Johnson Controls: Advocating the AI Agenda

Supported by a business with a long history of pushing boundaries, Mario Del Natale, Treasury Director, Global Digital Treasury, Johnson Controls, is open to exploring the potential of AI in treasury. He talks to TMI about his approach to this much-anticipated, and for some, even feared, technology. 

Johnson Controls is a world leader in smart buildings. The 140-year-old business offers the world’s largest portfolio of related technology, software, and services to clients in settings such as healthcare, education, data centres, airports, stadiums, hotels, and manufacturing. It employs more than 100,000 across 150 countries and more than 2,000 locations.

Back in 1885, when Warren Johnson set up the business in Wisconsin that would become the global giant that is today Johnson Controls, innovation was always at the heart of his endeavour. A serial inventor, his pneumatic tower clocks, electric storage batteries, wireless telegraph business, and steam-powered luxury cars and postal service trucks were ahead of their time.

The drive to push boundaries is so deeply embedded within the firm’s business philosophy that today, even its treasury operations are at the forefront of the imaginative use of technology. And with its own Global Digital Treasury locking it into progress, for Del Natale, one area where much could be gained is in the use of AI.

In doing so, we deliver secure, real-time ML-driven solutions that address the pressing challenges of sustainability, energy efficiency, indoor air quality, and smart, secure buildings. As the number of sensors we use rises, so the amount of data in buildings continues to grow exponentially. Resultingly, the value of processing data and applying intelligence at the ‘edge’ – essentially bringing computing closer to the data source rather than sending it on a slower journey to the cloud for processing – becomes increasingly important.

MDN: A couple of years ago, we started to refer to the blending of ideas such as integration, automation, and process streamlining with the core practices of coding and algorithm creation, as ‘digital’. Our own digital journey started back in 2009 with the implementation of a global payment factory. This is used not only by treasury but also by the entire organisation for any payments we process, and for any bank statements, including the intradays, that we receive.

As part of our phased approach to treasury’s digitalisation, we also implemented a TMS, a trade finance application, a bank account management application, and an in-house highly sophisticated analytical application using Power BI. All of these applications are interlinked and synchronised, and all Johnson Controls International (JCI) ERPs – currently more than 100 – are connected digitally to our payment factory. The payment factory itself is linked to more than 250 bank branches via SWIFT, using standard ISO 20022 messages.

We always had the idea to streamline our processes by automating and integrating data coming from multiple sources. To achieve that, we paid close attention to our static data set-up, establishing strict policies around these. This has been fundamental to the establishing of our future digital landscape.

Our ‘ultimate’ digitalisation solution is our Microsoft Power BI solution. Among many other possibilities, this provides real-time liquidity positions for around 3,000 bank accounts. It does this by deploying logic-based technical solutions that are hard to find or acquire from legacy software providers or even fintechs.

As a treasury, we are fully agnostic from IT and from our software providers when it comes to the coding and building of our dashboards. Our solution does not require any new set-up or coding if a new bank or bank account is added to our payment factory.

We are also fully agnostic from a static data point of view. Our Power BI app has a top-to-bottom approach meaning that from senior management to treasury analyst level, all can use the solution. It also ensures that aggregated data is available using a multi-layer ‘slicer’ solution that enables users to drill down to the most granular levels, such as bank statement transactions or payment details processed within our payment factory. It’s important to know that we’re using only the standard ‘out-of-the-box’ tools and features of our applications and Power BI, which means we can easily access all of our analytical tools from any device.

MDN: Some parts of our treasury are already using AI technologies. Our front office, for instance, uses applications such as Bloomberg or FXall dealing that have AI built into them. For use elsewhere, the question has to be which AI technologies to use, as there is an increasing number – maybe too many – of fintech solutions that are AI-powered. From my perspective, as a first step, although it’s already present, I’m looking at embedding AI deeper into our Power BI app to enhance our data analysis and forecasting.

The issue for me has never been whether or not to use AI, but rather one of understanding the models behind the technology. It’s important that if we use AI, we pay attention to data confidentiality, to the risks of biased outcomes, and of course to the retention of human knowledge so we do not come to rely blindly on a system.

Today, without AI, reaching a robust interpretation of our data requires its strict and detailed categorisation, achieved during static data set-up and then through dynamically or automatically tagging during data refreshes. I’m not sure if AI could ever bypass that data quality process, but it will speed up data refreshes and enable increased automation of data interpretation.

MDN: Our current use of AI is within our analytical tool. To get the most from it, we’ve found that this requires not only advanced technical knowledge across multiple areas, mainly around coding, but also a strong understanding of logic within staggered or sequential methodologies. But then I’ve found that to be the case too with ChatGPT. For it to be useful, you must go deeper and be more detailed with your questions and interactions to really achieve what you are looking for. If you don’t do that, it rarely provides the right answers, even if it does at least set you on the right path. Personally, I would say that AI is becoming an additional but ‘normal’ tool for me to use in treasury.

MDN: There’s no easy response to that because it depends on what type of AI we want to use. At this stage, I’m not sure that a solution exists that can be used ‘as is’. We will be introducing Microsoft Copilot 365, but it will not be a decision-making tool for us. Microsoft says Copilot 365 will bring “powerful new generative AI capabilities to apps millions of people use every day like Microsoft Word, Excel, PowerPoint, Outlook, Microsoft Teams and more”.

Independently, treasury, as with any other department, won’t be able to maximise the usage of AI if the source data is not cleaned and ready, as I’ve already stated. However, in our case, we are ready. How far will I go with AI? I’ll offer a consultant’s view: it depends on the needs, on the readiness of a potential solution, and on the acceptance by the company and users. We first have to understand the full spectrum of AI usage for treasury before planning our next steps.

Let’s say that a solution already exists, and we still have numerous reports to check and validate manually, and we are not using RPAs. I would think that an AI solution could be used to check the data behind our daily reports, highlighting errors, and even proposing next steps to correct the data, or perhaps sending automated emails or storing reports on a SharePoint, with charts and clear explanations, for analysts and others to use.

MDN: The adoption of AI necessarily brings a change in the paradigm of treasury’s profile. For the past two or three years, I have been arguing that treasury, or any other professional function, must develop a hybrid profile. It should combine deep business knowledge alongside the application and technical knowledge. It’s a role commonly referred to today as ‘data scientist’. I constantly re-enforce the need for such a blended profile and believe that the arrival of AI will make this transformation essential. 

Treasury must have all of these competences internally because we cannot fully rely on our IT functions any more, even if IT must remain a key partner. Within IT, there are often many different profiles because it is simply not possible to know every aspect. Treasury is a niche, and treasury applications are a niche within a niche. It follows that IT cannot be expected to specialise, so we help ourselves by bringing our own expertise.  

The emergence of AI, from my point of view, means that the ‘new’ profile and its required way of thinking and working is essential across every level of responsibility of treasury, even at the top management level. Bankers and software providers are often proposing new technologies. As treasurers, we must understand what they are offering to know whether or not their solutions fit the needs of our organisations. During an RFP process, for instance, we are used to challenging their marketing statements. When new technologies such as AI are tabled by them, we must be able to pay greater attention, and demonstrate a deeper understanding of the products if we are to properly assess them and see how best to use them within our organisations.

MDN: With ‘AI fintechs’ flourishing everywhere today, I believe that the best treasury solution currently takes a modular approach, where you choose the solutions that fit your needs and which can be seamlessly integrated with your other treasury tools. If you consider that ‘artificial intelligence’ as an output is the result of the interpretation of data based on models developed by coders using modern techniques and algorithms, tomorrow we will all have to pay much closer attention to AI and be able to choose the right solution. This is why it’s so important to develop the new professional profile that I was referring to earlier. 

Today, JCI is almost totally reliant on its treasury applications landscape because these applications are global in scope. It means that when we do introduce AI, we will have to go through a phase of deep regression testing to understand all of the outcomes and to interpret the resulting AI data, avoiding issues such as bias.

We are not yet there, and we’re not even sure how to get there. By definition, an AI solution should be able to self-learn to decrease our reliance on it. But then mitigation controls will become even more critical than they are today. These controls mean we will have to have informed people who can interpret the data, and who have a holistic view across both treasury and finance knowledge and processes. And that will mean treasurers becoming like Swiss Army knives, able to discuss and understand concepts outside of our comfort zone to avoid any potential issues such as results bias, over-reliance on the technology, or even the dumbing-down of treasury itself.

MDN: It may be optional today but it’s eventual adoption is inevitable. I cannot imagine legacy software providers and fintechs not marketing nor proposing new AI solutions. We have modern, flexible, and scalable solutions without AI today, but AI will be the next step. However, that step will not simply be like implementing new functionality or a new ‘well-known’ application. It will require strategic decisions that cannot be missed nor taken without thinking through all the pros and cons of its adoption.

At JCI, we are often recognised by our peers, bankers, and software providers as being an advanced treasury in terms of our applications. And yet the coming AI era for treasury will be completely new even to us, and we are currently blind as to which way we go.