How Much of Your Marketing Should a Machine Do?

This week's blog is a bit different. Most weeks I write about something I've seen enough times to have a firm view on. This week I'm writing about something we're in the middle of working out, and I'd really like your help with it.

‍

Two conversations have stuck with me. The first was with an AI engineer who's building a system to take a big chunk of LinkedIn content work off a client's hands: the research, the drafting, the approvals and the scheduling. The second was with the client the next morning, talking through how far we should let it go.

‍

Both were really good conversations. Neither ended with a neat answer.

‍

The question every board is asking

If you've raised in the last year, someone has probably asked what you're doing with AI. Usually it's some version of "can the team do more with what it's got?"

‍

It's a fair question. I've watched marketing delivery move further into automation for years, and a lot of what used to eat a marketer's week can now be done in minutes. Researching a person or a market, pulling quotes out of an article, cutting a 30-minute video into clips, scheduling a fortnight of posts. I'd happily hand most of that over.

‍

The harder question is what happens to the part that makes your content worth reading in the first place.

‍

Where the good stuff comes from

The project works like this. Our strategist interviews each person for half an hour, digs into their experience and builds the themes they'll post about. Those themes become a content calendar, and the calendar becomes posts.

‍

The best material comes from the moment someone says something slightly unexpected and the interviewer goes "hang on, tell me more about that." Then asks again, and again, until four questions later you've got the story only that person could tell.

‍

So the question we keep coming back to is whether a machine can learn to notice that moment.

‍

The engineer put the risk plainly. If every third person comes out of the process with the same topics, the whole thing is useless. You end up with general posts about cost optimisation, which is exactly what's already filling everyone's feed.

‍

What surprised me

I went into the conversation fairly sure the interview was the part to protect. I came out less sure.

‍

Carly Osman Holme did some research recently on how people now tell ChatGPT things they wouldn't tell their friends. Which made me wonder whether someone talking to an AI interviewer at 10pm on the sofa, with nobody politely waiting for them to finish so they can ask the next question, might be more honest than they'd be with us on a booked video call.

‍

Then the engineer asked something I hadn't considered. If people are that relaxed, is the video any use? LinkedIn is pushing video harder than ever (it's now got its own spot in the app's navigation), and a slouched recording from the sofa might not be something you'd want to post.

‍

I don't know the answer to either of those yet.

‍

What we're doing about it

We're treating the next group as a test, and we're being upfront with them that they're the first through it. In practice that means:

‍

• Getting the manual version out properly first, because it's hard to automate something you haven't yet done well by hand
• Giving the system the transcripts from our best interviews so it can learn what digging deeper looks like
• Asking our strategist to write down how she actually runs an interview, including the instincts she probably doesn't realise she has
• Keeping a person in the loop to review and sharpen whatever the AI interview produces
• Building a record of everyone's topics so the system can steer people away from saying the same thing as each other

‍

We train people to interview well, so I'm fairly confident we can train a tool too. What I can't tell you yet is how much of the result will still feel like it came from a person.

‍

The questions I'm still chewing on

• Which parts of your marketing would you hand to a machine tomorrow, and which would you protect?
• Is the time AI saves your team going into better thinking, or straight into more output?
• If your competitors are using the same tools as you, what's left that makes you sound like you?
• Would your customers tell an AI more than they'd tell your sales team?
• Is keeping a person in the loop a stepping stone, or the whole point?

‍

The one thing to do this week

Ask your marketing team to map every step it takes to get one piece of content out, from the first idea to it going live. Then mark each step: would we be happy for this to be automated, or would we lose something?

‍

The interesting conversation is in the steps where the team disagrees with each other. Or with you.

‍

Drop me a line and tell me where you've landed. I'll share what comes back (anonymously) in a future blog, along with how our test goes.

‍

PS. Working out where AI fits, and where it doesn't, is a conversation we're having with most of our clients at the moment. If you're having it with your own team and want a second opinion, get in touch.