Lead Routing and Lifecycle: Definitions That Survive Contact With Reality

Ask three people in a B2B company how many MQLs there were last month and you get three numbers, different by a factor. Marketing counts stage entries, sales counts leads they were told to work, and the board deck has a third nobody can derive.

That is not a reporting failure. It is what a lifecycle model looks like after being written once in a workshop, then edited by eighteen months of undocumented automation.

Lifecycle stages become fiction within two quarters

The decay follows one path. Six stages are agreed. Six months later a rep needs somewhere to put accounts that went quiet, so someone adds Recycled. Then Product Qualified arrives with the trial. Nothing is removed, because removing a stage means fixing the reports that depend on it. Meanwhile four automations and a nightly sync write to the same field, alongside manual edits by reps.

Count how many records sit in your largest stage, usually 40% or more of the database, how many lack an entry timestamp for it, and how many moved backwards last quarter unintentionally. A stage half your database sits in permanently is a bucket labelled "everyone", and every metric from it is arithmetic on a word that stopped meaning anything.

Definitions that hold have four parts, not one

Most lifecycle documents contain only entry criteria, which is why they rot. Each stage needs four before it exists.

Entry criteria checkable against a field or event. Not "shows buying intent" but something a query can evaluate: submitted a demo request, or fit score above a threshold with a pricing view in the last fortnight. If you cannot write it as a filter, it is not a criterion, it is a mood.

Exit criteria, and where exit leads. Every exit goes to exactly one other stage, a record is in one stage at a time, and the set is ordered. Backwards movement is explicit: "recycled" is not a stage, it is a transition with a reason code returning the record to nurture.

One owner of the transition. A system or a named role, never "marketing and sales". Two owners means two write paths, and the field holds whatever was written last.

An expected time in stage, which turns staleness into a query rather than sediment.

Keep the total at six or seven, and write a timestamp and reason code on every move.

Round-robin, territory and capacity each break differently

Round-robin assumes every rep is interchangeable and available. It assigns to the person on leave, distributes junk in equal shares until nobody trusts the queue, and puts two reps on one account. Its real flaw is answering "who gets this" without asking whether it deserves anyone now, which is a scoring question, not an assignment one.

Territory matches language, timezone and account ownership, which matters in European selling. It breaks on skew, when one territory produces 60% of volume and has one rep, and on drift, when a company relocates or is acquired. It breaks hardest on enrichment failure: an empty country field sends records to a default owner nobody watches.

Capacity-aware routing assigns to whoever has room, measured by open tasks or time since last assignment. The best of the three, with two traps. The capacity number must be maintained automatically: a manual availability field is wrong within a fortnight. And it punishes the fastest rep, who clears the queue and is rewarded with more, so it needs a cap.

What survives is a stack, not a choice. Account ownership is absolute and checked first: splitting an account across two reps costs more than any delay. Territory constrains the eligible set, capacity chooses within it, and a named fallback owner takes the rest, reviewed weekly.

Deduplication, and the merge you cannot undo

Duplicates are structural, not careless. Email is not a stable identifier: people change employer, use aliases, and submit forms with different capitalisation or plus-addressing. Domain is not a company: subsidiaries, freemail and agencies break it. And most form handlers create a new record rather than matching when anything differs.

Match in layers. Normalise the email first, lowercasing and stripping plus tags, and match on that exactly. Then on company domain plus a fuzzy name comparison, treated as a suggestion rather than a decision. Never auto-merge on name alone, however confident the score looks.

The merge deserves more fear than it gets. In most CRMs it is destructive and irreversible: survivorship rules decide which values win, and notes, attachments and custom object links can be silently dropped. Write them down. Original source takes the oldest record's value, the field everyone quietly corrupts. Contact details take the most recent non-null. The record on an open opportunity survives. Snapshot the loser first, and queue anything below your confidence threshold for a human.

Speed to lead, honestly

The famous five-minute multiplier comes from research on high-volume inbound web forms, most of it over a decade old and from a market that is not yours. Much of the effect is confounded too, because organisations that respond in minutes are better run in several other ways at once.

What survives scrutiny is a pattern, not a number. Response inside the first hour clearly beats response the next day, the curve is steep early and flattens fast, and the effect concentrates in high-intent inbound: demo, pricing and contact-sales forms. For a content download, an immediate call reads as surveillance.

So measure your own. Bucket last quarter's inbound by time to first human touch, five minutes, an hour, a day, longer, and compare meeting rates within each source. An afternoon's work, and the answer is about your market.

An SLA without teeth is a wish

Four things make an SLA real. A defined start event: when the lead became assignable after enrichment and deduplication, not when the form was submitted. That gap is worth reporting on its own. A stated target. Measurement by median and 90th percentile, because averages hide the tail. And an escalation that reassigns the lead on breach.

That last one is the teeth; the rest is a chart. Publish the numbers weekly by rep name and treat a miss as a process failure.

The audit: ten leads from last month

Pick ten leads from last month across your main sources. For each, reconstruct one row: the originating event's timestamp, when enrichment completed, when it was assigned and to whom, when a human first touched it, every stage transition and its reason, whether a duplicate existed, and who owns it now.

Most teams find all of these: records with no enrichment timestamp, leads assigned to someone who left, a stage set by an automation whose entry criteria the record fails, at least one duplicate, and a median time to first touch several times the SLA. Ten rows, an hour, more truth than a quarter of dashboards. It is the readiness audit applied to one layer.

Then the caveat. Routing and lifecycle hygiene stop leakage, they do not create demand, and faster delivery of bad leads changes little. An elaborate model that needs perfect enrichment also fails more often than a crude one. And stages are a compromise anyway, because real buying committees do not move through them. The model works to the degree that everyone reads it the same way, which argues for fewer stages, sharper definitions and a single record the rules compute from.

Frequently asked questions

Why do lifecycle stages stop being accurate?

Because stages get added and never removed, several automations and manual edits write to the same field, and the definitions were entry criteria only, with no exit criteria, no single owner of the transition and no expected duration. Within two quarters one stage holds most of the database and two teams report different monthly counts.

What is the best lead routing model?

Account ownership first, because splitting an account across reps costs more than any delay. Territory as a constraint on who is eligible rather than the rule, since territories skew and drift and empty enrichment fields create a dumping ground. Capacity chooses within that set on an automatically maintained measure, with a per-rep daily cap.

How fast do you have to respond to an inbound lead?

Inside the first hour for high-intent inbound such as demo requests and pricing enquiries. The quoted five-minute multiplier comes from old research on high-volume web forms in another market and is confounded, since fast responders tend to be better run generally. The curve is steep early then flattens, and content downloads differ: an immediate call there backfires.

How should duplicate leads be merged?

Normalise emails by lowercasing and stripping plus tags, match on exact normalised email first, then on company domain plus fuzzy name as a suggestion only, never on name alone. Write survivorship rules down: original source from the oldest record, contact details from the most recent non-null, and the record with an open opportunity wins. Snapshot the loser first, because CRM merges are irreversible.