“The moment you should use AI is the moment your first thought is ugh“
People of AI Search – Florian Fischer, Chief AI Officer at Slopelift
Welcome to People of AI Search, our new video series where we sit down with the people actually rebuilding search, marketing and agency work around AI – not the people posting about it.
First up: Florian Fischer, Chief AI Officer at Slopelift. Flo has been in the industry since 2008, started in PPC, and never really left it. Today he’s responsible for putting AI into the hands of roughly 80 people – deciding which tools are allowed, which data can touch them, and how to keep an entire agency moving in the same direction while the ground shifts underneath it.
We talked about turning a PPC department into an “AI-first” department (a rollout that started the week we spoke), why the SEO team adopted AI fastest, what the EU AI Act actually means for an agency’s tech stack, and the risk nobody’s talking about: AI burnout.
“Ugh” is the signal
You’ve been in PPC since 2008. What gets you excited about AI changing it?
I try to be a very efficient guy. What I love is getting the most out of your time, so you can be out of office early.
There’s a rule: the moment you should use AI is the moment you have a task where the first thing in your head is “ugh.” That’s exactly the task where you should put an agent on it, or figure out a way to make it more efficient.
What does that look like day to day?
Redundant tasks. Pulling negative keywords out of Google Ads campaigns. Checking performance at CPC level. All these things that require you to tinker in a very detailed way and take an unnecessary stretch of your time.
Making PPC an AI-first department
How is your PPC team actually using AI right now?
We’re in a complete transformation phase. We started by getting AI into the workflows – tests with audits, analyzing accounts. But starting this week, we’re turning the PPC department into an AI-first department.
You must use AI for your weekly account analysis. You have to use it for audits. You have to use it for performance reports. Team leads have to use it to summarize findings.
And what do you get back?
Time for strategic things. More conversations with the client – about what they actually want us to do, not the hygiene work on the account. The things that drive the business forward. That time is much better spent than getting a report right with the correct data.
The job of a Chief AI Officer
What’s the role, really?
It sounds nice – you have all the strings of AI in your hands. You’re the one who says “this tool is all right to use.” Which means governance and compliance come into play. With the EU AI Act coming in August, there’s a lot to think about. Basically I can use any tool – but what data can I use? Am I free to use this or that?
That’s one half: harnessing the AI. The other half is getting people involved without fear – without the fear that they’ll be substituted, that they’ll lose their job. It won’t go away. You will have to work with it. And this is the only harsh thing I’ll say: everyone who isn’t capable of using it, in our industry above all, will be left behind.
You dropped a buzzword – “harness engineering.” What is it?
It’s the evolution of prompt engineering. Because who’s doing prompt engineering today? Who’s writing your prompts?
Either me or the AI.
Exactly. That’s the point. The first thing I do in a new project with Claude is ask it to write me a CLAUDE.md for this project. Then you tinker with it. Harness engineering is reworking the first prompt and the backbone of the project – giving it the guidelines and the rails. What you can do, what you’re allowed to do, what you won’t do. You’re not allowed to search the web. Don’t use this data, only use this.
It’s getting Claude into a suit of armor.
Adoption: the slump after the surge
What’s the biggest challenge?
It’s not the enabling. The tricky part is keeping people working with it. You get a surge in usage when you announce something new – and then normally there’s a slump.
You have colleagues using it extensively, and colleagues going “bah, okay.” You need to get everyone to a certain level. It’s like a client conversation: we always try to get to eye level with the client, so I don’t have to explain every time what a Performance Max campaign does. Now we have to do the same thing internally with AI. Everyone talking on the same level – because then you can do larger rollouts. With nearly 80 people, that’s not easy.
How do you measure success?
The soft measurement is usage – how many prompts are there. But the point of truth is how many questions you get because something doesn’t work. That’s far more interesting. When everybody’s happy, nobody says anything.
When someone comes to you because they broke the system, that’s the best thing that can happen. You either know there’s something to fix – or you embrace it, because it’s something you hadn’t thought of yet.
Which team adopted AI fastest?
The AI team, obviously. [laughs] No – SEO. It’s simply closer to LLMs, to what we now understand as AI. It’s closer to SEO than it is to performance marketing. It’s in the DNA. It was only natural.
Governance in the EU
Do you hook Claude up to client data?
Because of the regulations in place – or coming – you have to set up and document the tech stack in detail. You evaluate every tool on its risk. If it’s high-risk: which data is used? All of it documented, so we don’t use data we’re not supposed to.
Claude has no servers in the EU, only in the US, which is a data-protection problem. So you need specific workflows. We use BigQuery, for example, to automatically anonymize some of our data. Financial data is completely out of context. There’s an anti-to-do list alongside the governance documentation.
There are predefined connectors people are allowed to use – your own calendar, your own Gmail. Some projects connect directly to Google Ads and Google Analytics accounts. But you have to build hurdles between the raw data and what you feed in.
A people’s business
How would clients describe Slopelift differently now?
I hope they wouldn’t. One of our key assets has always been eye-to-eye consultation and deep, long-term relationships. Running an agency is a people’s business. Maintaining clients is a people’s business.
If anything, I think it intensifies – positively – when people have more time to nourish those relationships.
The risk: AI burnout
Last one. What’s the biggest risk you see right now?
For high achievers – or people who want to be high achievers – AI can be a dangerous thing. You can do this, and this on the other side, and have another screen open running something else. This project isn’t done, so I start another one. It’s very easy to get not just AI fatigue, but a new kind of burnout.
Always keep it in scope. Always try to finish the project. Because the one thing AI always wants is: “Is there anything else I can do for you?” Sometimes you need to say – okay, now it’s finished.
And your last prompt in Claude?
“Make it more Viennese and not so German.”
People of AI Search is an OtterlyAI x Slopelift series. Stay tuned for the next ones.
