The lead management process, end to end

Lead management is a chain of custody. Six parties hold the record in turn — the capture surface, the identity layer, the qualification rule, the routing rule, the nurture engine and finally a human being — and at every handover some state has to travel with it. Leads are almost never lost inside a stage. They are lost crossing between two, when the source did not travel, or the assignment fired at somebody on leave, or the reply arrived on a channel the sequence was not watching.

Lead management is a custody chain, and the losses are at the joins

The reason this framing beats the usual stage list is diagnostic. A stage list tells you what happens; it does not tell you where to look when the number is bad. A custody chain tells you both, because every handover has three properties you can check independently.

• What must travel. The state the next holder needs and cannot reconstruct. Source is the classic example: it exists at the moment of capture and is unrecoverable an hour later.
• What proves it happened. A row, a timestamp, a status transition — something you can count. A handover with no artefact is one you cannot audit, which in practice means one you cannot fix.
• What happens if it silently fails. Almost every lead-management failure is silent. Nothing errors. The lead just sits somewhere until it stops being interesting, and the only symptom is a number at the bottom of the funnel that nobody can explain.

Hold those three questions against each of the five boundaries below and the process audits itself.

Handover one: capture, where source is recorded or invented later

Every route in should land in one place with its origin stamped at creation. Web forms, chat, inbound calls, event lists, referrals, replies to outbound, marketplace enquiries.

Two failures live here and both are mundane. Multiple destinations — a form to one inbox, chat to a Slack channel, calls to whoever picks up — means there is no single view and therefore no denominator to measure against. Missing attribution means you can never answer which channel funds itself, so budget decisions get made on impressions.

Record the source at creation. Retrofitting attribution is guesswork wearing a chart, and the guess is systematically biased toward whatever channel your team remembers.

There is a quieter loss at this boundary that most teams never see. An import or a capture that carries no usable identifier — no email, no phone number, no LinkedIn URL — cannot become a record at all. In Autocloz's shared import path such a row is counted as skipped rather than created, which is honest, but it means the number of rows you uploaded and the number of leads you now have are two different numbers and the difference is worth reading rather than assuming.

Handover two: identity, and the match rule that decides your errors

Before anything is decided about a lead, the system has to answer whether this person is already known. That is an identity question rather than a cleanup task, and the answer depends entirely on the match rule.

Autocloz's rule is a disjunction: a new lead matches an existing one in the same workspace if the email matches case-insensitively, or the phone number matches once every non-digit is stripped, or the LinkedIn URL matches as an exact string. First match wins.

Read that as a set of trade-offs rather than as a feature, because each clause has a specific failure.

• Email alone misses the same human who filled one form with a work address and another with a personal one.
• Adding phone catches that case, and collides two colleagues who both typed the company switchboard number into a form. They become one lead, and one of them is now invisible.
• Exact-string LinkedIn treats linkedin.com/in/name and www.linkedin.com/in/name/ as two different people, because they are two different strings.

None of those is a bug to be fixed; they are the errors that any disjunctive rule produces. The work is to normalise before you match — lowercase the email, strip the digits, canonicalise the URL form — and to decide which error you would rather have. A merge that loses a person is worse than a duplicate that annoys two reps, so if you must choose, choose the rule that under-matches and catch the rest with a review queue. The mechanics of getting a list into that shape before it ever reaches the CRM are in how to build a targeted B2B lead list.

Handover three: qualification, where a score becomes a queue order

Qualification exists to answer one question: who gets worked first. It earns its complexity only when you have more leads than capacity. Below that line, work them all and skip the machinery entirely.

The state that must travel across this boundary is not the score. It is the reason — which criteria fired, and when. A score of 68 tells a rep nothing; "matches the size band, title matched, replied to the last email nine days ago" tells them how to open. Autocloz returns the per-rule breakdown alongside the number for exactly this reason, sorted so the largest contributor is first.

Two design rules survive contact with real data. Keep fit and engagement as separate numbers, because a single combined score cannot distinguish high fit with no engagement — a nurture case — from low fit with high engagement, which is a polite decline. And re-derive the fit criteria from closed-won records rather than from a workshop, because the profile a team believes it sells to and the one it actually closes are rarely the same shape. The model itself, including where to put the threshold, is worked through in what lead scoring is and how to start.

Handover four: routing, and what the speed evidence actually says

Routing decides who holds the record next. Round robin, territory, vertical specialisation, or assignment by seniority — the rule matters less than whether the handover completes without a human in the loop.

The distribution is more instructive than the multiplier everyone quotes. In March 2011 Harvard Business Review published "The Short Life of Online Sales Leads" by James B. Oldroyd, Kristina McElheran and David Elkington. Alongside the decay study it is usually cited for, the authors audited 2,241 US companies by submitting a web-generated test lead to each and measuring the reply. Of those companies, 37% responded within an hour, 16% within one to 24 hours, 24% took more than 24 hours, and 23% never responded at all. Among companies that responded within 30 days at all, the average response time was 42 hours.

Two honest caveats, because this figure is repeated without either. It is 2011 data about web-form enquiries in a very different buying environment, so the exact percentages are period-specific. And the audit measures a test lead, which is a proxy for a real one rather than the thing itself. What survives both caveats is the shape of the distribution: a quarter of companies never replied at all, which is a routing and ownership failure rather than a selling one, and it is the failure a rule can actually fix.

The most common routing failure is not a bad rule. It is a correct rule firing at somebody on leave, with no escalation behind it, so the lead sits assigned and untouched and never appears on any list of problems. Build the escalation before you tune the rule; lead routing best practices covers the four assignment models and the visibility half that most designs forget.

Handover five: nurture, and the exit condition that makes it safe

Most leads are not ready, and the real choice is between nurturing them and discarding people you have already paid to acquire. Nurture properly means different things by segment: high fit with low engagement deserves genuine, low-pressure presence over months; low fit deserves a fast, polite disqualification, because leaving it in the machine consumes attention and inflates every funnel metric downstream; and previously engaged contacts who went quiet are the best list you own and the least worked.

What makes nurture safe is the exit condition, and this is the boundary where an honest description of a real system matters more than a principle. In Autocloz an email sequence stops on a genuine human reply and on a hard bounce, and it deliberately does not stop on an out-of-office, because the recipient is temporarily away rather than disengaged. That much is a sound design.

What it does not do is stop on a reply that arrives on a different channel. An inbound SMS or WhatsApp message does not touch the email enrolment at all; a recognised opt-out keyword adds a wildcard do-not-contact entry for that phone number and closes the conversation, which stops future texts and calls to that number, while the email steps continue on schedule. If your cadence mixes channels, treat cross-channel stop as a thing you check by hand rather than a thing you assume, and design the sequence so the email arm is short enough that the gap cannot embarrass you.

Handover six: the sales handover, and what has to travel with it

The last handover loses the most value, and it is almost always a context problem rather than a process one.

Define the qualification threshold explicitly and share it, then make the handover carry the history rather than a summary of it. The technical shape of that matters. Autocloz builds a lead's timeline by federating six separate sources at read time — operator events, cross-channel touches, chat messages, email threads and their messages, voice calls, and CRM activities — rather than reading one table. The practical consequence is that a handover is a change of owner, not a re-briefing, because nothing had to be copied anywhere for the next person to see it. The replies themselves land in one queue across every channel, which is what keeps the receiving rep from having to reconstruct the thread from three tools.

Where email lives in one tool, calls in another and forms in a third, the receiving rep gets a name and a number and opens the conversation by asking questions the person has already answered. That is the loss, and it is invisible in every report you have.

A worked audit that finds your leak

Numbers below are an illustrative worked example, not measured customer data. The method is the point; substitute your own.

Take one cohort — every lead captured in September — and count it at every boundary.

1. Submitted: 1,000. Every form fill, chat and inbound call recorded at the capture surface.
2. Became records: 916. Eighty-four carried no email, no phone and no LinkedIn URL, so nothing could be created. Loss: 8.4%.
3. Distinct people: 845. Seventy-one matched existing leads. On inspection nine of those matches were collisions on a shared switchboard number, so nine people are now invisible inside someone else's record.
4. Above the qualification bar: 340.
5. Assigned: 340, reachable owners: 314. Twenty-six went to a rep on leave with no escalation rule behind the assignment.
6. First-touched within the stated hour: 121.
7. Conversations: 44. Accepted by sales: 18.

Now read the ratios rather than the counts. Capture loses 8%, identity loses 1%, qualification is a policy decision rather than a loss, and routing loses 8% of what it was given. Then step five to step six loses 61% — 193 qualified, assigned, reachable leads that nobody contacted inside the window their own team published.

That is the leak, and it is not a copywriting problem, a scoring problem or a tooling problem. It is a capacity-and-alerting problem, and it is invisible in a top-to-bottom conversion rate that reports 1.8% and looks like a targeting issue. This is the entire argument for counting at boundaries: the same 1,000 leads produce a completely different remedy depending on where you measure.

If the number you want to move is the last one, the definitional work at the qualification bar is covered in SQL versus MQL and how to move leads between them, and the ICP generator turns a vague description of the target into fields a filter can actually act on.

Autocloz's free plan covers 5 users, 10 mailboxes and 100,000 contacts, with the pipeline, custom fields and the cross-channel timeline included — start free and run the audit above on a real cohort rather than a hypothetical one.

What the lead record will not tell you

Five limits, each specific, and each one a thing to know before you build a decision on top of it.

A recalculated score compounds. The scoring engine starts from whatever value is already on the lead and adds the matching rules' deltas to it, then writes the result back. Recalculate twice without the underlying facts changing and the number rises, clamped at 100. Treat the score as a ranking that is refreshed, not as a quantity that accumulates, and re-derive rather than re-apply when you change a rule.

The engagement rule measures your sending, not their interest. The engagement_recent rule fires when the lead's last-activity timestamp falls inside a window. That timestamp is stamped by anything that reaches the mail transport — a successful send and a hard failure alike — and it is also refreshed when an import updates an existing row. So a lead that has ignored you completely still looks recently engaged, because you emailed them. This is the single most common way a scoring model quietly measures its own output.

One shipped rule reads the composite score. The rule named email_verified compares the lead's overall score against a threshold rather than consulting any verification verdict. It is a score-above-threshold rule with a misleading name, so a model that includes it is partly scoring on its own previous output.

A tag with a capital letter is unreachable by the tag filter. The leads list lowercases the tag you filter by before matching it against the stored array, and the containment check is case-sensitive on the stored value. Store tags in one case, lowercase, and make that a written convention rather than a habit.

The CSV export is nine columns. Email, name, company, status, phone, LinkedIn URL, score, source and creation date, capped at 100,000 rows. Custom fields, notes, tags, owner and the entire activity timeline are not in it. If your exit plan is a CSV, that is the plan's actual scope, and it is worth knowing on the way in rather than on the way out — the same question is worth asking of whatever you are moving from.

None of the six handovers is a skill problem. They are all rule problems, which is the encouraging part: rules are cheaper than hiring, and the audit above tells you which one to write first.