 2026): High, Medium and Low score thresholds and the A1 to C3 fit and engagement grid.
• Brian J. Carroll, Lead Generation for the Complex Sale (McGraw-Hill): lead qualification and nurture for long buying cycles.
• Mark Roberge, The Sales Acceleration Formula (Wiley): agreeing lead definitions and service levels between marketing and sales.