What to automate and what to keep human

Most teams frame the automation question as how much. That is the wrong axis. A team that automates the right narrow things firmly will beat a team that automates a broad middle timidly, because the categories behave differently. Some follow-up work has one correct answer whoever does it. The rest is a judgment call, and a judgment call handled by a template produces a customer who knows they are talking to a system.

The two-question test

Before you automate anything in follow-up, ask two questions about the task.

1. Does it have one correct output that does not depend on reading the customer? Sending a reminder two hours before a meeting has one correct output. Answering a customer who says the price is too high does not.
2. Does it commit your company to anything? A price, a hold, a start date, a discount, a promise about availability. If the task can create an expectation you have to honor, a person approves it.

Yes to the first and no to the second means automate it. Anything else stays with a person, or becomes a draft a person approves.

Automate these without hesitation

Memory

This is the most valuable category and the one teams skip most often. Nothing a rep promised on a call should depend on that rep remembering it under pressure: the callback at eleven, the comparison by Thursday, the document the customer asked for, the call after they talk to their partner. A team that automates only this will get more from automation than a team that automates everything else and leaves memory to people.

Routing and assignment

Who gets the lead, who covers the shift, where an overflow call goes, which manager sees an escalation, what happens to a customer whose rep is out today. You can write these rules down, so a machine should run them. They should also be visible, so a rep can see why something landed in their queue. The glossary covers lead assignment in more detail.

Queue building

The list a manager works from in the morning: quotes with no next step, meetings with no confirmation, customers who replied and got nothing back, leads that went past your response standard. Building that list by hand takes an hour and is out of date by ten o'clock. Deciding what to do about each row takes judgment, so leave that part to the manager.

Logistics and confirmations

Scheduled reminders with fixed content: the time, the link or the address, what to have ready, who they will talk with. These pass both questions, provided they carry a real person's name and replies land somewhere a person reads.

Drafting

A drafted message or a drafted call list, prepared with the context of the last conversation, saves real time and raises the floor on quality. The condition is that a person reads it and sends it. An outbound draft that a person approves differs from an automatic send in one important way: who carries the responsibility.

Keep these with a person

• Anything that moves a number. Price, payment terms, discounts, fees. A number that changes is a negotiation.
• Objections. “The price is too high,” “I want to think about it,” “my partner hasn't seen it,” “another company is cheaper.” The right response depends on tone, history and what the customer meant, which is the information a template does not have.
• Complaints and frustration. A customer who is angry about a bill, a delay or a figure gets a person right away. A friendly automated reply to a complaint makes it worse.
• Commitments. Holding a price, promising a start date, confirming an approval, guaranteeing that an offer is available.
• Anything involving payment or account details. Some businesses, such as mortgage lenders and finance companies, fall under federal rules on safeguarding customer information, so decide on purpose what any automated channel may touch.
• The first conversation with a live inbound caller. Someone who dialed your company and is on the line wants a person. That differs from a call nobody answered.

The middle: draft and approve

Between fully automatic and fully manual sits the category many teams underuse. Software prepares the work and a manager or rep decides whether it goes: a drafted follow-up call list for the morning with a reason next to each name, a suggested reply to a text with the relevant part of the earlier call cited, a proposed next step on a quote that has gone quiet.

Follow-up often breaks down here, because the reps are busy and nobody has time to reconstruct fifty conversations. Preparing the work and asking a person to approve it fixes the time problem without moving responsibility onto a system. The glossary calls this an approved action.

What changes when the automation talks

Text and voice automation carry legal weight that a reminder in your own system does not. Under 47 CFR 64.1200, telephone solicitations to a residential subscriber may not start before 8 a.m. or after 9 p.m. local time at the called party's location, and the FTC's Telemarketing Sales Rule sets the same calling window at 16 CFR 310.4. Calls placed with an automatic telephone dialing system or an artificial or prerecorded voice carry their own consent requirements under the Telephone Consumer Protection Act, and in a February 2024 declaratory ruling (FCC 24-17) the FCC confirmed that the restrictions on artificial or prerecorded voice cover current AI technologies that generate human voices.

Three practical consequences. Decide with counsel which consent your company holds for each channel before you switch anything on. Make sure a stop request reaches every system that can contact the customer. And keep a record of what each automated system sent and when, because someday you will need to answer a question about a specific message. This is general information, not legal advice.

One customer, one week

A single case makes the line easier to see. Synthetic example: on Monday evening, Dana asks for a quote on a home and umbrella policy through an independent insurance agency's website.

• Automatic: the lead routes to the rep who covers Monday evenings and reaches her phone within seconds, with the request and the source attached.
• Human: the first call, and the first text that follows it.
• Automatic: a reminder on Wednesday morning, because the rep told Dana she would have the comparison by Wednesday.
• Human: the comparison conversation, because it is about price.
• Automatic: the meeting confirmation with the time, the video link and the rep's name, scheduled for Friday afternoon.
• Human: Dana's Thursday reply saying her husband thinks the premium is high.
• Draft and approve: the Saturday morning follow-up after she misses the meeting, prepared with the context of the week and sent by the rep after she reads it.

Nothing in that week is exotic. It works because the machine never speaks for the company about money, and the person never has to remember anything.

How to tell you automated the wrong thing

• Customers reply to your messages with questions a person would not need to ask, such as “who is this?”
• Your reply rate falls while your send volume rises.
• Reps start apologizing for messages sent in their name.
• Your notes look complete and your recordings tell a different story.
• Nobody can say what a given customer was told last week without listening to the call.

What to do this week

1. List every automated message your team sends and run each one through the two questions.
2. Turn off anything that answers an objection or quotes a figure automatically.
3. Automate one memory task completely, starting with promised callbacks.
4. Give a manager one prepared queue each morning and keep the decisions with people.
5. Confirm with counsel which consent you hold for each automated channel.