Every pipeline review eventually reaches the moment when someone asks the arithmetic question: how many leads do we actually need? The answer is usually a guess dressed as a number — a benchmark quoted from someone else's funnel, applied to a quota it was never derived from. The honest version of the answer is qualification math: a chain of stage conversions, each with its own benchmark band, run backward from the revenue target to the weekly lead volume the engine must produce. The 2026 data makes the chain computable end to end, and this piece walks it stage by stage — where each conversion sits, where it leaks, and how to run the backward math that turns a quota into a Monday-morning number.
The chain and its bands
The qualification chain runs visitor to lead, lead to MQL, MQL to SQL, SQL to opportunity, opportunity to won, and the 2026 benchmark bands are stable enough to plan against. Visitor-to-lead conversion for B2B sites runs 1 to 5%; lead-to-MQL runs 20 to 40% depending on traffic quality; and high-value inbound demands response inside five minutes (ev-blq-001). Downstream, blended lead-to-customer conversion averages 2-5%, but the blend is the trap — inbound leads convert around 5-10% while outbound runs 1-3%, which means the same lead target implies wildly different volume requirements depending on source mix (ev-blq-002). The middle of the chain carries the 2026 story: median MQL-to-SQL has fallen to 9.8% from 13.1% two years ago, while programs that rebuilt definitions around intent signals run 16.4% (ev-blq-004). Every stage has a band, a leak, and an owner — and the math only works when all three are named.
Stage one: visitor to lead — the 1-5% gate
The first conversion is the traffic-quality gate, and its 1-5% band (ev-blq-001) is less a performance target than a truth serum: a site converting far above 5% usually has a lead-definition problem (counting newsletter signups as leads), and one far below 1% usually has a traffic-mix problem (buying visitors who were never buyers). The leak at this stage is definitional — the form's promise versus the sales team's expectation. The owner is demand generation, and the repair is the form-to-definition audit: every form field maps to a qualification question sales actually asks, and every form's submissions route into the lead-to-MQL band with a scoring model attached.
Stage two: lead to MQL — the 20-40% filter
The second conversion is the qualification filter, running 20-40% of leads reaching MQL status (ev-blq-001). This stage exists to protect downstream humans from noise, and in 2026 the noise is industrial: AI-generated form fills and content-gated downloads inflate lead counts while the buying population stays flat. The leak is score drift — a scoring model calibrated eighteen months ago, rewarding behaviors that no longer predict. The owner is marketing ops, and the repair is the quarterly recalibration: sample MQLs that sales rejected, find the scoring signals that correlate with rejection, and cut them. A filter that passes 40% of everything is not filtering; a filter that passes 20% of the right traffic is doing its job.
Stage three: MQL to SQL — the 9.8% crisis point
The third conversion is where the industry's definitional rot shows up in the data. The median fell from 13.1% in 2024 to 9.8% in 2026 — a 25% decline in two years — while intent-signal programs hold 16.4% (ev-blq-004). The number is a governance indicator disguised as a performance metric: when marketing's MQL definition and sales' SQL definition describe the same buyer, the conversion runs in the mid-teens; when they drift apart — form fills counted as interest on one side, BANT-plus-timing demanded on the other — the conversion falls toward single digits regardless of effort. The owner is both sides jointly, and the repair is the shared definition with intent thresholds, reviewed quarterly. Teams that ignore this stage's band entirely — building plans on a 25% MQL-to-SQL assumption inherited from a 2022 deck — discover the error as a pipeline miss six months later.
Stage four and five: SQL to opportunity to won — the source split
The bottom of the chain is where source mix dominates. Blended lead-to-customer runs 2-5%, but inbound converts at 5-10% and outbound at 1-3% (ev-blq-002) — a five-fold spread that makes any blended plan a fiction. The leak at these stages is follow-through: the five-minute response discipline that makes high-value inbound nine times more likely to convert is an operational choice, not a natural constant. The owner is sales, and the repair is per-source math: every source carries its own conversion assumptions into the plan, and the aggregate pipeline forecast is the weighted sum of sources rather than one optimistic average applied to a mixed bag.
The backward math: from quota to Monday morning
The backward calculation is the discipline that turns bands into numbers, and one worked example shows the mechanics. Set a quota of $2M in new business at a $50K average deal size — forty wins required. With pipeline coverage at the benchmark 3.4x quota (ev-blq-003), the engine needs $6.8M in qualified pipeline, or about 136 opportunities at $50K each. At a 25% opportunity-to-win rate — the honest 2026 planning number for a mid-market motion — that requires 544 SQLs. Applying the 9.8% median MQL-to-SQL (ev-blq-004) demands roughly 5,550 MQLs; at the intent-led 16.4%, about 3,300 — the definition repair just cut the required top-of-funnel volume by 40%. At a 30% lead-to-MQL rate (ev-blq-001), the lead requirement is 11,000 to 18,500 depending on which MQL-to-SQL world you live in; at a 3% visitor-to-lead rate, that is 370,000 to 615,000 visitors a year — or a smaller number, if the traffic is richer and the definitions are tighter. Every line of that arithmetic is negotiable, which is the point: the negotiation happens in the open, against benchmarks, instead of hiding inside a spreadsheet's default assumption.
Three weekly recalibration drills
The math stays honest with three drills, run weekly and rotated. Drill one, the stage-stop audit: pull fifty leads that stopped at each stage boundary last week and read why — the reasons cluster, and the top cluster is the week's repair. Drill two, the source-split check: recompute last week's conversions by source against the per-source assumptions in the plan, and flag any source drifting more than five points from its band (ev-blq-002). Drill three, the definition vote: one boundary question — "would sales still call this an SQL today?" — asked of twenty recent records, with the yes-rate tracked as the definition's approval rating. Fifteen minutes each, one owner each, and the chain's assumptions stay live data rather than settled fiction (ev-blq-003).
The per-source discipline
The meta-lesson of the whole chain: blended numbers are for reports, per-source numbers are for running the business. The 2-5% blended conversion (ev-blq-002) describes an industry; it plans nothing. A plan built from per-source bands — inbound at its rate, outbound at its, events at theirs — prices each channel's contribution honestly and makes the budget conversation arithmetic instead of theology. The teams that run the chain this way share a quiet advantage: when a stage leaks, they know within a week, and the repair costs a definition change instead of a quarter. Qualification math is not the sexy part of demand generation, but it is the part that decides whether the rest of the machine was worth building.
The outbound variant of the math
The worked example above leaned inbound; the outbound variant prices the same chain differently and is worth running because it exposes the volume asymmetry. Take the same forty-win target at $50K. Outbound opportunities convert to wins at roughly one-third the inbound rate on longer cycles, so assume 15% opportunity-to-win: 267 opportunities instead of 136. Outbound SQLs-to-opportunities run tighter — a booked meeting is already a qualified conversation — so assume 50%: about 534 SQLs. But the SQL supply line is where outbound pays its tax: at a 1-3% lead-to-customer blend and meeting acceptance in the single digits per hundred touches, the activity math at the top becomes the real constraint (ev-blq-002). Sequences, call attempts, and list volume — not visitor traffic — become the denominators, which is why outbound plans fail on activity arithmetic rather than conversion arithmetic, and why the five-minute response discipline that governs inbound speed matters less than the signal discipline that governs outbound targeting. Both motions produce pipeline; they simply break at different stages, and a plan that cannot name its break-point stage has not been planned.
Three errors that break the math
Three recurring errors corrupt otherwise careful chains. The first is the inherited assumption: a conversion rate copied from a prior plan or a vendor deck, never re-derived from the company's own last four quarters — the fix is sourcing every band in the plan to either the company's trailing data or a named benchmark, never both silently. The second is the mix-free average: applying one blended conversion to a mixed-source pipeline, which the inbound-outbound spread makes a five-fold error at the extremes (ev-blq-002). The third is the static chain: bands set in January and never revisited, while the market's middle stage collapsed 25% in two years — the 9.8% median is itself a moving number, and a plan that assumes constants in a drifting market is quietly wrong every week it survives (ev-blq-004). The weekly drills exist to catch these three errors while they are still cheap.