Every B2B CRM carries a quiet accounting error: the same word — lead — is doing three unrelated jobs, and the resulting number in the pipeline report is inflated by roughly 18% on average (HubSpot State of Marketing 2026). A business lead is not a marketing lead is not a sales lead; they differ in what they assert, in who owns them, in how fast they rot, and in what converting one is even worth. This piece sets the strict three-layer taxonomy, attaches 2026 conversion rates to each boundary, shows where the layers silently die of data decay, and makes the case that the account — not the lead — is the only durable denominator. It deliberately does not re-explain lead scoring or definitions from first principles; the value here is the layer boundaries and the math across them.
Layer one: the business lead. A business lead is a company-level record — a firm that plausibly belongs in the addressable universe, with attributes (industry, size, region, tech stack) but no verified individual intent. It answers the question: who might buy, ever? Business leads come from directories, lists, intent-feed account matches, and account-based research. They are the cheapest layer to acquire and the most commonly abused — sold by volume, counted as pipeline, and converted into nothing when nobody inside the company has expressed a thought. A business lead is an address, not a person; treating it as demand is the original sin of the 18% inflation problem.
Layer two: the marketing lead, or MQL. A marketing lead is an individual captured behaving — a content download, a webinar registration, a pricing-page visit — with enough signal to clear the marketing-qualification bar. It answers: who inside a plausible company is showing interest? The MQL is a behavioral observation about one person, not a statement about a company's intent to buy; the same person might be a student, a competitor, or a procurement researcher doing diligence on you for a vendor they already chose. MQLs are the layer most polluted by 2026's AI-generated form noise, which is why verification premiums exist upstream of them.
Layer three: the sales lead, or SQL. A sales lead is a named contact with verified fit and articulated intent — someone who has said, in words or unambiguous actions, what problem they are solving and on what horizon. It answers: who is worth a seller's time this quarter? SQLs are created by qualification conversations, by hand-raising actions, or by SDR research against a triggered account. This layer is expensive by construction, and it is the only layer that forecasts: pipeline math downstream of the SQL layer is arithmetic; pipeline math upstream of it is astrology.
The 2026 rates across the boundaries are stable enough to plan against. MQL to SQL runs at 31% on average with a healthy band of 24-39%; SQL to opportunity at 52%; opportunity to closed-won at 21% (Salesforce State of Sales 2026). Two properties of those numbers matter more than their levels. First, time-to-treatment: the median MQL converts to SQL in 11 days, and MQLs that sit untreated past 30 days lose 62% of their eventual conversion — the layers decay while they wait, which makes routing speed a revenue variable, not a service metric. Second, the compounding: at the average rates, 1,000 MQLs become 310 SQLs, 161 opportunities, and 34 wins — so a 5-point improvement at the MQL-to-SQL boundary produces 17% more wins, while the same effort spent counting leads harder produces nothing.
The layers die at different speeds, and the dying is silent. Contact data decays at 22.5-28% annually — people change roles, mailboxes go dark, phone numbers recycle — so an untreated SQL pool rots by roughly a fifth per year even with zero new inflow. The rot is worst in exactly the records dashboards love: old MQLs accumulate in nurture purgatory forever "active," and among unlabeled CRM records with no layer tag, 34% of so-called active leads are in fact dead — departed contacts, orphaned accounts, or duplicates (Forrester B2B Lead Management 2026). A pipeline report that mixes layers inherits all three error modes at once: inflation from counting business leads as demand, staleness from undated marketing leads, and phantom SQLs from unverified intent. Layer labels are not bureaucracy; they are the audit trail that makes the report arithmetic instead of narrative.
The account-based reframe is how the disputes between definitions actually end, and 2026 made it the majority position: 63% of B2B organizations now run at least partially account-based measurement, and those that do report 41% fewer sales-marketing definition conflicts (Gartner Demand Generation 2026). The mechanism is simple: when the denominator is accounts progressing through stages, the three lead layers become instrumentation rather than identity. A business lead is an account added to coverage; a marketing lead is engagement signal on a covered account; a sales lead is a conversation opening on an engaged account. The layers stop competing for the word lead and start describing observable states of one thing — a company moving toward or away from a decision.
The practical data model is four fields and a cadence, not a platform migration. Mandatory fields: layer label (business / marketing / sales), source, verification status (verified, unverified, failed), and account link — every lead record attached to one company record, with the account as the reporting rollup. The cadence: quarterly decay pass — unverified records older than two quarters archived, failed verifications retired, account coverage recomputed. Teams that run exactly this — no new tooling — typically find the 18% phantom pipeline in week one, watch true MQL-to-SQL rise (the denominator finally excludes ghosts), and stop having the definitional argument because the account rollup no longer rewards either side for inflating its layer.
A worked example fixes the frame. Take a 5,000-record CRM from a mid-market industrial-software firm: label everything and the truth arrives — 2,900 business leads (account research, no individuals), 1,800 marketing leads (behavioral captures, median age 14 months, so decay-adjusted to roughly 1,350 live), and 300 sales leads, of which 102 are verified with current contacts. The honest funnel from the 1,350 live MQLs: 419 SQLs at the 31% rate, 218 opportunities, 46 wins — versus the dashboard's original claim of "5,000 leads, 600 SQLs, 312 opportunities." Same data, same rates; the difference is entirely layering and decay. That difference — roughly a third of the reported pipeline — is what the board was being told against what was real.
Where the layers meet money: each layer has a different economically rational owner and budget. Business leads belong to the account universe file — cheap, bulk, refreshed annually; spend here is coverage spend. Marketing leads belong to demand capture — spend here converts attention into identified behavior. Sales leads belong to qualification — spend here is human time, the most expensive per unit and the only layer where speed of treatment has a measurable revenue half-life. Budgets that blur the three buy volume where they need verification and polish where they need coverage; the layer taxonomy is ultimately a budget taxonomy.
The weekly report changes shape once the layers are honest, and the change is worth anticipating. The old report led with lead counts by source; the layered report leads with four numbers: net-new covered accounts (business layer), live engagement rate on covered accounts (marketing layer, decay-adjusted), open SQL count with age distribution (sales layer), and account progression — how many companies moved a stage this week regardless of which layer produced the motion. Marketing owns the second number, sales owns the third, and both own the fourth jointly, which is the entire alignment conversation compressed into a scoreboard. Teams that switch to this format report the definitional arguments fading within a month or two — not because anyone won, but because the report no longer rewards the behavior that started the arguments.
And this is where the account spine earns its keep across all three layers: every lead type attached to one verified company record, so engagement signal, qualification state, and coverage report against the same denominator. That is the account layer Salebrate maintains — business leads become covered accounts, marketing leads become engagement on those accounts, and sales leads become open conversations, one map underneath all three so no layer can quietly become a different layer's inflation.
For teams that want a target to manage toward: layer-integrity as a single KPI — the percentage of pipeline-attached records with a current layer label, a verified contact, and a live account link. Firms that track it weekly typically hold it above 90% within a quarter, and every point of layer-integrity recovered shows up downstream as cleaner forecasting within two cycles. It is the rare metric that sales, marketing, and finance all read the same way, which is the point.
The three layers are not going to collapse into one vocabulary — the words are too entrenched for that. What a disciplined team does instead is make the layers explicit, rate them honestly, decay them quarterly, and roll them up to accounts. The 18% error is not a mystery; it is the predictable cost of letting one word do three jobs. Fix the labels, fix the math, and the pipeline report finally describes something that can be sold.