Most SaaS revenue teams still run scoring models built for a world of static content downloads and quarterly newsletters, not one where prospects self-serve through a trial before a rep ever speaks to them. When a lead scoring model was designed for outbound and email nurture, it tends to reward attention rather than intent, and that mismatch is why so many “qualified” leads stall the moment an SDR picks up the phone. This post sets out a framework RevOps leaders can use to separate fit from behaviour, build scoring from real conversion history rather than guesswork, and route effort towards the accounts most likely to close.
Why Traditional Lead Scoring Models Fail Modern SaaS Teams
Classic point-based lead scoring assigns a fixed value to each action: five points for an email open, ten for a webinar attendance, twenty for a form fill. The scores accumulate over time, and a lead who has been on a nurture list for a year and opened dozens of emails can easily outrank someone who registered for a trial yesterday. Total points alone say nothing about whether the number is high because the account matches the ideal customer profile, or because the same person has clicked every newsletter link for months without ever engaging the product. Handed a list sorted by that single number, an SDR has no way to tell those two situations apart.
This model was largely inherited from a marketing motion built around gated content and email cadence, where the only signals available were content interactions. Product-led SaaS companies now generate a much richer stream of behavioural data: trial activations, feature adoption, seat invitations, integration requests. Folding all of that into one composite score, using the same arbitrary point logic that once scored a whitepaper download, throws away the information that actually distinguishes a ready buyer from an idle browser. The result is a familiar pattern: pipeline volume looks healthy on a dashboard, but conversion at the discovery-call stage tells a different story, because the qualification step never separated who the account is from what the account is doing.
Splitting Fit Scoring From Behaviour Scoring
The structural fix is to stop treating fit and behaviour as one number and instead track them as two independent axes that are only combined at the point of defining a sales-qualified lead. Fit describes whether an account resembles your best customers on attributes that rarely change week to week. Behaviour describes what an individual or buying group is actually doing right now, which changes constantly and should be allowed to decay when activity stops.
What Counts as Fit
Fit scoring draws on firmographic data such as industry, employee count and revenue band, technographic data such as the tools already in a prospect’s stack, and any explicit ICP criteria a company has defined, such as a minimum team size for the product to deliver value. Because these attributes are stable, fit scores are usually maintained as CRM properties updated through enrichment rather than recalculated on every interaction. A large enterprise account can score extremely well on fit purely by matching the target profile, and that says nothing yet about whether anyone there intends to buy.
What Counts as Behaviour
Behaviour scoring tracks readiness-to-buy actions: trial activation, depth of feature adoption, multiple named users logging in independently, integrations being connected, and specific high-friction pages such as pricing or security documentation being revisited. Unlike fit, behaviour should decay over time if nothing new happens, because a burst of activity two months ago that has since gone cold is a weaker signal than the same score would suggest if it were still accumulating. A smaller account that scores modestly on fit can still show strong, current behavioural signals, such as a trial user inviting teammates and connecting a billing integration in the same week, and that combination is often a better predictor of an imminent deal than fit alone.
Keeping the two as separate CRM fields, merged only when an SQL threshold is defined, gives sales and RevOps a shared vocabulary for why a lead is flagged: is this account well matched but passive, or a weaker match showing urgent intent? Those two situations call for different plays, and a single blended score erases the distinction entirely.
Using Historical CRM Data to Build Predictive Scoring
Point values assigned by committee are guesses. A more defensible approach is to mine closed-won and closed-lost history in the CRM for the behavioural attributes that actually preceded conversion, then weight the scoring model on that evidence rather than intuition. This is a straightforward retrospective analysis: pull the trial and engagement history for every closed opportunity from the past year or two, segment by outcome, and look for the actions that show up disproportionately in the won cohort.
The main trap in this exercise is survivorship bias. If you only study the won deals, every behaviour looks predictive, because you have no comparison against similar accounts that showed the same behaviour and never converted. A rigorous version of this analysis compares won opportunities against a matched sample of lost or no-decision opportunities with similar fit scores, so that the behavioural difference between the two groups is what actually drives the new weighting, not just a list of things winners happened to do.
A scoring model built this way still needs to be revisited on a schedule, because SaaS buying patterns move as fast as the product does. A pricing change, a new packaging tier, or a shift in the ideal buyer persona can make last year’s weighting stale within a quarter, and a model that isn’t recalibrated will keep confidently ranking leads on signals that no longer correlate with revenue. HubSpot’s own documentation on managing custom contact and company properties is a useful reference point for teams building this kind of scoring logic directly into CRM fields (developers.hubspot.com/docs/api/overview).
Why Intent Signals Beat Vanity Engagement
Vanity engagement costs a prospect nothing: opening an email, clicking a link, sitting through a webinar are all low-friction actions that require no real commitment. Intent signals, by contrast, require the prospect to spend something, whether that is time, internal political capital, or exposure to a colleague. Starting a trial, inviting a second user, requesting a specific integration, or repeatedly returning to a security or pricing page are all actions that cost the prospect effort, and that effort is precisely why they correlate more strongly with genuine evaluation.
Recency matters as much as the action itself. A cluster of high-intent behaviours within a short window, such as several colleagues logging into a trial in the same few days, usually indicates a buying committee assembling and moving quickly. The same total activity spread thinly across several months is a much weaker signal, even if the raw point total ends up identical under a naive scoring model that never accounts for timing.
Because intent signals are time-sensitive, scoring them is only useful if the routing is near real time. A nightly batch job that recalculates scores overnight means a prospect who showed strong intent at 9am is not flagged to an SDR until the following morning, by which point the buying window may have narrowed. Workflow automation tools such as n8n can watch for specific trigger events, such as a second user seat being activated, and push an immediate alert into the CRM or a Slack channel rather than waiting for the next scheduled sync (docs.n8n.io).
Designing a RevOps-Ready Scoring Framework
Bringing these pieces together means embedding three things into daily operations rather than a one-off configuration project: separated fit and behaviour scoring, weighting derived from historical conversion analysis rather than guesswork, and routing logic that prioritises current, high-effort behavioural signals over cumulative low-effort ones. The diagram below shows how the two scoring tracks feed a single SQL threshold decision, which then branches an account either to the AE queue for immediate follow-up or into a nurture track if the threshold isn’t met.
Governance Steps That Keep Scoring Honest
A scoring model earns trust through governance, not through a clever initial build. Four practices keep it accurate over time. First, document the SQL threshold formula explicitly, stating the minimum fit score and minimum current behaviour score required, rather than leaving the combination rule as tribal knowledge. Second, apply decay rules to the behaviour score so that stale activity stops inflating rankings months after the fact. Third, build a feedback loop where sales can flag a lead as a false positive directly in the CRM, and RevOps reviews those flags on a regular cadence to adjust weighting. Fourth, treat any predictive or machine-learning layer as an input to be reviewed rather than a black box to be trusted; a model can surface correlations that drift as the market shifts, and a human review catches that drift before it wastes a quarter of pipeline.
One governance question RevOps leaders face repeatedly is what to do with a low-fit account that shows strong intent. The right answer depends on the underlying revenue model rather than the scoring framework itself. If the business can profitably serve smaller accounts, a low-fit, high-intent lead is worth an SDR’s time regardless of what the fit score says. If the model depends on large enterprise contracts to hit efficient unit economics, that same lead is better held in a lighter nurture track so sales capacity stays focused on accounts that can support the target deal size. The scoring framework should surface this tension clearly rather than resolve it silently by burying a low-fit account under a blended score.
Behavioural scoring also touches data protection obligations, since it typically involves tracking individual-level activity across marketing and product tools and combining it into a profile used to prioritise outreach. UK organisations processing this kind of personal data should have a clear lawful basis and be able to explain the logic to a data subject if asked, and the ICO’s guidance for organisations is a sensible starting reference when designing or auditing this part of the pipeline (ico.org.uk/for-organisations/).
Related Reading
For more on this, see more on lead generation and outreach, including Salesloft’s SDR to AE Shift: Restructuring SaaS Sales for 2025, Understanding Lead Generation: What Is It and Why It Matters for Startups, and LinkedIn Lead Gen Forms for Hospitality SaaS.
Frequently Asked Questions
What is the difference between a fit score and a behaviour score?
A fit score reflects how closely an account matches the ideal customer profile on stable attributes such as industry, employee count and technographic data, and it rarely changes week to week. A behaviour score reflects current readiness-to-buy actions such as trial activation, feature adoption and multi-user engagement, and it should decay over time if activity stops. Keeping them as separate CRM fields and only combining them at the SQL threshold prevents a well-matched but passive account from being treated the same as one showing active intent.
How often should a lead scoring model be recalibrated?
There is no fixed interval that suits every company. Recalibrate whenever pricing, packaging or the target segment changes materially, and review the weighting on at least a quarterly cadence even if nothing obvious has changed, since buying patterns can drift gradually rather than in one visible event.
Should SDRs pursue low-fit accounts that show strong intent?
It depends on the revenue model rather than the scoring framework itself. If the business can serve smaller accounts profitably, strong intent from a low-fit account is worth engaging. If the model depends on large enterprise deals to hit target unit economics, that same account is better held in a lighter nurture track so sales capacity stays focused on accounts of the right size.
Can AI or predictive models replace manual scoring review?
No. Predictive models are useful for surfacing correlations across historical data, but without regular human review they can drift or exaggerate patterns that no longer hold as the market or product changes. Treat any predictive layer as an input that RevOps reviews on a set cadence, not as a fully automated replacement for governance.
Leave a Reply