Before AI Wins, Modernize the Mess

Play Audio Version

I recently wrote here on this blog about the possibility of a new ‘AI Winter’ that reduces the investment and interest in AI systems. There is a genuine fear that most companies have yet to find a use for AI that generates value for their business.

MIT even published research suggesting that 95% of companies running pilots and tests still can’t see any business value from AI.

But I would caution jumping on the anti-AI bandwagon. It is a transformational technology that will kick-start entire new businesses and business models. It’s too simplistic to judge this based on questions such as ‘does this generate more revenue for our business immediately?’

The real problem for most companies today is that they were designed before AI became so easily available and easy to use. They are not designed for AI.

AI requires thought, planning, and infrastructure. To really gain value from a language model you need to train it in all the various areas that it may need to engage with. For example, if you are building a chatbot that can answer customer questions then it needs to be trained on every detail of your products and services and then constantly refreshed and updated with changing details – such as store opening hours or special offers.

This means that you may need to get your data in order before ever embarking on the AI project or pilot.

Then there is the issue of legacy systems – often unconnected or connected manually through the use of employees cutting and pasting information between applications.

If your CRM is not connected to your logistics and warehousing system and there is no simple way to query delivery status then it becomes very difficult to apply tools such as agentic AI.

Agentic offers the opportunity to allow bots to handle processes on behalf of employees or customers. Imagine a situation where a customer says that I want to buy a return flight from Paris to Sydney in July, preferably with only one stop, and for less than $1,000.

An agentic agent can do the research, find the options, and keep monitoring the prices constantly. You can even give the bot permission to make a purchase if all the conditions are satisfied.

But think about the APIs and data flows needed to make this work. You can’t rely on an employee manually punching in customer numbers or preferences in this automated environment.

This is the real problem for most companies. They have legacy systems that may go back decades and their data is not structured in a way that makes it easy to train AI.

Last year I wrote an article for Planet Mainframe magazine celebrating the 65th birthday of the COBOL programming language – a language I did actually use earlier in my career. If you look at just the US alone then the federal government has systems with over 220 billion lines of COBOL. 95% of card reading systems at ATMs will use COBOL. 43% of the core American banking system is built with COBOL.

Just look at statistics like this and the problem becomes obvious. AI works. It can be a powerful tool to create new opportunities, create new insight, and create productivity gains – but it doesn’t exist in a vacuum. It has to connect into the rest of your business so data can be exchanged and the AI can be trained.

Most companies still have huge data silos – different departments have different systems that are not interconnected. They also create and store data locally in each team. Batch processes are common in many older systems – they are designed to process information overnight, not in real-time. Data is also unstructured and in many different formats and systems.

If your core applications were installed in the 90s and just maintained since then, how do you think an incredibly new AI system will connect into this?

What you really need is a data and application modernization strategy – so the core infrastructure of your business can catch up and get ready to use AI.

This is not easy. It requires a focus on processes and systems, but also people too. Your internal team has to become familiar with new procedures. Budget will be required for change that may see very few immediate results – it is preparing the infrastructure for future change. You also need to plan for drift and decay – this is not a one-off change, you need a plan for constant updates in future.

The headlines about an AI winter and all the AI project failures fit a certain media narrative. Many people love to see new technology failing, but the failure here is not with the AI. New systems require careful thought and a planned environment.

Running small pilots on SaaS systems is fine and can prove the concept, but when you need to scale up those pilots to the enterprise level then your new AI systems have to be planned for and managed – usually this means modernizing what you already have before moving forward.

IBA Group has an experienced practice focused on application modernization, data management, data analytics, and AI. Check the website here for case studies and examples of projects that have been designed and delivered in addition to ideas, insights, and project suggestions.

Follow IBA Group on LinkedIn for regular updates and comment. For more information on technology strategy and how tech connects to real business solutions please click here