Show Notes
Agencies are rushing to adopt AI, and most of them are skipping the step that decides whether it works. AI is a prediction machine – it has no nuanced understanding of how your agency does the work, so it needs every process spelled out in frankly painful detail. The problem is that in creative businesses the processes usually aren’t written down at all; they live in people’s heads. In this episode, Gareth puts Callum in the hot seat on why SOPs come before AI rather than after it, what to do when an agency tells you it has no processes to show you, why prompt libraries turned out to be a dead end, how to get a process out of the person who has been doing it for fifteen years, and why the agency time-based model is the thing really under threat here.
In this episode
- Why SOPs precede AI rather than follow it – AI is a prediction machine working on probabilities, so without the scaffolding of a defined process you will never get consistent outputs, no matter how good the tool is
- Why standard operating procedures are so uncommon in agencies – in creative businesses the process is tacit knowledge sitting in people’s heads, and one founder told Callum outright that “that isn’t terminology we use in our business”
- What that actually costs you, before AI even enters the conversation – no accountability for who does what at which stage, no way to train new people consistently, no way to know whether you are over-servicing, and one core process running in ten slightly different flavours because everyone adds their own approach
- Why prompt libraries were a dead end – workflows are a step above them, because a standardised process means AI already knows every step rather than waiting to be told what to do each time
- Learn through AI rather than from it – its own data on how agency work should be done will give you very average outputs, so the process has to come from your senior person, the one who now teaches it rather than does it. Get that and it is close to downloading their brain into the system
- How Callum runs a blueprint session – start where the pain is, not with the process, then sit down with the department leader for an hour and a half and get to the five to ten real steps. Plus the quantitative and qualitative split, and the trick of asking AI directly what it is still unclear about so it tells you where your gaps are
- Don’t put a bullet train on steam engine tracks – or on no tracks at all. Timesheets were built for human speed, and every time AI speeds up a process your charge-out hours fall, which is exactly why you need something documented to measure the before and after against
- Why having it written down is what gives you the freedom to deviate from it – structure is not rigidity, and you can only improve a process you can actually see
