How to automate lead intake

Somewhere in your business, right now, there is an enquiry waiting. Someone filled in a form, or sent an email, or left a message, and they are quietly deciding how long they will wait for you before they try the next name on their list. Lead intake is the unglamorous work of catching that person before they cool off — capturing what they said, working out how urgent it is, filing it where it won't get lost, telling the right human, and making sure someone actually follows up. It is mostly plumbing. But bad plumbing loses deals, and most small teams are running it by hand without realising how much it costs them.

This page is a straight walk through how to automate that job properly — what to hand to software, what to keep for a person, and the one narrow place where AI genuinely earns its keep. We build these pipelines for a living, and this exact one runs the contact form on this very site, so none of what follows is theory.

Why leads cool off

A lead is at its warmest the second it lands. The person has a problem in their head right now, they have decided to do something about it, and they have taken the trouble to write to you. Every hour after that, the temperature drops. They get distracted. A competitor replies first. The urgency that made them fill in your form fades back into "I'll sort it later." Every enquiry that sits in an inbox over a weekend is a deal quietly cooling off.

Here is the maths that most owners never sit down and do. Say your average job is worth £1,500. Say you get a genuine, ready-to-buy enquiry roughly once a fortnight, mixed in with the tyre-kickers. If you miss one of those a quarter — one Friday-afternoon message that you only saw on Monday, by which point they'd booked someone else — that is £6,000 a year walking out of the door. The fix is not a new salesperson or a bigger marketing budget. The fix is a bit of automation that makes sure no message ever waits on a human to be noticed. That asymmetry is the whole argument: the cost of one lost lead dwarfs the cost of the plumbing that would have caught it.

The other quiet cost is the tax on your own attention. When intake is manual, part of your brain is always half-watching the inbox, because you know that missing one is expensive. That low hum of vigilance is exactly the thing a small team should be automating away. We've written before about what actually gives a small team its Friday back, and it is almost never the flashy stuff — it's removing the ten little jobs you didn't know you were doing.

The five steps of a lead's journey

Every enquiry, whatever your business, travels the same five steps. Naming them is half the work, because once you can see them separately you can see which ones are safe to automate.

1. Capture

Getting the enquiry out of wherever it landed — a web form, an email, a DM, a phone note — and into structured data you can act on. The trick here is to ask for less, not more. A form with fourteen fields gets abandoned; a form with three short questions gets finished. Capture the minimum you need to triage, and get the rest in the conversation.

2. Triage

Reading what they actually said and deciding what it means: is this urgent, is it a good fit, roughly how big is it, what are they really asking for. This is the one step that needs judgment, and we'll come back to it — it's where AI fits, and nowhere else does.

3. File

Putting a proper record somewhere durable, tagged and searchable, so that "did we ever hear back from that plumber in Espoo?" has an answer. An inbox is not a filing system. A record in a CRM — even a free Notion one — is.

4. Notify

Telling the right person, in the place they actually look, fast. For most small teams that's a Slack card or a phone notification, not another email buried in the same inbox that caused the problem. The notification should carry enough context that they can decide in five seconds whether to drop everything.

5. Follow up

Making sure the ball never gets dropped after the first reply. This is where most manual systems fail — not at capture, but three days later, when the lead is waiting on you and there's no task anywhere reminding anyone. A tracked follow-up task with an owner and a date is what turns "we meant to get back to them" into revenue.

Where AI belongs — and where it absolutely does not

This is the part most people get backwards, so let's be precise. Of those five steps, four are deterministic: given the same input, you always want the same output. Filing follows fixed rules. Prioritising follows fixed rules. Notifying follows fixed rules. Creating the follow-up task follows fixed rules. None of these needs intelligence — they need reliability, and reliability is exactly what plain automation gives you.

Only one step is genuinely AI-shaped: reading a messy, freeform enquiry and summarising it into something a human can act on in ten seconds. When someone writes three rambling paragraphs about their situation, pulling out "they need a booking system, they're losing walk-ins, they want it before the summer season" is a language problem. That's what large language models are actually good at. So that one slice — turning waffle into a tight triage brief — is where we point AI, and we point it nowhere else.

Why not just let AI do the whole pipeline, since it can? Because you'd be paying for a Ferrari to do the school run and adding failure modes for free. A rule that says "budget over £5k, tag as high value" is right 100% of the time, costs nothing, and never has a bad day. Ask a model to make that same call and you've introduced latency, an API bill, and a small but real chance it decides £4,900 rounds up. Forcing AI onto deterministic plumbing is the classic small-team money-waster. We wrote a whole piece on how to tell the two apart — AI versus automation, and how to know which one you actually need — and the short version is: use AI only for the step that requires reading and judgment, and let boring, testable rules do everything else.

The goal isn't a system that never needs a human. It's a system that only needs a human when it should — and that never makes a human do what a rule could.

We run this on ourselves

The contact form on this site is not a demo. It is the real pipeline, running live, and it is the flagship example of the exact work we sell. Here's what happens when you fill it in.

You answer three short questions — that's the whole form, deliberately. The moment you submit, AI reads what you wrote and writes a short triage brief: what you seem to need, how urgent it looks, whether it's a fit. A tagged record files itself into Notion — our CRM — with a respond-by date already set based on how the enquiry reads. The team gets a Slack card with your brief attached, so we can decide in seconds whether this is a drop-everything one. And follow-up tasks are created automatically, assigned, so nothing waits on someone remembering. By the time a human looks, the thinking is done and all that's left is the part that should be human: the reply.

If you want the full breakdown — the fields, the rules, the fail-soft behaviour when the AI is unavailable, the actual Notion structure — we've published the whole thing as the lead-operations teardown. We dogfood it because it's the honest way to sell it: the plumbing that runs our own business is the plumbing we'll build for yours.

Start without AI at all

Here's the advice that costs us money to give and is still the right advice: build the automation-only core first, and don't add AI until the boring version is solid.

The core is four of the five steps — capture, file, notify, task — with a human reading the raw enquiry themselves. No AI anywhere. A form drops a structured record into a CRM, tags it by a couple of simple rules, pings you, and creates a follow-up task with a due date. That's it. That version alone kills the weekend-inbox problem, because the lead is now filed and flagged and owned the instant it arrives, whether or not anyone's summarised it yet. You read the raw text yourself for now — it's ten seconds — and the machine handles the parts you used to forget.

We've packaged exactly this as a free Notion lead CRM you can start using today. It is the boring, reliable core with none of the AI — capture, file, notify, track — and it's genuinely enough for a lot of small teams. Get that running, feel what it's like to never lose a lead again, and only then ask whether the ten-second read is worth automating. Usually it becomes worth it around the point where you're getting enough enquiries that the reading itself starts to eat your day. Not before.

The rules that do the "thinking" without AI

People assume that triage — deciding what a lead is worth and how fast to respond — must need intelligence. Mostly it needs a handful of if-this-then-that rules. Here are the ones that carry the load:

• Dedupe by email. If an enquiry comes in from an address already in the CRM, attach it to the existing record instead of creating a duplicate. One person, one file. This is a lookup, not a judgment call.
• Priority from timeline. Ask "when do you need this?" as a multiple choice — "this month / this quarter / just exploring" — and map the answer straight to a priority. "This month" jumps the queue. No reading required; the person told you.
• Deal size from budget. A budget band on the form maps to a value tag. Over a threshold, it's flagged high-value and routed to a senior person. A number compared against a number.
• A respond-by SLA date. The moment the record is created, stamp a respond-by date on it — say, next business day for normal enquiries, same day for high-priority ones. Now "are we late on anyone?" is a filter, not a worry.

Notice what every one of these has in common: the input is already structured, or trivially structured, and the output is fixed. That's the signature of a job for automation, not AI. The intelligence you thought you needed was mostly just asking the right question on the form so the answer arrives pre-sorted.

Common mistakes

Most broken intake fails in one of four predictable ways.

• The "form that just emails me" trap. A contact form that fires off an email and does nothing else has automated precisely nothing — it's moved the problem into the same overloaded inbox that was losing leads in the first place. The enquiry still isn't filed, tagged, prioritised, or tracked. It just feels automated.
• No owner. A lead that belongs to "the team" belongs to no one. If a record lands without a named person attached, everyone assumes someone else has it. Assignment has to happen automatically, at intake, every time.
• No tracked follow-up task. The first reply is easy to remember; the second one, three days later, is where deals die. If there isn't a task somewhere with a due date, "I'll circle back Thursday" is a wish, not a plan.
• Over-automating the human reply. The opposite mistake, and just as costly. The one thing you should not hand to a machine is the actual conversation with the prospect. Auto-sending a canned "thanks for your enquiry" is fine; auto-sending the substantive reply reads as exactly what it is, and people can smell it. Automate everything up to the reply. Let a human write the reply.

A worked example, with real numbers

Take a small studio getting around 10 enquiries a week through a plain contact form that emails the owner. Today, each one costs roughly 5 minutes of read-it, decide-if-it-matters, copy-the-details-into-a-spreadsheet, remind-yourself-to-reply. That's 50 minutes a week — call it 43 hours a year — of pure administrative shuffling, none of which is talking to a customer.

Worse than the hours is the leak. About once a fortnight, a good lead lands late on Friday. The owner glances at it on their phone, thinks "Monday," and by Monday two of those have gone quiet because someone else replied over the weekend. At a £1,500 average job, that's the £6,000-a-year hole we opened this page with.

Now run the same 10 weekly enquiries through an intake pipeline. Capture is instant and structured. Filing, tagging, priority, respond-by date, Slack ping, and follow-up task all happen with no human touch. The owner's job shrinks to reading a tidy brief and writing the reply — the only part that was ever really theirs. The 50 weekly minutes drop to maybe 10. The Friday lead is filed, flagged high-priority, and sitting on a phone as a Slack card with a same-day respond-by date before the owner has even put their coffee down. The leak closes. The recovered £6,000 pays for the build many times over in the first year, and the 40-odd reclaimed hours are just gravy.

How it connects to the rest

Lead intake isn't a closed loop — it's the front door of a longer hallway. A lead that turns into a "yes" should hand off cleanly to whatever comes next: onboarding, a project kickoff, an invoice, a welcome sequence. If your intake system and your delivery system don't talk, you've just moved the manual re-typing from the front of the process to the middle — someone copies the won deal's details into a fresh project by hand, and errors creep in at exactly the moment a new client is forming their first impression.

The clean version: when a record's status flips to "won," that same automation kicks the next process into life — creates the project, files the client's details, sets up the first onboarding tasks, all from the data you already captured at intake. No re-keying, no gaps. The information a prospect gave you in three form fields on day one should still be doing useful work on day thirty. That's the difference between a form and a pipeline — and it's the difference between chasing leads and simply, quietly, catching every one.