MQL-to-SQL Leak Patterns 2026

Direct answer

In 2026, the MQL-to-SQL leak in mid-market B2B is still mostly definitional. Six recurring patterns account for most of it: definitional drift at the boundary, response latency, bad-fit capture, routing and territory collisions, the incentive seam between marketing and sales, and lifecycle decay between the MAP and the CRM. Two surfaces didn’t exist at this scale before 2024: AI-agent scoring and phone-first cadence discipline. Both fix leaks; both create new ones if rolled out sloppily.

If your MQL-to-SQL conversion sits below 35%, one of these six is the cause. Usually more than one.

The 2026 baseline numbers

Four primary sources anchor the rest of this article. None of them are quoted at full length; the patterns below show how they apply.

• Salesforce State of Sales, 8th edition (Feb 2026): Nine in ten sales teams now use AI agents or expect to within two years. AI agents span planning through quoting, and most teams are still growing pipeline year-over-year despite capacity constraints.
• HubSpot State of Marketing 2026: Sixty-one percent of marketers call AI the biggest disruption to marketing in twenty years; 80% use AI for content; 75% for media production.
• Chili Piper on speed-to-lead (Aug 2025 update): The InsideSales / Oldroyd rule — contact within five minutes is roughly 21× more effective than thirty minutes, and businesses that respond in five minutes or less are 100× more likely to connect and convert — is still the headline number. The operational floor is unchanged: the B2B median response time is about 42 hours, and 38% of inbound leads never hear back. Industry splits: Healthcare ~2h 5m, Telecom ~16m, small companies (1–300 employees) ~48m, mid-market (301–2500) ~1h 38m, enterprise (2501+) ~1h 28m.
• Cognism State of Outbound 2026 (2025 internal data; n=451,895 calls, 39,679 meetings): Industry cold-call connect only edged from 2.3% in 2025 to 2.7% in 2026. Cognism SDRs running on verified contact data hit a 13.3% answered rate on cold calls — nearly matching the 14.4% AEs get on warm calls. Industry email reply average sits near 5%; Cognism AEs ran 28.38%, SDRs 8.98%. Top outbound cadence is 57% phone / 27% LinkedIn / 15% email; meetings held across the dataset was 85.94%; weighted talk-track usage for MEDDPICC discovery hit 77%.

Where these sources disagree — most notably on the exact speed-to-lead threshold and the cold-call connect rate — the disagreement is flagged below, not papered over.

The six leak patterns

1. Definitional drift at the boundary

Marketing calls it an MQL; sales calls it “not ready.” The handoff breaks because the criteria live in someone’s head. Cognism’s 2025 MQL vs SQL guide defines the boundary in behavioral terms: high-intent inbound actions (demo request, pricing page revisit, competitive comparison, sales-replied email) trigger SQL; gated content downloads and newsletter signups remain MQL. Most teams don’t codify this. The result is a contested boundary where marketing’s “MQL volume” gets silently rejected by sales, and both sides report a conversion number that doesn’t reconcile with the other.

The test: can an SDR look at a record, see a score, and predict the rejection reason in under sixty seconds? If not, the definition is contested.

2. Response latency, not speed-to-lead

The Oldroyd / InsideSales finding — contact within five minutes is roughly 21× more effective than within thirty minutes — is still the cited rule across 2026 vendor benchmarks and is the empirical basis for “respond in five minutes = 100× more likely to connect” claims. The operational reality hasn’t moved much: B2B teams still average ~42 hours to first response, and 38% of inbound leads never hear back at all.

Industry splits from the HubSpot-via-Chili-Piper dataset:

Mid-market is the worst of both worlds: not small enough to have a founder doing first touches, not large enough to staff a 24/7 inbound desk. The leak isn’t that nobody knows the rule — it’s that the team can’t operationalize it. Chili Piper reports a first-minute follow-up can lift conversion by ~391% versus waiting a few minutes longer. That number is a marketing claim, but the directional case is solid: high-intent inbound decays fast.

The test: what is your median T2T (time-to-touch) for demo requests? If it’s over thirty minutes, this is the leak.

3. ICP-fit at capture, not lead source

Marketing optimizes for form-fill volume; sales rejects most of it as wrong-ICP. Across published demand-gen benchmark syntheses for mid-market SaaS, healthy teams run MQL-to-SQL in the 30–45% band; dysfunctional teams sit at 13–25%. Variance tracks ICP-fit at the front of the funnel — does this lead match employee band, industry, geo, and tech stack — not lead source quality. A “lead source” report that doesn’t segment by ICP-fit is theatre.

The 2026 fix is gating on firmographic fit before the MQL flag fires. A demo request from a 12-person company in a non-target geography is not an MQL — it’s noise that costs an SDR an afternoon.

The test: what’s the rejection rate of MQLs by reason? If “wrong ICP / bad fit” is over 40%, fix the gate.

4. Routing and territory collisions

Right lead, wrong owner. Round-robin assigns an SDR who doesn’t own the segment; geo/IP mapping misses; named-account lists collide with product-led signups. Salesforce’s 2026 framing puts AI agents at the center of routing, quoting, and territory planning, but only teams that re-engineered routing logic since 2023 are seeing lift. If your routing hasn’t been rewritten since pre-2022, it’s leaking in two places:

• Named-account collisions (the same account going to two reps). Anything above ~5% is the threshold most operators treat as broken.
• SDRs working accounts outside their ICP-fit because round-robin doesn’t know what an SDR owns.

The test: what’s your named-account collision rate? What’s the share of MQLs that hit an SDR whose owned-segment list excludes the account’s industry or size band?

5. The incentive seam between teams

Marketing gets paid on MQL volume or MQL→SQL conversion; sales gets paid on closed-won. They optimize for different things at the exact boundary this article is about. The 2026 fix isn’t a metric change — it’s a co-owned metric: SAL→SQL conversion (sales-accepted to sales-qualified), reported weekly, with both functions signed onto the SLA. Cognism’s State of Outbound 2026 reports that best-in-class outbound teams run an SDR-AE buddy system with weekly insight loops; weak teams hand off a record with no context and hope. The leak at the boundary is mostly a context leak, not a definition leak.

The test: when an SDR rejects an MQL, is the rejection reason a free-text field, or a required picklist the marketing team reads weekly? If it’s free-text, the feedback loop is broken.

6. Lifecycle decay between MAP and CRM

A six-week-old MQL is increasingly the wrong artifact. Buyers research off-site, return directly, and re-enter via a different channel. HubSpot’s 2026 data shows AI-generated content flooding the market; trust signals shift away from gated PDFs toward communities, podcasts, and direct brand revisits. The leak: nurture tracks keep emailing a “lead” who is already mid-evaluation on a competitor. Scoring decay — reduce scores by X% every Y days without engagement — is a real fix. AI re-scoring in 2026 can do this automatically; the catch is that the model needs to be grounded in SQL acceptance, not just engagement signals, or it will keep awake leads that have already gone dark elsewhere.

The test: of the MQLs from 90 days ago that never converted, how many are still receiving active emails? If the answer is “most of them,” the scoring decay isn’t decaying.

The new surfaces in 2026

AI-agent scoring. Salesforce State of Sales 8th (2026): 9 in 10 sales teams now use or expect to use AI agents within two years. HubSpot 2026: 80% of marketers use AI in workflows. The interesting question isn’t whether to use AI for scoring — it’s whether to trust the score. The 2026 default rollout pattern that has held up is parallel-running: AI-scored and rules-scored side by side for one full quarter, with SDR accept-rate by score band as the tiebreaker. The model with the higher accept-rate at the SQL boundary wins the seat; the loser retires. Don’t trust either alone.

Phone-first cadence discipline. Cognism’s State of Outbound 2026 finds that top outbound teams run a 57 / 27 / 15 channel split (call / LinkedIn / email), with 3–5 touches per prospect, and book meetings no more than four days out. Meeting show rate on bookings >4 days out averaged ~60% across the dataset; a structured pre-meeting cadence (case study two days before, day-before confirmation) lifted show rates to 80–82%. For inbound, this maps to: phone-first within minutes for high-intent; SDR triage within 24 hours for mid-intent; no human touch for low-intent until behavior changes.

Talk-track usage as a quality metric. Cognism’s 110-AE analysis tracks weighted talk-track usage rates — next-steps/process 91%, pain/value 82%, product/differentiation 79%, MEDDPICC 77%, objections 74%. This is the 2026 version of “are reps actually running the framework.” If you can’t measure it, you can’t enforce it, and unenforced frameworks are slogans.

SLAs that hold in 2026

Three numbers to put on the wall:

• T2T (time-to-touch) — median minutes from form fill to first human touch. ≤5 min for high-intent (demo request, pricing page revisit); ≤30 min for mid-intent (whitepaper, webinar, comparison guide); no SLA for low-intent until behavior changes.
• T2DQ (time-to-disqualify) — median hours from MQL to a clean Yes/No. If sales can’t disqualify inside 48 business hours, your scoring is lying, not your reps.
• T2M (time-to-meeting) — median days from MQL to first qualified meeting. Cognism’s outbound benchmark is 7–10 days at 3–5 touches; inbound high-intent should be sub-48 hours.

Two quality checks that catch what the SLAs don’t:

• Meeting-held rate. Cognism’s outbound number is 85.94% across 39,679 meetings — nearly 9 in 10 booked meetings actually happen. If yours is below 70%, your booking qualification is broken, not your cadence.
• SDR acceptance rate. Percentage of MQLs marked “working” versus “disqualified.” Below 50% means the MQL definition is contested, regardless of what the SLA doc says.

Qualification framework reality

BANT survives at the SMB / low-ACV end as a gut-check. MEDDIC / MEDDPICC is the 2026 default for mid-market and enterprise SaaS at $50K+ ACV — the “Paper” and “Process” components matter more in 2026 because procurement is in more deals than it was in 2019. GPCT (Goals / Plans / Challenges / Timeline) still appears in HubSpot-led GTM motions but has weak adoption outside that ecosystem. Custom / AI-scored qualification is growing fast as agents pre-qualify on firmographic fit plus real-time intent before routing.

The framework matters less than enforcement. Cognism’s internal data shows MEDDPICC discovery talk tracks running at 77% weighted usage when measured — not assumed. Pick one. Measure talk-track usage at the call level, not in the manager’s gut. The team that measures is the team that runs it.

30 / 60 / 90 plan to plug the leaks

• Days 1–30. Codify the MQL/SQL boundary in writing, in CRM fields, not a doc. Get marketing and sales to sign a one-page SLA covering T2T, T2DQ, T2M, acceptance threshold, and disqualification categories. Identify the named-account collision rate. Stand up the three SLA dashboards. Make the rejection reason a required picklist, not free text.
• Days 31–60. Deploy AI scoring as a shadow model alongside rules. Instrument accept-rate by score band. Add a phone-first step within five minutes to the top three inbound sources. Make disqualification reasons required CRM fields with a fixed taxonomy (bad fit / not ready / no budget / wrong owner / no response). Cut inbound cadences to 3–5 touches over 7–10 days.
• Days 61–90. Pick one qualification framework (MEDDPICC is the 2026 default for mid-market SaaS) and train the team. Stop booking meetings more than four days out without a confirmation cadence. Retire the nurture path for any lead silent 60+ days unless AI re-scoring keeps them live. Re-run the leak audit and compare to the Day 1 baseline.

The biggest unlock in 2026 isn’t a tool — it’s disqualification hygiene. Most mid-market teams over-rotate on lead-gen volume and under-rotate on fast, clean No. That is the actual leak, and the cheapest one to fix.

Leak audit checklist (60 minutes)

1. Pull MQL volume and MQL→SQL conversion for the last 90 days. Below 30% is a leak; below 20% is a fire.
2. Pull T2T by lead source for demo requests. Anything above 30 minutes is the leak.
3. Sample 20 rejected MQLs. Code each into: bad-fit / not-ready / no-budget / wrong-owner / no-comment. The dominant category names the leak.
4. Pull SDR acceptance rate (% of MQLs marked “working” vs “disqualified”). Below 50% means definitional.
5. If you run AI-scored and rules-scored, pull accept-rate by score band. Higher accept at SQL wins.
6. Pull named-account collision rate. Above 5% means routing logic is broken.
7. Pull 90-day-old MQLs that never converted. If they’re still being emailed, scoring decay is the leak.
8. Pull meeting-held rate. Below 70% means booking qualification is broken, not your cadence.