I recently watched a sales rep named Marcus blow off an inbound lead with a lead score of 94 out of a possible 100.
We were doing pipeline review and his manager pulled up the lead score column to ask why he hadn't touched the hottest MQL in the queue.
Marcus had spent his time on a 61 instead.
The 61 was a Series B company nobody had heard of, a lower-level contact, fewer page views. On paper it was a worse lead in every measurable way. His manager asked him to explain why, and Marcus pulled up LinkedIn. The 94's founder had posted three weeks earlier about a hiring freeze. Their last funding round was 2021. Glassdoor was a hazmat zone. The 61, on the other hand, had just hired a VP of RevOps, their CEO was posting twice a week about scaling go-to-market, and they'd quietly added Gong and Clay to their stack, which you could see in their job listings.
The manager insisted Marcus reach to the 94, despite what his research indicated. He closed the 61 that quarter. The 94 never returned a call.
Here's the thing about that pipeline meeting: the model wasn't wrong about the data it had. It was wrong about the data it didn't have. Marcus qualified that deal the way every good rep qualifies a deal: by reading the company. The score qualified it the way software reads a record: by counting things. And the gap between reading and counting is the entire reason your best reps quietly ignore the number you spent six weeks building a calculation for.
For the first time, we can close that gap inside HubSpot. Not with a better scoring model, but with no scoring model at all.
What Traditional Lead Scoring Measures (and Misses)
Pull up any lead scoring model and you'll find some version of the same ingredients.
Title keywords
Company size and industry
Revenue and headcount
Page views (especially pricing and demo pages)
Email opens and clicks
Form fills and content downloads.
Add it up, weight it, and you get a number.
Every one of those inputs is a proxy. A form fill is a proxy for interest. A pricing page view is a proxy for intent. A "VP" in the title is a proxy for authority. None of them is the thing you care about. And yet, we built scoring models around them.
And to be clear, that wasn't a mistake. Proxies were the best we had. You cannot build a deterministic rule that reads a founder's LinkedIn post and infers a hiring freeze. You cannot write a workflow filter for "their last funding round was four years ago and the energy feels off." So we counted email opens, because email opens are countable, and sometimes that count was correlated with buying interest.
Most teams are still optimizing the count. They tune weights, add decay, A/B test thresholds, polishing a measurement system built entirely on surface signals. The model is only as good as the signals feeding it, and the signals have always been a compromise.
That's the gap. Your scoring model measures what's easy to count. Your reps act on what's hard to read. The deals that matter live in the second category.
What Smart Properties Do
HubSpot Smart Properties are CRM fields filled by AI instead of by a human, a form fill, or an enrichment tool. You write a prompt, point it at a data source, and the Breeze Data Agent generates the value and writes it to the record.
When you create a Smart Property, you choose where the agent looks: Web research (the open web), Company website (their site specifically), Property data (other fields and associated records already in HubSpot), or Call transcripts (your recorded conversations). For the first time, a CRM field can be populated by something reading unstructured data and reasoning about it. Which is the exact move your reps take manually.
Two caveats before you get excited:
They are not real-time. A Smart Property fills when you tell it to. That can be on demand, on record creation, or on a schedule via workflow. It is not watching LinkedIn for you. Think of it as a research assistant you dispatch, not a live feed. If a company's situation changes the day after you fill the property, the property doesn't know until you tell HubSpot to refill it.
They cost credits. Every fill burns HubSpot Credits, and at scale that's a number with dollars attached. We're going to spend a whole section on credits later in this post, because it's the difference between a clever enrichment strategy and a $5,000 surprise on your invoice.
Four Smart Property Prompts That Beat Your Lead Score
These are written for the Breeze Data Agent and built to be copy-paste-ready. Before I share the actual prompts, here is one prompting rule I follow, and you should too.
Pin the output to exact allowed values. The agent will happily write you a paragraph when you wanted one word. So tell it the only acceptable outputs and forbid everything else. Pair every rating property with a companion "Reasoning" Smart Property that captures the why in plain language, so reps get a verdict they can act on and an explanation they can argue with. Set the field type to match: single-line text or a dropdown for the rating, multi-line text for the reasoning.
Now for the fun part.
1. LinkedIn Activity Index
Employee-level LinkedIn activity is one of the most predictive and most ignored buying signals in B2B. A company where the VP of Sales, three AEs, and the founder all post weekly is a company that buys software to grow. A company with radio silence is a company in survival mode. I look at this type of signal extensively in my work as Head of RevOps at FirstTouch.
Data source: Web research.
CONTEXT:
You are evaluating a company to determine how actively its employees
participate on LinkedIn. High employee LinkedIn activity is a strong
signal of a growth-oriented, marketing-mature organization that invests
in go-to-market. You are researching the company named in this record.
Company: {{company.name}}
Domain: {{company.domain}}
WHAT TO RESEARCH:
- Whether individual employees (not just the company page) post original
content, comment, or share on LinkedIn with any regularity.
- The apparent cadence: do multiple employees post weekly/monthly, or is
the account dormant?
- Whether leadership (founders, VPs, C-suite) are personally active.
- Distinguish brand-page posting from genuine individual employee activity.
Individual activity counts far more than brand-account activity.
SCORING GUIDE:
- High: Multiple employees, including leadership, post original content
regularly (roughly weekly). Active commenting and engagement.
- Medium: Some employee activity, sporadic. A few people post occasionally;
most are passive.
- Low: Little to no individual employee activity. Brand page may post, but
humans are largely silent.
- Unknown: Insufficient public information to make a determination.
OUTPUT FORMAT:
Return exactly one of these four values and nothing else:
High
Medium
Low
Unknown
Do not include reasoning, punctuation, explanation, or any other text.
Output a single value only.2. Estimated Sales Cycle Length
This one tells a rep how to pace the deal before they've spoken to anyone. A 180-day enterprise procurement slog and a 30-day founder-led purchase get worked completely differently, and most reps don't find out which one they're in until week three.
Data source: Web research and Property data
CONTEXT:
You are estimating how long a typical sales cycle would be for this company
as a B2B buyer. Use company size, industry buying norms, procurement
sophistication, and apparent tool-stack maturity to infer the likely length
of their purchasing process.
Company: {{company.name}}
Domain: {{company.domain}}
Employee count: {{company.numberofemployees}}
Industry: {{company.industry}}
WHAT TO RESEARCH:
- Company size and structure. Larger orgs and regulated industries imply
longer procurement, security review, and legal cycles.
- Industry norms for software purchasing (e.g. healthcare/finance/government
skew long; startups and SMB tech skew short).
- Procurement signals: dedicated procurement/vendor-management functions,
security/compliance pages (SOC 2, ISO), or RFP-driven buying behavior.
- Tool-stack maturity: a sophisticated, modern stack often correlates with
faster, more confident buying decisions.
OUTPUT FORMAT:
Return exactly one of these five values and nothing else:
Under 30 days
30-60 days
60-180 days
180+ days
Unknown
Do not include reasoning, ranges, hedging, or any other text.
Output a single value only.3. Events and Webinars Cadence
A company that runs its own webinars and events has a marketing budget, a content engine, and a reason to talk to a vendor like you. A company that has never hosted anything may not have the funds to buy.
Data source: Web research plus Company website.
CONTEXT:
You are determining whether this company runs marketing events, webinars,
or virtual programming. This signals marketing maturity and budget.
Company: {{company.name}}
Domain: {{company.domain}}
WHAT TO RESEARCH:
- Whether the company hosts its own webinars, virtual events, conferences,
workshops, or recurring programming (check the website's events/resources
pages and recent public activity).
- The cadence: ongoing series vs. one-off vs. nothing.
- Whether they only SPONSOR or appear at others' events rather than hosting
their own. Sponsoring is different from hosting and should be classified
separately.
SCORING GUIDE:
- Frequent: Runs its own recurring events/webinars (e.g. a regular series
or multiple events per quarter).
- Occasional: Hosts its own events sporadically; no clear ongoing cadence.
- Sponsors Only: Appears at or sponsors others' events but does not host.
- None: No evidence of hosting or sponsoring events.
- Unknown: Insufficient public information to determine.
OUTPUT FORMAT:
Return exactly one of these five values and nothing else:
Frequent
Occasional
Sponsors Only
None
Unknown
Do not include reasoning, explanation, or any other text.
Output a single value only.4. Recent Engagement Intent Score
This one reads the activity data you already own — and it's where your old scoring model goes to die.
Data source: Property data and associated engagement records.
CONTEXT:
You are scoring the recent buying intent of a company based ONLY on its
engagement activity inside our CRM over the last 90 days. Do not use web
research. Reason about patterns in the activity, not just raw counts.
Company: {{company.name}}
Use associated contacts' recent activity, page views, email engagement,
form submissions, and meeting/call records from the last 90 days.
WHAT TO EVALUATE:
- Multi-contact clustering: are MULTIPLE people from the same company
engaging in the same window? Coordinated engagement from 3+ contacts is
a far stronger signal than one person clicking repeatedly.
- High-intent page activity: visits to pricing, demo-request, comparison,
or implementation/onboarding pages weigh more than blog or top-of-funnel.
- Timing and acceleration: is engagement increasing recently, or is it old
activity that has gone cold? Recent and accelerating beats high-but-stale.
- Recency: weight the last 30 days more heavily than days 31-90.
SCORING GUIDE:
- 80-100: Multiple contacts, high-intent pages, accelerating in last 30 days.
- 50-79: Solid recent engagement OR multi-contact, but missing one element.
- 20-49: Some engagement, single contact, or cooling off.
- 0-19: Minimal, stale, or low-intent activity only.
OUTPUT FORMAT:
Return a single integer between 0 and 100 and nothing else.
Do not include a percent sign, reasoning, label, or any other text.
Output one integer only.Or: Build a Score They Won't Ignore
Everything above is modular by design: four signals, four columns, mix and match. But if you want to simplify things, how about one Smart Property that does the whole qualification in a single pass and outputs a verdict your reps can act on immediately.
Feed it your ICP. Let it research. Have it return a score, the reasoning behind the fit, the timing read, the single strongest signal it found, and a recommended next action, all output as structured JSON.
Field type: multi-line text.
Data source: Web research and Property data.
CONTEXT:
You are a senior B2B sales qualifier. Evaluate this company against our
Ideal Customer Profile and return a single qualification verdict. Research
the open web and use the CRM data provided. Reason like a rep doing manual
discovery, not like a scoring model summing points.
OUR ICP:
- Target: B2B SaaS and tech-enabled services, 50-1,000 employees.
- Strong fit: growth-oriented, marketing-mature, modern GTM tool stack,
active leadership, recent funding or hiring momentum.
- Weak fit: stagnant headcount, dormant online presence, signs of
contraction (layoffs, hiring freezes, leadership churn).
Company: {{company.name}}
Domain: {{company.domain}}
Employee count: {{company.numberofemployees}}
Industry: {{company.industry}}
WHAT TO RESEARCH AND WEIGH:
- Fit: size, industry, and ICP alignment.
- Momentum: recent funding, hiring, leadership additions, or product launches
vs. layoffs, freezes, or churn.
- Marketing maturity: employee LinkedIn activity, events/webinars, content
operation, modern stack.
- Timing: likely buying-cycle length and any signals of active evaluation.
OUTPUT FORMAT:
Return ONLY valid JSON in exactly this structure, with no markdown fences,
no preamble, and no trailing text:
{
"score": <integer 0-100>,
"fit_summary": "<one sentence on ICP fit>",
"timing_summary": "<one sentence on buying timing/cycle>",
"top_signal": "<the single strongest positive or negative signal found>",
"recommended_action": "<one of: Prioritize now | Nurture | Monitor | Disqualify>"
}
The recommended_action MUST be exactly one of the four listed values.
Output the JSON object only.Run this on every new MQL. Refresh it monthly on target accounts. You can read the JSON on the record as-is, split each key into its own companion Smart Property for cleaner reporting, or use it as a workflow action that routes the record based on recommended_action.
How to Manage Credit Spend
Everything above is cheap per record and ruinous at scale. You need to understand the meter before you turn it on.
A Smart Property fill costs 10 credits. Credits run $10 per 1,000, so one fill is about ten cents per record. That's the number to keep in your head: a dime a fill.
Your plan comes with a monthly allotment:
Starter includes 500
Pro includes 3,000
Enterprise includes 5,000
Credits do not roll over. They reset on your usage-period start date and unused ones expire. So your budget is your included credits plus whatever extra credits you've bought.
Worth noting too: all your AI features share one credit pool. Heavy Smart Property usage in marketing means fewer credits for a sales team's Prospecting Agent. Decide who owns the budget before someone burns it on day one.
What not to do
When you create a Smart Property, HubSpot will helpfully offer to fill it across your existing records right there in the setup flow.
Decline.
Create the property empty. Then enroll records through a workflow instead. The bulk-fill button is exactly how the expensive mistakes happen.
Picture 50,000 companies in your CRM and one Smart Property you're excited about. You create the property and hit Smart fill.
50,000 records × 10 credits = 500,000 credits = $5,000. One click.
Now imagine you built all four properties from this newsletter and ran them across the whole database: 50,000 × 4 × 10 = 2,000,000 credits = $20,000.
I hope I'm wrong about how often this happens. I don't think I am.
Who To Enroll, and Who Not To
The fix isn't running less research. It's running research on the right records. Three tiers:
Always enrich: MQLs and named target accounts. These are the records where a dime of research drives decisions. Spend freely here.
Conditionally enrich: Accounts showing fresh engagement. They've crossed a threshold worth a closer look, but aren't automatic. Gate these behind a trigger.
Don't enrich: Cold records, out-of-ICP companies, and anything you'd disqualify on a deterministic rule anyway. Researching a record you'd never work is setting money on fire, at volume.
Three Workflow Patterns Worth Building
Stop bulk-filling. Enroll through workflows so you control the pace and never touch a record that doesn't qualify. Use the Data Agent: Fill Smart Property workflow action. It respects the property's configured data source, including web research.
New MQL Enrichment. Trigger: contact reaches MQL lifecycle stage. Filter: company is in-ICP and has a valid domain. Action: fill LinkedIn Activity Index, Estimated Sales Cycle Length, and their reasoning companions on the associated company.
High-Intent Engagement Refresh. Trigger: contact views pricing or demo page, or 3+ contacts from one company engage within 7 days. Filter: company not enriched in the last 14 days. Action: refill Recent Engagement Intent Score and reasoning.
Target Account Quarterly Refresh. Trigger: scheduled, quarterly. Filter: company is on the named target-account list. Action: refill all web-research properties so the research stays current on the accounts that matter.
What This Replaces (and What It Doesn't)
Smart Properties replaces proxies with intelligence and provides reasoning for it.
Keep your deterministic rules for the hard edges. Geography you don't sell into, roles that can't buy, active competitors. Those are binary disqualifiers, they cost nothing, and you don't need to pay credits into an LLM relitigating them. What you gain in the trade is the thing our sales rep, Marcus, had and the legacy lead scoring model didn't: a verdict with reasoning attached. Something a rep can read, agree or disagree with, override, and learn from. The old number was a black box that reps never trusted. The new one is a colleague's opinion they can check against their own.
So here's the experiment. Take your next 100 MQLs. Run them through your current scoring model and write down the number. Then run them through the consolidated Smart Property and write down that one too. Put both in front of your reps for a week.
Watch which number they act on without being told to.
That's your answer about which one is worth sticking with.
Best, Ryan
