Three leaks, and they are always the same three
Your enquiries are not disappearing at random. They drain out at three specific points, and none of them is where teams usually look.
Leak one — the unwatched channel. A web form posts to a shared mailbox nobody owns. A LinkedIn message sits in someone's personal inbox. A missed call never reaches the CRM because your telephony and your CRM have never been connected.
Leak two — the unowned record. The lead is in the CRM, but the owner field is empty or set to a team. Everyone assumes a colleague has it. Nobody does.
Leak three — the one-touch lead. Someone replies once, no next action is written down, and the record ages quietly until it is dead.
None of this is a software problem. It is an ownership and memory problem — which is exactly why buying another CRM will not move the numbers.
Three numbers that will tell you what you are losing
Before anyone talks about AI, pull three figures. You already have them.
Median time to first response, measured from the timestamp on the enquiry, not from when someone opened it. Share of enquiries with no named owner — count empty and team-level owner fields as unowned. Share of open opportunities with no activity in 21 days.
All three come out of your mail system and your CRM with a saved view or a simple export. No instrumentation, no project.
Their value is that they are arguable. Your team can dispute the number. Nobody can dispute the question. And in most cases they show the bottleneck is not the volume coming in — it is what happens in the first 48 hours after it lands. That is good news: follow-up is far cheaper to fix than acquisition.
What the system actually does, step by step
Four mechanics, in order. This is not a black box.
Capture. Web forms post through a webhook. Mailboxes connect over the Graph or IMAP API. Call logs come in from your telephony platform. Every channel lands as one record, deduplicated on email address and company domain so the same prospect does not enter three times.
Enrich. The system maps incoming fields to your CRM objects, attaches the source, the campaign and the full thread, and flags what is missing before a human has to ask for it.
Prioritise. A weighted score runs against criteria you defined — segment, deal size, urgency signals, fit. It orders the queue. It does not decide.
Assign and schedule. An owner is set by your rule — territory, segment or round-robin — and a next-action date is written as a required field. An SLA timer starts on first response.
Most of the value sits in capture and assignment. A lead that enters with an owner and a due date stops getting lost, whatever the score says. The full mechanics are on the Sales & Lead AI page.
A score sorts a queue. It does not qualify.
Be clear about what a score is: a weighted sum of criteria you chose. Its quality depends on those criteria, not on how clever the maths is.
Run this test before you trust it. Take the last six months of closed-won deals and check them against their scores. If your wins were not concentrated in the high band, the criteria are wrong — not the tool. Fix the criteria, re-run, check again.
And hold one rule: a low score means *look at this later*. It never means *discard*. The most expensive mistake we see is treating an automatic ranking as a decision.
What stays with your team, and why
The conversation, the negotiation and the decision to walk away stay human. That is not a courtesy — those are precisely the moments when the information that decides the deal is not in the CRM.
Contact rules stay human too, and they go into the workflow as hard stops, not preferences: opt-out lists, suppression lists, accounts flagged as owned by a named person, restricted contacts. Every one is enforced before an action is proposed, and every action is written to an audit log.
A system that can route around those rules is a liability, not an asset.
Why we switch automatic sending off at launch
Yes, the system can send follow-ups by itself. We recommend you do not let it, on a first deployment.
The reason is economic, not ideological. The measurable gain comes from capture and prioritisation. Sending adds risk without adding much of it. A generic message at the wrong moment costs you more than a follow-up you forgot — especially in a market where your name travels.
So the first release drafts and queues; a person approves. Once your criteria have held up over several months and the approval rate is stable, we switch on automatic sending for narrow, bounded cases — and only with your written agreement.
How to start without wasting a quarter
One channel. One segment. Four weeks of measurement against the three numbers above. Then adjust the criteria and widen the scope.
Open every channel at once and you get noise, plus no way to attribute any improvement to any cause. Narrow first, prove, then extend.
Two grounds of application are documented: lead follow-up for estate agencies and sales follow-up for B2B services. Illustrative example: results depend on your pipeline, your criteria and your team.
Next reads
Where to place human validation, and how hard stops are enforced: our guide to human validation.
Why an assistant that answers is not a system that acts inside your tools: chatbot, assistant or agent.
Want the prioritised version for your own processes? The AI Diagnostic returns three opportunities in a few questions.
Illustrative example of how the process can work.