Lead scoring is the practice of ranking prospective customers by how likely they are to buy, using a number or letter grade attached to each record in a database. Points are added for attributes that resemble past customers, such as job title or company size, and for actions that signal interest, such as a visit to a pricing page. They can also be taken away. Once a record crosses an agreed threshold it is handed to a salesperson; below it, the prospect stays in automated email programmes. The method exists because lead generation produces far more names than sales teams can call, and most of those names will never buy.

How a scoring model is built

Most models measure two things separately. Fit, also called explicit or profile data, describes who the prospect is: role, seniority, industry, headcount, revenue, country. Engagement, or implicit data, records what the prospect has done: email clicks, page views, form submissions, event attendance. Webinar platforms log time in session, poll answers and questions per attendee, often rolled into one engagement score and pushed into a customer relationship management (CRM) system.

Oracle's Eloqua rates profile from A to D and engagement from 1 to 4, producing a 16-cell grid in which A1 is the most qualified lead and D4 the least. A B1 record, a good fit and very interested, goes to the sales queue, according to Oracle documentation, while a C1 stays in nurture. HubSpot's scoring tool, rebuilt in 2025, uses a three-by-three matrix from A1 to C3 for combined scores and High, Medium and Low labels for single-dimension ones. Administrators cap total points and points per criteria group. Score decay reduces points for actions as they age, over one, three, six or twelve months, so a white paper downloaded last year counts for less than a demo request this week.

A hypothetical model for business-to-business (B2B) software shows the arithmetic. A director-level title earns 20 points, a company of 200 to 1,000 employees 15, a target industry another 15. A pricing page visit adds 10, webinar attendance 15, an email click 2. A free webmail address subtracts 10; an unsubscribe subtracts 20. With the threshold at 65, a director at a 500-person company in a target sector who attended a webinar reaches 65 and becomes a marketing qualified lead(MQL). Without the webinar, the same person sits at 50 and waits. Forrester's Laura Ramos gave similar examples in 2008: "+5 for downloading a white paper and +15 for attending a webinar".

Once flagged, the record moves along a lifecycle that most CRM systems track as lead, MQL, sales qualified lead (SQL), opportunity and customer. Marketing operations teams write the rules in marketing automation software such as Eloqua, Marketo, HubSpot or Salesforce's Account Engagement; sales teams act on the result.

Predictive lead scoring swaps hand-set points for a machine learning model trained on which past leads converted. Salesforce's Einstein Lead Scoring needs at least 1,000 leads created in the previous 200 days and at least 120 conversions, according to Salesforce documentation. Below that, it falls back on a global model built from anonymised data across many Salesforce customers, and scores are refreshed every 10 days or so. HubSpot's AI-generated scores require a sample of at least 50 contacts, 25 converted and 25 not.

Where scores meet the ad platforms

For paid media, the question is whether a lead turned into anything. Scores and lifecycle stages increasingly travel back to ad platforms as offline conversions, so that bidding systems chase qualified leads rather than form fills. Without that feedback, Meta's lead delivery is indifferent to the value of a result and optimises for volume.

Google introduced enhanced conversions for leads in 2022, matching hashed email addresses or phone numbers from the CRM to earlier ad interactions. Google Ads API v24, released on April 22, 2026, added nine conversion types splitting lead events into generate, qualify and close stages. Attaching a monetary value to each stage lets value-based bidding favour valuable leads over raw volume.

Meta's Conversion Leads performance goal sets entry conditions, according to its developer documentation: at least 200 leads a month, CRM uploads at least daily, a target stage reached within 28 days of the lead being created and a conversion rate for that stage of between 1% and 40%. It works only with Instant Forms. LinkedIn announced Qualified Leads Optimization in December 2024, fed through its Conversions API with at least five qualified leads every two weeks, according to Social Media Today. LinkedIn claimed early reductions in cost per qualified lead of upwards of 39%.

Third parties now sell the score itself as a bidding input. Scowtt, which partnered with LiveRamp in February 2026, trains models on more than 50 CRM variables and sends conversion probability and customer value scores to Google, Meta and TikTok roughly 15 minutes after a CRM event.

Origin and evolution

The threshold logic descends from the Demand Waterfall, a funnel model from the research firm SiriusDecisions that popularised the terms MQL and SQL. Forrester, which agreed to acquire SiriusDecisions in November 2018, dates the original waterfall to 2006; Wikipedia gives 2005. The model was rearchitected in 2012, and in 2017 Terry Flaherty and Kerry Cunningham unveiled the Demand Unit Waterfall, which counted buying groups rather than individuals.

Oracle agreed to buy Eloqua in December 2012 for about $871 million, and Adobe agreed to pay $4.75 billion for Marketo on September 20, 2018. Ramos captured the ambition of the rules-based era: "Using numerical, quantifiable scores to grade leads turns the art of marketing into a science."

Machine learning followed. HubSpot launched predictive lead scoring in 2015, its first application of the technique, and on August 23, 2018 announced a Likelihood to Close score from 0 to 100 that it said was more than four times as accurate. Salesforce unveiled Einstein at its Dreamforce conference in 2016. HubSpot later retired its original score property altogether: creating new custom score properties was disabled on May 1, 2025, and existing scores stopped updating on August 31, 2025, with no one-to-one migration.

Why it matters for marketers

Lead generation budgets are judged on cost per lead, a number that says nothing about revenue. Scoring is the bridge. Google framed its 42 lead generation launches at Google Marketing Live 2026 around the argument that lead quality problems are measurement, quality and design problems, not sales problems.

Consistency is the difficulty. In a LinkedIn measurement guide, Alex Venus of Personio said MQL conversion rates vary from 4% to 20% or more within the same organisation. Automation is far from universal. A Firmable survey of 222 B2B sellers, reported today, found 23% using AI to score or prioritise leads, while respondents estimated that 32% of their contact and account data was inaccurate, incomplete or outdated.

Limitations and disputes

The weights are frequently arbitrary. Forrester, in a blog series on abandoning MQLs, argued that profile and engagement points are typically based on guesses and that the process converting early interest into revenue fails more than 99% of the time. Its B2B Revenue Waterfall, launched in 2021, moved measurement to the opportunity.

The unit of analysis is contested too. A score attaches to one person, yet purchases are made by committees. Intentsify, launching QuantumDemand on August 11, 2026, proposed marketing qualified accounts in place of MQLs, citing buying committees that grew from 6.8 to 10 stakeholders and journeys lengthening from 211 to 272 days. Chief executive Marc Laplante said B2B marketing had "optimized for the metrics technology made easy to measure: clicks, leads, impressions, and MQLs." Much research also happens out of sight: LinkedIn research in June 2026 found 94% of B2B buying groups use large language models before contacting sales.

Engagement data is polluted as well. Bots and fabricated submissions inflate behavioural scores, and Google has acknowledged that its invalid traffic filters do not prevent invalid leads on their own. Predictive models inherit such errors, and a global model trained on other companies' data may not reflect one firm's buyers.

Regulation draws a further boundary. The Court of Justice of the European Union (CJEU) ruled on December 7, 2023, in case C-634/21, that a credit score can amount to an automated decision under Article 22 of the General Data Protection Regulation (GDPR) when it plays a determining role. On May 28, 2026, Berlin's administrative court upheld a warning against a solar leasing firm that ran Schufa credit checks on prospects before any site visit, finding neither pre-contractual necessity nor legitimate interest provided a lawful basis.

Not the same as

Lead grading - a fit-only rating. Salesforce's Account Engagement grades prospects against profiles covering location, company size and job title, separately from its behavioural score.

Lead validation - checks that a submission is genuine, through measures such as reCAPTCHA or double opt-in, before any score is worth calculating.

Account scoring - ranks companies rather than people, usually with intent signals aggregated across a buying group.

Lookalike modelling - scores people not yet in the database. LinkedIn Predictive Audiences uses lead form and conversion data as a seed to find similar members.

Recent developments

Ad platforms have begun scoring leads themselves. Google announced Lead Intent Scores in May 2026, labelling submissions as high, medium or low intent from its own signals, alongside a native leads screen sorting records into raw, qualified, converted and lost. Since April 21, 2026, Google Ads has also used AI to qualify phone call leads from recordings in the United States and Canada, replacing call duration as the main test.

The plumbing is shifting at the same time. Google stopped accepting new adopters of offline conversion imports through the Google Ads API on June 15, 2026, directing them to the Data Manager API.

Timeline

  • 2005 or 2006: SiriusDecisions introduces the Demand Waterfall, source dates differ
  • 2008: Forrester's Laura Ramos describes numeric lead scoring as key to lead management
  • 2012: SiriusDecisions releases the Rearchitected Demand Waterfall
  • December 20, 2012: Oracle agrees to acquire Eloqua for about $871 million
  • 2015: HubSpot launches predictive lead scoring, its first machine learning application
  • 2016: Salesforce unveils Einstein at Dreamforce
  • 2017: SiriusDecisions unveils the Demand Unit Waterfall built around buying groups
  • August 23, 2018: HubSpot announces the Likelihood to Close predictive score
  • September 20, 2018: Adobe agrees to acquire Marketo for $4.75 billion
  • November 2018: Forrester agrees to acquire SiriusDecisions
  • 2021: Forrester launches the B2B Revenue Waterfall, centred on opportunities
  • March 2022: Google launches enhanced conversions for leads
  • December 7, 2023: CJEU rules in case C-634/21 that credit scoring can be automated decision-making under GDPR Article 22
  • December 17, 2024: LinkedIn announces Qualified Leads Optimization
  • May 1, 2025: HubSpot disables creation of new legacy score properties
  • August 31, 2025: HubSpot legacy scores stop updating
  • February 2026: LiveRamp and Scowtt partner on predictive CRM scores for ad platforms
  • April 21, 2026: Google Ads introduces AI-qualified call conversions in the United States and Canada
  • April 22, 2026: Google Ads API v24 adds nine lead stage conversion types
  • May 2026: Google announces Lead Intent Scores
  • May 28, 2026: Berlin administrative court rules on pre-visit Schufa checks of solar prospects
  • June 2026: Google Ads introduces a native leads screen with four lead stages
  • June 15, 2026: Google closes offline conversion imports via the Google Ads API to new adopters
  • August 11, 2026: Intentsify launches QuantumDemand and marketing qualified accounts
  • September 26, 2026: Firmable survey finds 23% of B2B sellers use AI to score or prioritise leads

Summary

Who: Marketing operations teams configure scoring models and sales teams act on them, using marketing automation and CRM software from vendors including Oracle Eloqua, Adobe Marketo, HubSpot and Salesforce. Ad platforms such as Google, Meta and LinkedIn consume scores as conversion signals, and Google now produces its own.

What: A method of ranking prospects by likelihood to buy, combining fit data about who a prospect is with engagement data about what the prospect has done, either through hand-set points or through machine learning models trained on past conversions. Records above a threshold become marketing qualified leads and pass to sales.

When: The underlying funnel dates to SiriusDecisions' Demand Waterfall of 2005 or 2006. Rules-based scoring spread with marketing automation in the late 2000s, predictive scoring arrived in the mid-2010s, and ad platform scoring and CRM feedback loops expanded through 2026.

Where: Inside marketing automation platforms and CRM systems, with scores and lifecycle stages sent to Google Ads, Meta, LinkedIn and other ad platforms through offline conversion imports and conversions APIs.

Why: Lead generation produces more names than sales teams can contact, and cost per lead says nothing about revenue. Scoring directs sales effort and gives bidding systems a value signal, though arbitrary weights, individual rather than buying group focus, polluted engagement data and GDPR limits on automated decisions constrain what a score can prove.