Casla helps B2B companies with complex propositions to elevate their brand and generate demand. Since opening its doors in Sheffield in 2023, the marketing agency has taken an open-minded yet thoughtful approach to AI adoption – identifying and experimenting with the ways it can save time and genuinely add value, whilst retaining the integrity of high-quality, ‘human-generated’ content. In doing so, the team admits AI has changed their proposition as an agency, and has enabled them to achieve things they simply couldn’t have done before. In this post, Millie Hurst, Head of Content at Casla, shares an honest and detailed account of where AI fits into the agency’s work, and where it doesn’t have a place.

When ChatGPT first launched in 2022, I felt like it was copying my homework. I’d spent years writing for various online magazines – which made up a small percentage of the copy it was being trained on. But despite the fact I still have to bat off well-meaning questions about whether AI will take my job, I’ve definitely come round to the whole thing.

It’s brilliant at tasks where there is an objectively right or wrong answer or output, and much less good at subjective tasks where there could be many different right answers or outputs. It’ll summarise what a company does in seconds, but if I were to ask it to write me an outbound message for a client, it’d produce something that sounds tonally off.

At Casla, we predict that AI is going to take over a lot of the marketing tasks that people currently do in the next few years – but that it’ll mainly be the boring, repetitive, unglamorous stuff. CRM automation, large-scale research tasks, writing meeting summaries. And that as much, if not more, new tasks will emerge that people are needed for as a result of AI.

We think that while there is a significant short-term competitive advantage to be gained by early adoption of AI, over time, this opportunity will fade away and people will remain the competitive differentiators. Human judgement, subjectivity and skill will remain essential to marketing effectiveness.

Researching industries and honing messaging

We use AI to get up to speed on a client’s industry before introductory meetings and to learn about an aspect of their technical world. I’ll basically have a conversation with Claude, asking specific things about a niche topic. For example, I used it recently when drafting messaging for our client TJ Digital Systems, a manufacturing consultancy based here in Sheffield.

I was trying to make the messaging more ‘on the nose’ so the person receiving it would turn around and say, ‘This is relevant to me, we should have a conversation.’ I asked Claude to explain what shop floor automation might actually look like for a manufacturing company in the UK with 200 employees, made handwritten notes on the answers to help me remember it, and asked it to explain any acronyms and technical terms I didn’t know.

You can then keep digging deeper and ask things like, ‘Can you list the five biggest problems that a manufacturer with 200 employees would have?’ You can break it down by job title and find out about issues a CEO is facing versus what frustrations a Production Director might have – and on it goes.

A research task where I’m building a picture of a prospect’s world in order to make our messaging more powerful – for me, that’s an area where Claude comes into its own.

Content interviews

We do regular content interviews with clients – a half-hour Google Meet call that gives us material for blog posts and ghost-written LinkedIn posts. AI helps us to research the topic – for example, why our client Scintam’s solution is an innovative use of electrical discharge machining (EDM) – and it helps us to write a list of questions we think will create the best content.

For Casla Conversations, our series where we interview people in B2B and write up a blog post based on the conversation, we rely on the automatic transcription and summary from Google Meet. This frees up time and headspace during the call so I can be more present and ask better questions, and it can then be fed into AI to summarise conversation and highlight the main themes covered. This gives us a head start when coming up with an angle for the blog post and working out how to structure it to make it as engaging as possible.

If five years ago, someone had told me I’d never have to manually transcribe an interview again, it would have felt like Christmas. As a journalist I spent years conducting interviews where I’d record the conversation on my phone, while an online voice recorder ran on my laptop as a backup, and I’d spend at least half a day transcribing it and mapping out the key topics.

Google Meet’s transcription might struggle with the acronyms of the B2B world, and the AI mapping might not be perfect, but going back to the old way would be like returning to the horse and cart.

AI makes the infeasible, feasible

While AI often simply takes over tasks that we would have previously done ourselves, the full picture is much more nuanced, complex and, in some ways, surprising.

We find ourselves working in a different way than we would have done ‘pre-AI,’ because certain things that would have been infeasible to do as a person, have become feasible with AI.

To give a specific example, we recently qualified and scored 1,500 contacts for a client, with AI reading every word of each contact’s LinkedIn profile: their profile headline, their about section, their job title, and their responsibilities in their current role. It’s not that this would have been impossible to do ourselves, but it would have probably used up half a month’s budget in human time. Using AI, this cost about $20–25 in tokens and half an hour of thinking to come up with the right prompt and technical set up.

Building the tools

Over the past year or so, we’ve started to build bespoke software for ourselves and for clients, and it’s become a core part of the value we deliver.

For example, for one client we built a basic but functional CRM over a weekend, from scratch, with email and calendar integration. Even six months ago, we wouldn’t have believed that would be possible.

Our LinkedIn InMail / messaging sending app has been a real hit with clients (described by one person as ‘freaking amazing’) – it’s a super simple tool, but turns what was once a fairly onerous and fiddly task into something simple, quick and dare we say it, satisfying. 

What we don’t use it for

We don’t ever use it to write copy from scratch. A piece of content where someone’s simply briefed AI to ‘write something about [topic]’ is easily detectable and sends really poor signals. 

In the past, I’ve finished a blog post or a LinkedIn post, pasted it into Claude and asked it to edit it, and it’s done a startlingly good job. But when I posted the version edited by Claude, I instantly regretted it.

It just feels like an overly sanitised version of our voice, stripped of any personality, because AI smooths out all the crinkles that shape how we express ourselves as individuals. And the whole purpose of our LinkedIn posts is to build awareness and trust. If, as a content person, I’m outsourcing my writing skills to AI, it feels dishonest.

It’s been interesting to see the backlash to AI-generated content – LinkedIn even has a new ‘Seems like AI slop’ button to flag posts that look like AI, which reflects just how much is being produced. When I was discussing this article with my colleague Richard, he summed it up as, ‘If I wanted to know what ChatGPT or Claude thinks, I can just ask it myself. What I want to hear is someone’s own perspective.’

The counterintuitive and perhaps heartwarming upshot is that due to the proliferation of AI slop, original, high-quality, ‘human-generated’ content will become even more valuable.

Where it goes wrong

We haven’t built anything that has outright failed, which is the boring but honest answer. The more subtle danger of AI has been getting excited about what the technology can do and losing sight of the outcome the client actually needs. Richard admits to falling into this trap occasionally, but as a small business working with clients that want results, fortunately there isn’t much room for us to get distracted by technology.

Closing

AI has gone from being helpful peripherally to being a core part of how we get work done; from being a helpful assistant, to working on its own and completing large and complex tasks for us.

Conservatively, we estimate that we’re able to deliver 50–100% more deliverables and outcomes per ‘human hour’ than we were pre-AI. This has enabled us to take on more clients and do considerably more for the ones we already have. Our scopes of work look radically different from early 2025, and the value a client gets per pound spent is much higher.

It’s also changed what we see as our proposition as an agency – what clients buy from us now is a bundle of human time, plus custom-built technology, plus data, plus AI token usage. To the extent that somewhere between a quarter and a half of how we deliver projects now consists of non-human labour.

There’s already been so much change in how we work due to AI, and this is just the beginning.

You can learn more about Casla on their website and hear their latest updates on their LinkedIn page. Follow Millie on LinkedIn for B2B content posts and advice.