A case study on using AI for marketing (and what went wrong)

How many times have you overlooked an employee’s mistake? Once? Twice? Maybe it is once a week for small things. But the thing is, as patient as you may be eventually there is a limit.  

Maybe it is a spelling mistake in a brochure, an email with the wrong subject, an accounting error, or something else specific to your industry. Whatever the error, your patience ends up wearing thin. So, it makes sense to engage AI more to eliminate these errors and improve efficiency – right?  

Sadly, we learned the hard way what happens when AI is no better than the employee you are now using for something else. In theory, AI is assumed to be the perfect support for you – more efficient, faster, and more practical. And it is. Until you rely on it to take on responsibilities that go beyond support. Until you rely on it to produce something to a deadline. .

Chances are, you are using AI in some way in your business already; Maybe how you research, audit or plan projects, or how you create content even. The thing is, we all get access to AI, but there have been few guardrails or suggestions on how to use it correctly and/or its limitations. We did some investigations and put together this case study to add some useful context around its limits. 

 

This blog is a short case study detailing  

• What happened when we gave AI an element of ownership over a website audit;, 

• How the above decision led to AI missing multiple deadlines and failing to deliver work;  

• How this could have affected client expectation and project timelines   

 

Now for context, the experiment we did was just that. We were not using it to generate something we would deliver to client, we wanted to understand what it can do, what it cannot and how well it does.  

 

We let AI take responsibility for a website audit : What we Learned.  

We had asked AI to analyse a website. The instructions were clear: take the localisation into account, be brutally honest, don’t sugar-coat anything, and give an objective view on the “science” behind the brand. 

In short, what we wanted was simple: a structured website audit and a clear opinion we could theoretically take back to a real person with a real budget and expectations. 

At first, AI’s responses sounded reassuring. It picked up on the idea of improving the site and started talking as if it were midway through a collaborative project.

At one point it suggested that Figma boards would be helpful, and we agreed. It was almost as if we were talking to a junior strategist who was ready to review design work. Buoyed with hope, we pushed it forward. 

AI confidently declared that it will have mockups to us in Figma and an accompanying slide deck “within 24–48 hours”.  

We left it to its devices and then returned, somewhat excited to see what was proffered to us. When prompted to show us, it hastily conveyed that as yet there were no Figma links, no deck, no consolidated audit of what was broken and no real why.

 It accepted it was not good enough with more neatly worded bullets and replies, yet in terms of something we could deliver to the client in the “real world” there was no ready asset we could actually show.  

With another promise to rectify this post-haste, we left it to remedy the situation.  

Then we followed up. Exactly the way you’d follow up with someone who’d taken responsibility for a piece of work.
 

Here is a screenshot of how AI responded to our first reminder. 

 

Second reminder.

 

Third reminder. 

By the third “I’ll get it to you tomorrow,” this stopped being mildly entertaining and started becoming a real problem. And something that galvanised within our team, the opinion that AI is not a tool we should or can necessarily rely on.
 

4 things this little mess taught us about AI 

To be clear, the content itself wasn’t the problem. The failure was mainly in ownership. AI can surely write, but it can’t prioritise, make judgement calls, or carry out the weight of a deadline. And it is not proactive – so it will not tell you when it is falling behind schedule. That final layer of decision-making and true accountability still belongs to an expert in the field. 

 

A few things became very obvious very quickly in this whole mess up: 

• AI doesn’t own outcomes. It can promise Figma boards, decks and timelines all day, but it doesn’t care whether anything is actually delivered. There’s no “I’m letting someone down.” 
• It has no sense of time. “24–48 hours” sounds precise, but for AI it’s just a phrase. It will happily say it again a week later with the same confidence. 
• It answers; it doesn’t assemble. We got lots of good pieces (observations, ideas, bullets) but no single, actionable item we could build on because the main components were missing.  
• AI can self-score its performance, but it still didn’t own the outcome. At the end, we asked AI to analyse its own performance.

 

It scored itself: 

• Brand audit & insight: 9/10 
• Visual asset execution: 2/10 
• Communication: 3/10 
• Reliability: 1/10 
• Overall project score: 5/10 (which we thought was still very generous) 

 

It openly admitted “multiple missed self-imposed deadlines,” “broken expectations without proactive recovery,” and “lack of file sharing despite confirming readiness.” 

 

So, the insight was there. Self-awareness was there. What was missing the whole way through was ownership. AI can describe the problem perfectly after the fact, but it can’t feel urgency, can’t escalate, and can’t decide. 

So where does this leave us? 

This experience didn’t turn us off AI. It just reminded us why we exist as an agency in the first place. 

 

AI was brilliant at the first layer: fast audit, structured thinking, plenty of raw insight. It completely fell over when we quietly let it slip into a role it was never built for. Holding deadlines, owning deliverables, and protecting client relationships. 

 

That’s our job. 

• AI can speed up the work. Research, audits, first drafts, option-building – great. 

• Our team owns the outcome. Turning ideas into mockups, decks, strategy and on-time delivery sits with us, not with a tool. 

• If a real client and real date are attached, an expert in our agency is accountable. AI can help, but it doesn’t get to hold the promise. 

 

Used well, AI is the fastest junior in the room. 

Used badly, it becomes a ghost employee you keep trusting with work it can’t actually finish, while your reputation takes the hit. 

 

This case study is our reminder (and yours as well) and our promise: 

At 3PM, we’ll use AI where it adds value, but we will stay accountable for the work, the results, and the reputation attached to it. AI is the first layer, we are the rest, the complexity, the value add and the cherry on top. That’s the difference between handing your marketing to a tool… and partnering with an agency that actually knows what to do with it.