Chapter 29

Automation

Automate the noticing, not the talking.

What changed

Marketing automation is usually built as workflows. Someone fills in a form, and five emails follow. Points pile up against a contact until a score tips them over to sales. The work is judged by how many emails go out and how many leads cross the line.

That misses what automation is good at. A machine never forgets to look. It can see a quote viewed three times and not answered, usage halved, or a renewal sixty days away. Most businesses miss those moments, because nobody's watching.

What's new is that an AI assistant can now hold the conversation too, and many firms are handing it the front door. So the question of who answers, a machine or a person, has become the one most teams get wrong.

What it is for

Automation has one job: to notice a moment while it's live, and get the right answer to the buyer in time. The answer may come from a machine or a person.

It serves three stages most directly. Answered, because each moment carries a question, and automation makes sure it gets a reply. Bought, because many sales are lost to delay and silence, not to rivals. And Kept, because falling usage and an approaching renewal are the moments where customers decide whether to stay.

It touches Believed, because a wrong answer from your bot is your wrong answer. A Canadian tribunal held Air Canada to what its chatbot told a customer (Chapter 16).

It does little for Known, Found or Preferred. It works on people who have already raised a hand.

Speed matters because a moment is short. An audit published in the Harvard Business Review analysed 1.25 million sales leads (Oldroyd, McElheran and Elkington). Firms that tried to reach a lead within an hour were "nearly seven times as likely to qualify the lead" as firms that waited an hour more. It's a correlation: fast firms may simply be better at selling.

The anatomy of a good one

A good automation runs left to right, from something the buyer did to a record of what happened next.

A signal: something the buyer did [Answered]. A click, a reply, a login, a quote viewed. Never trigger a person on an email open alone; Apple's privacy feature can fire an open when nobody looked.

A rule: the moment it means [Answered]. "Pricing page twice within 48 hours, after an enquiry" means "this buyer is comparing prices". Write the question down. If you can't say what the buyer is asking, you've only got a click.

The route: machine or person? [Bought]. My working rule is to route on stakes for the buyer. Stakes are high when the money is large for them, the decision is hard to reverse, they're upset, or the answer depends on their circumstances.

A simple answer, by machine, from the one truth [Answered · Believed]. It doesn't improvise. If the answer isn't in the approved sources, it says so and offers a person.

High stakes: a named person, with context [Bought · Kept]. One name, not a shared inbox, with everything the machine knows, so the buyer never repeats themselves.

Record what happened, and learn [Kept]. Read the records monthly, and change the rules that aren't working.

The mistake most teams make

They send sales the highest lead score. Most scoring, as run, counts activity: everything a person has done, over months. A moment is a particular combination of actions, close together, that means a buyer is at a particular step. The finance lead who visits the pricing page twice in one afternoon may score 10 points. The junior analyst who has read every blog post for a year may score 140. The score sends sales to the analyst.

The second mistake is putting a machine between the customer and a person. Gartner found 64% of customers would prefer companies did not use AI for customer service, and the top concern was that it would be harder to reach a person (Gartner). That's a stated preference, but it names the fear precisely.

How you know it's working

Keep workflows, emails sent and "deflection" as diagnostics; a high deflection rate can mean a trapped customer. Judge automation by the share of each moment's signals answered within its time limit, the share that moves to the next step, how often a buyer has to repeat themselves, the assistant's accuracy against the one truth, and deals lost where a signal fired and nobody acted.

An example

Fieldline, the invented scheduling-software firm, used to run a lead score. In one deal, a junior coordinator downloaded three guides, crossed the line and got the sales call. She had no budget. Meanwhile the finance lead viewed the proposal four times in two days, going back to pricing each time. Nobody saw it, because viewing a proposal earned five points. The deal went quiet.

Fieldline rebuilt around moments. In the next deal, the finance lead's third view reached the account manager within the hour. Her first line was a question, not a report: "Most finance leads at this stage want year one all-in. Would that help?" She never mentioned what she'd seen.

The full playbook. The Automation Playbook, with the signal-to-moment map, the handover rule sheet and the guardrails spec for a customer-facing AI assistant, is free at rocmode.com/playbooks.

Whose evidence this is. Gartner sells research to customer service leaders, and its customer figures are a stated preference from late 2023. The lead-response audit appeared in HBR, not in a peer-reviewed journal, and I found no peer-reviewed replication.