AI automation that removes the manual work between a lead and a sale

How we scope it

Automation projects fail for a predictable reason: the process gets automated as described rather than as performed. Those are rarely the same thing, and the difference is where every unpleasant surprise lives.

So the first thing we do is watch. Someone does the task three times while we take notes, and we write down each step as a sentence with a verb in it. Not "lead management" — "someone opens the shared inbox, reads the enquiry, checks the ZIP against the coverage map". Categories cannot be automated; steps can.

That exercise reliably turns up something nobody had documented. A spreadsheet one person maintains privately. A rule about which customers get called back first that exists only in someone's head. A check somebody does out of habit that turns out to be the reason a particular mistake stopped happening. Those are load-bearing, and an automation that ignores them fails in week two.

We also time each step, because the thing that feels most annoying is frequently not the thing costing the most hours. Our post on what to automate first covers the four tests a good candidate has to pass.

What we build on

We work with the tools you already pay for rather than proposing a migration. In practice most builds sit on a workflow layer such as Zapier, Make or n8n, connected to whatever CRM, calendar and inbox you run.

Where a language model is genuinely useful — reading an unstructured enquiry and pulling out the fields, drafting a reply, summarising a long thread — we use one. Where a plain rule does the job, we use the rule, because rules are cheaper, faster and do not surprise you.

Everything is built so you can see what it did. Every automated action leaves a record, which matters the first time something goes wrong and you need to work out whether the system misfired or the input was odd.

Where automation is the wrong answer

Three situations where we would tell you not to bother, because they come up often enough to be worth stating.

The process is still changing. If how you handle enquiries is different this month from last, automating it locks in a version you are about to abandon. Settle the process first.

The volume is genuinely low. A task done twice a week that takes four minutes is twenty minutes a month. That is not worth a build, whatever the demo looked like.

The step needs judgement every time. If a human has to look at each case and decide, the automation can prepare the decision but cannot make it. That is still useful, and it is a smaller project than people expect.

We would rather say this at the scoping stage than take on work that will not pay back.

What the first project usually looks like

Lead intake, most often, because it is the step that touches revenue directly and the one with the clearest before and after.

An enquiry arrives at nine in the evening. Within seconds there is a structured record with name, phone, service type and source. An acknowledgement goes out telling the person when they will hear back. It is routed to whoever covers that service area. Nobody touched it.

The next morning your team opens a queue that is already sorted rather than an inbox to triage. Two days later anyone who has not replied gets a follow-up automatically, which is usually the part that recovers the most revenue, because those are the enquiries that used to fall through the gap between busy days.

The two numbers to watch are response time and follow-up rate. Both are measurable before you start, which is what makes the second project easy to justify.