Why Your Business is Already Sitting on an AI Goldmine (And Most Companies Don't Realise It)

Over the past few months, there has been an increase in the number of publications and prominent people calling the current version of AI a bubble. Recently, Nvidia share price volitility, which is nothing new, is partly attributed to investors' belief that the AI bubble is about the burst.

There is no doubt that there are a range of companies or startups raising eye-watering sums of money on very flimsy promises. The major issue is that the development of these models and their utility is constantly improving, eating up some of these startups along the way with each new model release.

However, we've seen some flatlining in terms of the exponential improvements the models used to achieve between releases. The image and video models have a way to go still, but my view is that within the text-to-text space, the models are at the point of usefulness where they should be implemented into business workflows. The challenge for businesses is that this isn't always a technical task but rather one that requires a collaborative effort between technical teams and process owners. This was covered in my article about operational leaders being the key to unlock AI's potential.

Where is the usefulness with AI

Within any business, there are processes and ways of working that truly make the business tick. These processes will have imperfections, but those imperfections will simply just work. This makes it challenging to find a ready-made AI solution on the market that can do what you need and that will work with your process imperfections.

Most SaaS requires some adaptation to work and often requires that the business make changes to fit within the tool. This has multiple disadvantages, one being that company-wide buy-in can be affected. The power of AI is in its ability to pick up on the repetitive stuff in any process. This is why we have a handy calculator on our homepage to allow you to see the power of marginal savings across a business.

My philosophy is very much that AI is the enhancement, not the replacement. But with the ability to find newfound productivity, we do also have to ask ourselves about how the future of work looks.

In summary, the usefulness with AI - for me at least - is seen in the margins of the process and not in trying to replace a whole process with a SaaS AI solution that will be under-promising and likely underdeliver.

Quick wins with AI

The likelihood is that you're already sitting on a wealth of AI opportunity. That opportunity is the data that results from the processes that your business has and can bring you immediate gains through fine-tuning a model to replicate that data process and create the same results. The fine-tuning process is cost-effective for a business and has high accuracy in producing results that mirror the inputted data. If this data is sensitive and can't be passed to an online model, you can always train an open-source model on the same data with the same outcomes, which becomes an onprem solution.

The reality is that most businesses are sitting on goldmines of structured data without realising it. Your email communications contain patterns of how you respond to different types of customer inquiries, supplier negotiations, or internal requests. Your CRM data shows how successful deals progress versus ones that stall. Your customer service tickets reveal the language patterns that lead to quick resolutions versus escalations.

Financial records, invoice processing, HR documentation, project management updates, sales reports, marketing campaign responses - all of this data represents processes that someone in your business has mastered through experience. The beauty of fine-tuning is that you can teach an AI model to recognise these same patterns and replicate the decision-making or output generation that your experienced team members would typically handle.

Implementation Roadmap

The mistake most businesses make is trying to boil the ocean with AI. My approach is to start small and prove value before scaling. Begin by identifying one repetitive process that involves data you already have - something that currently takes a skilled person 15-30 minutes but happens multiple times per week.

Map out exactly what inputs go into that process and what the desired output looks like. Gather 6-12 months of historical examples where this process was completed successfully. This becomes your training data. The next step isn't to build anything yet, it's to validate that the process is consistent enough that an AI model could learn it.

Once you've proven the concept works with a small dataset, you can then look at scaling it across similar processes. The key is to resist the urge to automate everything at once. Each successful implementation builds confidence and knowledge for the next one.

Integration Challenges

Here's where things get messy, and it's rarely the AI that's the problem. Most businesses have systems that were never designed to talk to each other, let alone integrate with AI tools. Your CRM doesn't play nicely with your email system, your project management tool is completely separate from your invoicing, and everything requires manual export and import.

The temptation is to rip everything out and start fresh, but that's usually overkill. Instead, focus on the data flows that matter for your specific AI use case. Sometimes a simple daily export from one system and import into another is enough to get started. The perfect integration can come later once you've proven the value.

The bigger challenge is often cultural rather than technical. People have workarounds and processes they've developed over years. Your AI solution needs to work with these human systems, not against them, at least initially.

Training and Upskilling Needs

The skills gap isn't usually where people think it is. You don't need everyone to become prompt engineers or understand machine learning. What you need is for people to understand how to work alongside AI tools effectively.

This means training people to recognise when a task is suitable for AI assistance versus when human judgement is essential. It means helping them understand how to validate AI outputs and when to trust versus question the results. Most importantly, it means helping them see AI as amplifying their expertise rather than replacing it.

The people who are already good at the processes you're looking to enhance with AI are your most valuable assets. They understand the nuances, the edge cases, and the business context that makes the process work. These are the people who should be involved in training and validating your AI implementations.

Future-proofing Considerations

The pace of AI development means that what's cutting-edge today might be standard functionality in 18 months. This is actually good news for businesses because it means you don't need to bet everything on one particular technology or vendor.

Focus on building internal capabilities around understanding your data and processes rather than becoming dependent on any single AI tool. The businesses that will thrive are those that develop a culture of continuous process improvement and data-driven decision making.

Keep your implementations modular and avoid vendor lock-in where possible. The AI tool you use today might not be the best option next year, but the process improvements and data quality work you do now will carry forward regardless of which tools you use.

Most importantly, remember that AI is moving fast, but businesses change slowly. The fundamental processes that make your business successful aren't going to be revolutionised overnight. Focus on gradual improvements that compound over time rather than trying to transform everything at once.