AI & Data

What is everyone missing about AI in Enterprise Business?

Reuben Boughton, Marketing Director

16 May 20248 minute read

I want you to think about the things that you excel at.

It might be baking, golf, math, writing, reciting the John Barnes Anfield Rap, but for a moment just imagine that one thing that you excel at.

Now think about, what makes you so good at that skill.

Is it a god given gift? Is it that you understand the technicalities better than anyone else? Or is it the purposeful practice you put in from an early age to unlock your potential at the delivery and execution?

In most cases, unless you are Mozart, it’s the latter. Matthew Syed wrote about this in his brilliant book, Bounce. If you want to excel, 10,000 hours of training will take you much further than your natural abilities. It’s hard to argue against the practice, practice, practice argument.

Mozart using a macbook laptop in an old cafe

“The harder I practise, the luckier I get.” - Gary Player

So how and why should we relate the bounce theory to AI in Enterprise business?

One of the leading authorities in AI right now is Ethan Mollick. In his most recent book “Co-intelligence” he sets out four principles that will improve every users use of AI.

One of his founding principles is 'Invite AI into everything that you do'. Mollick believes at a bare minimum every person in your business should be given an OKR (Objectives and Key Results) of spending 10 hours utilising AI in your everyday work tasks and report back on the findings. This isn’t anywhere near the 10,000 hours that Syed discusses in Bounce, but the theory remains the same. The more I practice the luckier I get.

What are Mollick’s Four Principles of AI use?

1: invite AI to everything that you do

AI is a tool, it can help us maximise the volume of ideas we create & the velocity of experimentation around those ideas. In essence, this is why medicine and education will be the industries that see the quickest returns. Immerse yourself in it and utilise it, test it, and iterate with it. The more I practice the luckier I get.

Immerse yourself in it and utilise it, test it, iterate with it. The more I practice the luckier I get.

2: Be the human in the loop - pick the things that think you are important and double down utilising AI.

It's a human trait to dismiss new technology. But every new version of these LLMs gets better and we should utilise them for everything we do, and make it a habit. And frame every prompt in a specific way, testing the outcomes as you go.

Instead of “Give me ten ideas to help chefs save time and money in the kitchen” try: “Give me ten ideas to help chefs save time and money in a 10-metre x 10-metre kitchen, using only electrical appliances making soups and jacket potatoes.”

Iterate, iterate, iterate, iterate.

3: Treat AI as a person.

This lowers the fear factor. Give the AI as much context as you can. Tell it who it is i.e. “you are a French chef looking to create gourmet food at cut prices and you want to deliver this in a down-to-earth pub venue….”

4: Presume this is as bad as AI is ever going to get.

The larger the model the better the model. As an example of this, @Bloomberg spent 10 million dollars on creating their own LLM and it's excellent at many things, but ChatGPT 4 beats it on every characteristic. ChatGPT 4 also beats every average doctor's consultation given and fed the correct data. So if your answer is simply just to create your own LLM it’s not the holistic solution. You’ll need a hybrid approach using multiple LLMs, the best people you have, and your own data sets and then you are just getting started.

So, where the bloody hell are we and how does business see value from AI?

Most enterprises are creating use cases in the back office, some are even creating models that help sales teams make better decisions.

But most enterprises have charged middle management task forces to scope out the AI value for the business, this will likely lead to some large-scale RFP that will be sent to the usual players, asking for the answers that nobody truly knows yet.

A higher value will come once we get to the front office, using multiple data sets, images and behaviours to extrapolate unique insights into our customers but at a pace. But in the meantime, the back-office use will supply us with great use cases and invaluable learnings. Nobody truly knows how long it will take for a hierarchy of service providers to appear. The major players continue to build products at a rapid rate and scale, this is land-grab territory and can often lead to very fast failure.

Fast failure how?

A prime example of this is Tyler Perry’s recent venture. Perry was in the process of adding 12 sound stages to his studio but on seeing OpenAI’s video generator Sora (Launched in February 2024) halted all expansion indefinitely.

Perry was quoted by the Hollywood Reporter:

“All of that is currently and indefinitely on hold because of Sora and what I’m seeing,” Perry said “I had gotten word over the last year or so that this was coming, but I had no idea until I saw recently the demonstrations of what it’s able to do. It’s shocking to me.” - Tyler Perry

Photo credit: tylerperry.com

The AI tool was launched on 15 February – its ability to produce realistic footage a minute long from simple text prompts. This all makes a lot of sense regarding the recent writer strikes, artists and creatives want to protect their IP, but more than that, their voices, performances, and future-proof their earnings.

Who has the upper hand in the AI innovation game & how big is the change to come?

It’s certainly the 1920s & 1840s again regarding a technological revolution, but this time it’s on performance enhancers. CEOs will seriously need to rethink corporate organisations and processes.

Regulations will almost certainly lag, meaning the smaller independent companies can experiment and test without being weighed down by governance, internal politics and PR scaremongering.

Bias and hallucination are still an issue but much less so with the new models. Mollick again in his recent book continually sights the difference between ChatGPT 3.5 and 4 as vast. We must use the Frontier models, GPT-4 (OpenAI), Claude 3 (Anthropic), and Gemini Ultra (Google).

"Friends don't let other friends use ChatGPT 3.5" says Mollick.

Everyone is making mistakes including Google

Earlier this year it was reported that Gemini (Google's AI Chatbot) had gone off the rails regarding image creation.

Images showing people of colour in German military uniforms from World War II that were created with Google's Gemini chatbot have amplified concerns that artificial intelligence could add to the internet's already enormous pools of misinformation as the technology struggles with issues around race. The New York Times wrote in a story under the headline Google Chatbots AI Images Put People of Colour in Nazi-Era Uniforms. Not a great look.

Let us not forget, however, Rome was not built in a day, and when it comes to winning in the long term these vendors understand iteration is key.

So should we sit in and follow or get into the reeds regardless of a nervy marketplace and incoming regulations?

CEOs, CIOs, and department heads, get all of your people regardless of function as part of their OKRs and professional development to use Chat GPT-4 (Or a competing LLM) in their role and write learnings and efficiencies and report back. Get accustomed to writing and iterating prompts. This is not going away. This is an efficiency tool.

Create different customer personas in your prompts to market test products and product ideas. Question your current methodologies. Many organisations are dropping the agile methodology because why do things that way when you can have AI review screen designs from the perspective of different personas and automatically generate feedback while automating the documentation phase. Think about how you can reward AI use for the greater good of your enterprise. BMW recently offered staff a financial incentive. Money saved by AI utilisation was a percentage given back to the larger team.

Expert prompting will be the biggest short-term gap. Coders are proven to be the worst prompters. Treat Chat GPT as a person, find your best prompters in the business and pass on the learnings.

When you next write a job spec, spend time researching what part of that spec can be automated via AI and then reframe the spec. Be clear about the impact on roles that AI can have, why not also ask candidates about how they see automation playing a role in what they do. Don't get carried away by privacy. Tick the privacy boxes, AI LLM's should be seen as a product and as an efficiency multiplier not as a super spy. It’s only the same as utilising Dropbox. This is how we should frame AI in the business context. AI is lazier in December, and we have no idea why, this is new territory and much is unexplainable regarding the role AI has to play in our businesses and society at large. So as always, beware of the know it all car salesmen, if Google are dropping the ball, our car salesmen will be dropping all sorts.

Finally always fact-check and then check again, AI can be manipulated via prompt injections and is vulnerable to expert use.

Enjoy getting those hands and feet dirty kids, this is not going away, and those who sit it out and watch on the sidelines, have lost the war already. As always any feedback is welcome.

Thanks, Reuben.