The lead lifecycle, drawn to scale

Most lifecycle diagrams are drawn by marketing and contain about six boxes, evenly spaced, arrows going one direction. They are wrong in two ways: the states are not evenly sized, and the arrows go backwards constantly.

Here is the version that matches what actually happens, with the automation that belongs at each edge and, more usefully, the places where things fall out.

The states that actually exist

Anonymous. Somebody is reading your site and you know nothing about them. The only automation that belongs here is capture, and the only decision is how aggressively you interrupt them. Most sites over-interrupt, converting a small number of visitors into leads and a larger number into people who will not come back.

Identified. You have an email address or a company. This is a much bigger state than most models allow, because the majority of identified leads are not remotely ready and treating them as pipeline is how forecasts become fiction. The automation at this edge is enrichment and scoring, and the decision is whether this person goes to a human at all.

Engaged. They have done something that costs them time: replied, booked, attended, asked a question. Engagement is the first state that means anything, and the edge into it is the single highest-value automation in the whole system, because response latency here has a steep curve. Route and notify within seconds, not minutes.

Qualified. Somebody human has established that there is a problem with a consequence, a person who owns it, and a plausible path to a decision. Note that this is a judgment, not a score. Systems that automate the transition into qualified are producing a number that feels like pipeline and is not.

In process. The long middle, where most of the calendar time lives, and where lifecycle diagrams are least useful because the state does not change for weeks at a time while the deal is either progressing or dying. The automation that matters here is not advancement, it is detection of stalling, which nothing in the state model itself will tell you.

Committed. A decision has been made and the remaining work is procurement, legal, and signature. Deals die here more often than anyone expects, and almost always for administrative reasons rather than commercial ones.

Won, or lost, or the third one nobody models: dormant. Dormant is not lost. It is a deal with a real problem, a real person, and no current path, and it is the single most under-exploited asset in most CRMs because the state does not exist so everything lands in closed-lost and is never looked at again.

The edges where things fall out

The diagram is only useful if it shows the leaks, so here they are in order of size.

Identified to engaged is the biggest and it is mostly a latency problem, addressed above. Everything else at this edge is secondary to answering fast.

Engaged to qualified is where most bad pipeline is created rather than lost. The failure is optimistic promotion: a friendly conversation gets marked qualified because it felt good, and four months later it is a no-decision loss. The automation that helps is not a rule, it is a required field, specifically a free-text field for the stated consequence. If a rep cannot fill it in with something the prospect actually said, the deal is not qualified, and making that field required at the transition catches an enormous amount of wishful thinking.

In process to committed leaks through single-threading, which the system can detect and almost never does. A deal past a certain age with exactly one associated contact should be flagged automatically. This is a two-line rule and it is worth more than most of what sits on a sales dashboard.

Committed to won leaks through administrative silence. Somebody is waiting on legal, nobody is watching the clock, and three weeks pass. Automation here is straightforward: any deal in committed with no activity for seven days gets a nudge to a human.

And the missing edge, dormant back to engaged, does not exist in most systems at all. Building it is mostly a matter of tagging losses by reason and setting a date, which is the cheapest pipeline generation available to any team and gets skipped because it feels like admin rather than selling.

What not to automate

The transition into qualified. Every attempt to automate this produces a lead-scoring model that eventually correlates with something irrelevant, and the sales team learns to distrust the score, and then you have a system nobody uses that everybody maintains.

Stage advancement based on activity. A deal does not progress because somebody sent an email, and systems that advance on activity produce forecasts that look healthy right up until the end of the quarter.

Closing anything. A rule that auto-closes stale deals will, on a long enough timeline, close a real deal that was quietly progressing over email your system could not see. Flag it, surface it, make a human press the button.

The general shape of the answer is that automation should be excellent at detecting and terrible at deciding. Anything that measures elapsed time, counts relationships, notices silence, or reconciles two systems is worth building. Anything that forms a judgment about whether a human being intends to buy something should stay advisory, because it will be wrong in ways that are invisible until they are expensive.

Draw the states you actually have. Instrument the edges rather than the boxes. And add dormant, because the deals sitting in your closed-lost pile with a date attached are the cheapest quarter you will ever have.

— DealArena Team