What Is an MQL (Marketing Qualified Lead)? Scoring, Threshold and SLA

An MQL (marketing qualified lead) is a lead that the marketing team has qualified as ready to hand over to the sales team 1 . Salesforce draws the same line from another angle: an MQL is a lead that fits your product or service but is not yet ready to buy 9 . This status is assigned by criteria written in advance, not by intuition. This page doesn't stop at the definition; it builds the mechanism that produces MQLs: scoring, threshold, handoff, measurement.

In Turkish search, the abbreviation MQL often doesn't point to marketing at all. We classified the top 20 results from Semrush's Turkish organic data, pulled on August 2, 2026, ourselves: only 1 covers Marketing Qualified Lead; the other 19 are about MetaQuotes Language, a cryptocurrency called Miraqle or industrial Minimum Quantity Lubrication 11 . This page covers only the marketing meaning.

What do the top 20 results for "mql" in Turkish search cover?

An MQL is not just a label but an internal contract: when marketing applies the label, sales is expected to call that record. If what the label means isn't written down, the two teams mean different things by the same word and the reports clash. The rest of this page explains how to build that written definition.

The one question that separates an MQL from an SQL: who did the qualifying?

The difference lies not in the lead itself but in who applied the label. In HubSpot's product documentation, an MQL is a record marketing considers ready for sales, while an SQL is a record the sales team has qualified as a potential customer 1 . In Salesforce's framework, an SQL both fits the product and is ready to buy 9 . That's the essence of the distinction; for the concept of a lead itself and its subtypes, see our definition of a lead.

Three things the abbreviation MQL gets confused with in Türkiye

• MetaQuotes Language (MQL4/MQL5): the programming language of the MetaTrader forex platform.
• Miraqle (MQL-USD): an asset traded on cryptocurrency exchanges.
• Minimum Quantity Lubrication: a minimal-quantity lubrication method used in machining.

This confusion has a practical consequence. Teams that use "MQL" as a filter in search and ad data pull traffic with nothing to do with marketing into their reports. Two simple safeguards are enough: spell out the abbreviation the first time it appears in internal documents, and use a fully written label such as "marketing qualified lead" in CRM field names. The same discipline also ends definition debates between teams.

An MQL criterion is the intersection of two questions: "could this person be our customer?" and "is this person interested?" The first is fit, the second engagement. HubSpot's scoring tool keeps the two as separate scores and combines them into a single score 2 . Three different product documentations build the same two-axis model in different words, so the concept is independent of the tool.

Writing criteria starts with writing your ideal customer profile (ICP). Open the deals you've won recently and find the common denominator across title, industry, company size and geography. That list is the raw material of the fit axis. Traits that recur in lost deals become the raw material for negative scores. Write both lists together with the sales team; criteria written by one side collapse at the first objection.

Fit signals: title, industry, company revenue (explicit data)

Fit is data that describes who the lead is. Oracle Eloqua calls it explicit data and gives title, industry and company revenue as examples 5 . HubSpot ties the same axis to demographic attributes such as age, title, company size and annual revenue 2 . Adobe Marketo Engage does not treat the demographic score as optional; it states that it must be added alongside the behavioral score 4 .

• Is the title at decision-maker or budget-holder level?
• Are company size and revenue within the band you serve?
• Is the industry one where you can produce references?
• Is the geography within the market you can sell to and deliver in?

Engagement signals: site visits, content downloads, email opens (implicit data)

Engagement is data that describes what the lead does. Eloqua defines it as implicit data: site visits and responses to campaigns 5 . HubSpot builds its engagement score from actions such as visiting the site, subscribing to the newsletter, clicking a call-to-action button and opening a marketing email 2 . Adobe defines behavioral scoring as assigning numerical values to trackable actions that show purchase intent 4 .

A model that leans on one axis breaks down quickly. If you only look at engagement, someone who doesn't fit your product at all but downloads every guide becomes an MQL and eats up sales time. If you only look at fit, a decision maker at a company that has never heard of you also counts as an MQL. Unless the two axes are read together, scoring doesn't produce a priority order; it only produces a ranking, and that ranking doesn't tell sales whom to call first.

Writing criteria is a separate job from deciding which channels bring leads into this pool. We cover the channel side and the steps of creating demand on our lead generation process page; here our focus is how to qualify an incoming lead. The only link between channel and criteria is this: if a channel keeps bringing leads from off-target industries, you fix it with targeting, not with a scoring rule.

The cleanest way to make scoring concrete is to build a two-axis grid. Oracle Eloqua gives a lead the letter A, B, C or D based on profile criteria 5 and the number 1, 2, 3 or 4 based on engagement criteria 5 . The two scores are combined into a single lead score: A1 is the most qualified lead, D4 the least 5 .

How to read the grid: A1 is the most qualified, D4 the least

The letter answers "is this person our customer?", the number "is this person interested?" The right person at the right company who is quiet stays in nurturing; someone very engaged but outside your audience is not sent to sales. HubSpot builds the same logic: its combined score labels run from A1 to C3, with the letter representing the fit score and the number the engagement score 2 .

The scales are not the same. Eloqua splits the engagement axis into four 5 , HubSpot's combined score into three 2 . In Eloqua a lower number means higher engagement 5 ; when you build your own scale, write down the direction from the start. When building your own model, carry over the logic, not the scale; how many tiers you use depends on how many distinct actions your sales team can tell apart. Building more tiers than you can distinguish makes the model unreadable.

The grid is not a report line but an action table. Oracle Eloqua also recommends mapping scores to the appropriate follow-up action once the model is defined 5 . In practice this comes down to three buckets: hand over to sales, keep nurturing, archive. Decide together with the sales team which cell falls into which bucket; otherwise the grid never becomes more than a nice graphic on the wall.

Fit (A-D) × engagement (1-4) grid: lead scoring

Example scoring table: how many points for which signal

The grid stays abstract until you write down how many points each signal gets. The points below are not universal values taken from primary sources but roicool examples showing how the model is built; you are expected to replace them with your own closed-deal data. The only source-backed rule is this: Adobe Marketo Engage recommends giving actions that lead directly to a sales conversation, such as a demo request, a score equal to the handoff threshold 4 .

Where do you get the point values? The soundest method is to work backward. Look at what actions your won deals took in the last weeks before handoff; actions that recur among winning leads deserve big scores. Actions common among lost or never-responsive records get small scores. Do this exercise once, then update it on a quarterly rhythm. Keep the table to one page; a scoring table nobody maintains is more misleading than having no scoring table at all.

Example MQL scoring table: how many points for which signal

Negative scoring: the signals that break the model

Points don't only get added. An off-target industry, an email from a competitor's domain, a careers page visit or a long stretch with no engagement go on the minus side. A model with no negative scores bloats over time: everyone crosses the threshold, the MQL count rises and sales stops trusting the list.

The second job of negative scoring is decay. If a lead crossed the threshold in the past but hasn't produced any signal for months, that score reflects its past, not today's reality. When you add a rule that erodes the engagement score over time, the list cleans itself. These examples are not a standard taken from sources but the framework we apply.

Adobe Marketo Engage recommends that beginners start with a threshold of 100 points and develop the model from there 4 . That number is not a universal standard but a starting point for calibration. The example table above is built on a smaller scale; if you work with those values, the threshold falls in the 50-60 range rather than 100. What matters is not the number itself but that the threshold requires more than one meaningful touchpoint. The handoff rule is clear: the lead is handed to the sales team when it has shown enough interest by taking the actions you defined and its total score crosses the threshold 4 .

The size of the threshold translates directly into workload. A threshold that is too low floods the sales team with leads not yet ready to talk, and the team stops taking the list seriously. A threshold that is too high keeps genuinely interested leads waiting. The threshold should be high enough that it can't be crossed by a single email open; the lead should have more than one meaningful touchpoint with your brand before handoff 4 .

Choose the threshold together with your scoring table. If the highest value of any single action doesn't exceed half the threshold, no lead can become an MQL in one move; at most companies, that is the desired behavior. The one exception is direct intent signals: Adobe recommends giving actions such as a demo request a score equal to the handoff threshold 4 . Someone who requests a demo crosses the threshold in one move and goes straight to sales. If you don't build in this exception, your hottest lead sits waiting just because they haven't downloaded enough content.

Calibrating the threshold: an update rhythm driven by data

Adobe writes that scoring is an iterative process and that both scores and thresholds should be reviewed continually as data accumulates 4 . In practice this is a simple quarterly check: look at the scores of closed deals at the moment of handoff.

If most won deals cluster just above the threshold, the threshold is in the right place. If nearly all winning leads are far above the threshold, it is lower than it should be and the gap is eating into sales time. Conversely, if sales sees only a handful of MQLs a month and closes all of them, the threshold is higher than it should be. You don't need an external benchmark for this check; your own closed-deal data is enough.

A word of caution: the handoff threshold and the score produced by predictive scoring models are not the same thing. The handoff threshold is a rule you write; a predictive score is a probability estimate derived from historical data. The two are produced on different scales and are not interchangeable; we cover predictive scoring separately in the next section.

A threshold is not a rule on its own; it needs to be tied to a document both sides accept. HubSpot defines an SLA (service level agreement) as a contractual document that sets specific, measurable standards between parties 10 . Interdepartmental SLAs, in turn, create alignment and accountability, especially between sales and marketing teams 10 .

The four items the SLA must cover

HubSpot's definition also provides the skeleton of the document: clear handoff procedures, lead quality standards and response time expectations 10 . Add a feedback obligation and the agreement becomes two-way.

• Handoff procedure: at what score threshold, with which fields filled in and through which system the lead is handed over 10 .
• Quality standard: the mandatory fit criteria for a record to count as an MQL, that is title, industry and geography 10 .
• Response time: how quickly sales will contact a handed-over lead. Derive this from your own team's capacity, not from benchmark data 10 .
• Feedback: sales flags the reason for every rejected lead in the same system.

Both sides of the agreement make commitments. Marketing commits to volume and quality; sales commits to contact and feedback. A one-way SLA, meaning just the sentence "marketing will produce this many MQLs", produces a list that hits its target in the first quarter but that sales never calls.

Keep the SLA somewhere both teams can see it and review it quarterly; the document should carry a date, a version number and the approval of both sides. The system in which the handoff happens should also be written down: is it a stage change in the CRM, a task assignment or a notification? When this detail isn't written down, the handoff gets lost in email and no one takes responsibility. Write in the same document who can revoke the MQL label and in which situations; a model without a revocation rule can't clean up its mistakes, and the numbers lose their meaning over time.

What happens to a rejected MQL: the feedback loop

A rejected lead doesn't go in the bin; it goes back into the model. Sales' rejection reasons, such as wrong industry, no budget or not a decision maker, are the cheapest source for updating the scoring rules. Adobe's principle of an iterative process applies right here: scores and thresholds are reviewed as data accumulates 4 .

Sales validates the lead with its own framework. In Salesforce's training content, one of these frameworks is BANT: budget, authority, need and timing 9 . When you record which item a rejected lead failed on, that item can be turned directly into a fit scoring rule. If leads without budget keep coming from the same industry, the problem lies not in scoring but in targeting. In this way the SLA stops being a document and becomes a loop that feeds the model.

Building an MQL system in five steps

As hand-written scoring rules grow, they get harder to maintain. A 2025 peer-reviewed study published in Frontiers in Artificial Intelligence writes that with traditional scoring models, sales reps spend a lot of time on a pile of unqualified leads that won't convert, and that these models can turn into inaccurate, arbitrary and biased methods 8 . The same study defines lead scoring as using data analysis to estimate each lead's weight according to its likelihood of converting 8 .

Predictive scoring takes over this maintenance. In Microsoft Dynamics 365, predictive lead scoring calculates the score of open leads with a machine learning model based on historical data 6 7 . A terminology note: Microsoft's official Turkish documentation uses "müşteri adayı" (literally "customer candidate") as the Turkish term for "lead" 7 ; the Turkish version of this page uses the same term.

A rule-based model is not a bad option at small volumes; its biggest advantage is that it is readable. Sales can see at a glance why a lead counts as an MQL, and there is a rule they can challenge. A predictive model loses some of this transparency, and you need to explain the reasoning behind the score separately. The choice depends not only on technology but also on how much the team trusts the model.

The minimum data for predictive scoring: 40 + 40

The model has a concrete data threshold. According to Microsoft's documentation, the organization must have created and closed at least 40 qualified and 40 disqualified leads during the time period selected for training the model 7 . For a team that hasn't reached this number, predictive scoring is not the next step; first you need to run the rule-based model and accumulate closed-deal data.

The output is a probability ranking. In Microsoft's example, the model calculates a score of 80 for Lead A and 50 for Lead B; Lead A is more likely to convert into an opportunity 7 . Don't confuse this score with the handoff threshold described above: it is produced on a different scale with a different logic, and it does not replace the threshold.

On the academic side the results are promising but narrow in scope. The same 2025 study compared fifteen classification algorithms using the CRM data of a single B2B software company 8 ; the best performer, the Gradient Boosting Classifier, achieved an average accuracy of 98.39% 8 . This is not an industry benchmark but the result of a single-company case study, and there is no guarantee it will give the same result on your data.

According to Forrester's 2022 waterfall benchmark data, the conversion rate from inquiry to closed-won in an MQL-centric process is below 1% 3 . It's important to state what this figure is not: it is an end-to-end conversion, covering the whole path from first interest to closed deal. It is not the MQL-to-SQL conversion rate and cannot be used in place of that rate.

The same analysis also points to the model's weak spot. According to Forrester, the points assigned to profile and engagement factors and the threshold that determines when an MQL is passed on are based on guesses and arbitrary estimates rather than on analysis of actual buying propensity 3 . This critique does not refute the scoring model we built above; it says that unless the model is calibrated with closed-deal data, it remains a list of assumptions.

A buying group, not one person: where to read the signal

According to Forrester, more than 80% of B2B purchase decisions are made by a buying group of more than three people 3 . The same analysis says that one person downloading four white papers is at best a sign of mild interest for sales, and that the truly strong signal is engagement from several members of the buying group 3 .

In practice this means: aggregate scores not only on the person record but also on the company record. Several people from the same domain visiting the pricing page in the same week is a stronger signal than one person downloading documents back to back. Most scoring setups do the opposite and reward the single most active person.

The way to turn this critique into the model is this: set point weights with closed-deal data rather than guesswork, and add a second, account-level threshold alongside the person score. The person score tells the sales rep whom to call; the account score tells you which company to open a campaign for. When both are kept in the same table, the MQL report becomes a document you can make decisions from again. In practice, this starts by merging records from the same company under a single account; without that merge, the buying group signal is invisible anyway.

The conclusion is balanced: the MQL isn't dead, but it isn't enough on its own. Its role of defining the handoff point between marketing and sales still stands. But when it isn't used together with signals aggregated at account level and a reading of the buying group, the reporting you build on it measures not the buying side but the person who downloads the most content.

A scoring model only works if conversion data is collected without loss. If CRM, ad platform and site data don't come together in one place, the MQL definition stays on paper: you are missing the signals to assign points to, and you have no closed-deal data to calibrate the threshold. The order is clear: measurement infrastructure first, then the scoring model.

In the accounts we manage, we build the measurement infrastructure in this order: we set up Google Ads accounts from scratch and optimize them, and we record conversions with server-side API integration (CAPI) 11 . The aim is to tie to data not just the moment a form is submitted but also what happens after that form. An ad platform only starts bringing similar people once it sees the data of quality leads.

On the measurement side, don't start scoring before you set up three connections. Match the conversion in the ad platform to the record ID in the CRM, keep the MQL label in a single field in the CRM, and write the closed deal back with the same ID. When these three work, you see which campaign brings not forms but won deals; the data to calibrate the scoring table also comes from here. Every scoring debate held before these connections are in place is a debate of opinions.

Cost per MQL, not CPL: changing the optimization goal

On the ad side, the fastest gain comes from changing the optimization goal. Our cost-per-MQL case study is a concrete example: ad optimization was run not on cost per form (CPL) but on cost per qualified lead 11 . The goal was to make cheaper not the number of forms but the number of leads sales found worth talking to.

Even with the same service set, the cost per lead is not fixed; in the health tourism accounts we manage, it ranged between 11 and 16 USD 11 . Cost per click falling while cost per lead rises is a common picture. That's why an improvement in cost per click alone does not count as a sign of success.

Building the MQL definition is not a reporting job but a decision mechanism: it determines which lead gets called, which budget goes where and which channel gets shut down. When the definition isn't written down, all these decisions are made on gut feeling and debated again every month. If you want to build the model and tie it to your ad goals, our lead generation service covers exactly this scope.