The brief that never arrives

I’m going to open with a prediction: this week AI will cost one of you some work and you won’t even notice. A client will have the sort of question that used to come straight to you. Maybe a strategy question, something about their shifting market, a technical problem, a new landing page. The sort of thing you are very good at. A few months ago you’d have been their first port of call and there would have been a good chance that conversation led to paid work. This week they’ll ask Claude/ChatGPT instead and you’ll never know the opportunity even existed.

It’s a prediction, but one I am confident in because I already see work being lost this way. I survey agencies every quarter to better understand what is happening in the space and agency leaders are already telling me about this happening. In one founder’s words, “We have lost a lot of business to clients trying to use AI to perform the work themselves.” Another quoted their client’s words: “We’re now getting AI to do some of the creative [written and visual] work that you used to do…” A third had watched a client rebuild their own website using Claude, which “changed the nature of the SEO retainer we are able to offer.”

We know it is happening already, but the cases we know about are potentially just the tip of a worrying iceberg. When clients get an answer they think is “good enough” from their AI sidekicks, the opportunity just disappears. Even when the client isn’t in a position to understand whether that answer is really good enough at all.

Questions lead to briefs

Agency leaders love to tell me that most of their business comes from “repeat and word of mouth”. I’ve written before about how little those labels tell us, and this is another case when those lazy labels are hiding information. A lot of that repeat business started with a client asking for our view. A worry about the marketing, a pattern they’d seen, something a competitor is doing. We answer the question well, show our expertise in the area and suggest a next step, and that turns into a piece of work. We rarely think of these conversations as working on new business as it just feels like being helpful. When we think about the conversations that lead to work, though, a surprising amount starts out in this way.

Whoever answers the question is effectively writing the first draft of the brief. When that is AI rather than us we’re far less likely to land work.

Order-takers lose out first

The agencies who are most exposed to this shift are the ones David C Baker would call “order-takers”. The agencies reliably turning around whatever work the client asks for. It feels like the way to build a productive business, but when the model is “Take the order. Deliver the thing” we’re competing with everyone and everything on delivery. That was sometimes tough when the competition was offshore. It’s going to be brutal when we’re instead competing against data centres full of the fastest order-takers ever built.

Order-taking agencies face a double-whammy too. Our clients are rarely experts in the things we do. When we deliver work directed by non-experts it’s usually the wrong work and very rarely the best work we should be doing. That means it delivers less for the client, which makes us look less valuable and further closes the gap between us and our client’s AI. A dangerous spiral we don’t want to get caught in.

The gap widens again when our clients see us as trusted experts. We need clients to understand that our expertise will be deeper, more informed and more based in our particular understanding of their business than the answer their chatbot will serve them. More importantly, they need to understand that our answers will pass through a filter of judgement and experience that AI can’t yet match and that we’ll apply that judgement rather than blindly agreeing with every idea the client has. This is where the value lies today. When producing ideas and content becomes so cheap, the value shifts to being the filter.

Is our judgement really safe?

Many would say that our thinking, taste and judgement are only a temporary moat and that AI will cross that soon enough. No one can say for sure, but I personally believe that AI is nowhere near as close to crossing those moats as some would have us believe. Talk to enough people about AI and you will soon spot the pattern of people believing AI is great at everything except the one thing they are personally good at. When that many people say the same thing about their own field, you have to conclude there is a ceiling on what AI can currently do in most areas. AI is exceptional at code, which probably alters the perception of how capable it is to those in tech. In other areas I think that ceiling is lower than many suggest.

The problem is that AI will never admit where that ceiling is. I don’t recall an LLM giving me a response along the lines of “I’ve done that, but I don’t think it is very good”. Instead it acts like the most overconfident guy in the pub with a couple of pints of bravado inside him. It’s never “I’ll try, but you might want to check the results with someone who knows better”, but instead more “Hold my beer”. It’s a very human trait to confuse confidence with competence. So when AI gives a client its confident recommendation about an issue they are not expert in, it might not be obvious when those recommendations don’t hold as much value as ours would.

So the thinking might be safe, but our place as the first port of call for answers might not be. Whether or not we are able to give a better answer is irrelevant if we are never asked or the client doesn’t trust that our answer will be better. I am sure the marketers reading this will be champing at the bit to shout “It’s all down to positioning” and they’re right.

How agencies undermine that position

Our natural position as agencies is to be that trusted expert for our clients, yet it’s easy to undermine it and become order-takers. Reports are a classic example of this and I’ll call out the SEO agencies for being the group with the highest proportion of offenders. The default report in the SEO agency space has page upon page of automated data no one cares about drowning out the tiny proportion of analysis and direction that made it into the report. If we’re now competing against clients’ own AI tools for the work, we really have to stop leading with the parts it can replace and burying the parts it can’t. Instead we should be opening with the thinking: What are we seeing? What are we suggesting next? Data gets moved to an appendix or supporting dashboard and we ensure that clients who only read a single page are getting the part their AI isn’t going to give them.

Reports are an easy example, but account management calls are often as bad. Short chat, lots of recap, read out some data and then ask if there is anything the client wants us to focus on. It’s the same trap as the SEO report - heavy on the replaceable, light on the high-value parts. The trap is evident all over the agency space: websites that are lists of services, email shots that are roundups of uninteresting news, case studies that talk about the output but never the decisions, proposals that just echo the brief and add a price. All of these examples signal that the doing is what is important and not the thinking.

Agencies falling into these traps need to reverse the direction to protect their value. There are things you could do this week to increase your chances of being the first port of call with a client’s next valuable question: a change to report formats to open with recommendations, an unprompted note about changes you have seen, a proactive suggestion for a change on the account. None of these take long and each one shows the client something we can give them that AI can’t.

At the start of this article I predicted that AI would cost one of you some work this week without you noticing. What are you going to do to make sure it’s not you?