Filip Tichý | 3.4.2024 |
The authors of this article, Filip Tichý (Partner at Grant Thornton Slovakia) and Jakub Chudík (Co-Founder at Assetario), take you through the world of artificial intelligence in the AI Breakfast series. This article was written without the use of AI.
Many companies are considering how to suitably implement artificial intelligence ("AI") into their business. The implementation of any modern technology is a challenging process accompanied also by risk. This is especially true with a technology that will paradigmatically change our everyday life. The fundamental pillars for a successful AI implementation in any form are: strategy, data, infrastructure and skills & culture. Each of these pillars would easily populate a separate book, therefore to simplify and generalize, we would characterize them as follows:
The first AI pillar to start is the strategy. Every company should evaluate and analyse the impact of AI on its industry, its business model, or its operations. Based on these considerations, form your goal and the path to achieve it, decide which steps to take (and which not to take). Which areas and activities should be automated and how (the question of this article -buy or build), what is the next step, what are the associated risks and so on.
The second AI pillar is data. Data is the main “fuel" for artificial intelligence, the more (and the better the quality) the better. The company should already have identified all possible data sources – internal and external, recorded and unrecorded, already available and also potential data sources.
The third AI pillar is infrastructure. The way in which the company’s data is consolidated, what are the data flows, how is the IT infrastructure set up, the level of digitalization, ERP system and other applications. However, when it comes to this AI pillar, it is not only about the infrastructure in the sense of IT, but also about the company itself and the way it is “built" and how it is managed. The implementation of AI is, to a big extent, an experimental and iterative process. The first implementation process will, with a 100% certainty, not represent the final form of the optimal implementation. It is therefore essential for the company to be ready (and, in operating terms, capable) for rapid changes and iterations.
The fourth and final AI pillar are the skills and culture of the people in the company. Artificial intelligence is often very superficially associated with replacing humans. But it is the people working for companies, who have to think creatively about implementation, utilisation and collaboration with this revolutionary new technology. This is the most abstract and the "softest" AI pillar. It is necessary to develop so called "fusion skills" - the ability to work effectively in an AI-human interaction environment.
The hallmark of successful AI implementation preparation is that each one of these pillars is properly captured and not neglected. These pillars are not independent, they interact and influence each other. AI strategy defines access to data and infrastructure, capabilities and culture also influence strategy, etc.
One of the AI strategy’s options is the technical choice of AI implementation. Once the areas and activities to automate, available data, what can be acquired internally and what externally has been identified, the infrastructure has been set up and an enthusiastic AI team has been launched, the question of practical implementation follows. Many companies are already in the position of knowing which area they want to automate and are faced with the question of how to do it technically.
There are two main approaches, how the company can implement AI: build their own AI model ("build"), or buy an existing "off the shelf" AI application or tool ("buy"). How to decide? The main differences are obvious: building your own AI model is more of an option for large organisations with huge amounts of data and transactions. Smaller companies without large amounts of proprietary data will in turn search the market for the most suitable AI app.
Typically, the "build" approach of building a custom AI model will be taken up by companies that already have an existing data analytics department, which are already working systematically with the company's data and uses simpler expert systems ("plain" or "non-intelligent" algorithms). When such a company feels that it has reached the top in increasing its performance or efficiency by leveraging and using data, it will start building its own AI models and thereby creating an "AI superstructure" on top of existing data analytics. This is the most clear case of the "build" AI implementation being the most appropriate.
On the other hand, small (or very small) companies may ask a different question. Is it even worth it for my mini-company to think about implementing AI? The answer is a solid “yes”. First and foremost, AI will change both the environment and the industry in which a company operates, and even a small organisation will need to perceive and respond to these changes. AI as a technology will soon be available to everyone even the individual users, for example Microsoft Copilot or a multitude of other small AI apps. It is quite likely that Copilot will soon be like MS Excel - everyone has it on their computer, it has huge potential and with expertise it can be used as a platform to build any functional tool, with most users only using only a minimum of its potential.
The decision between the "build" and "buy" strategy could be simplified through the relationship between the maturity of two AI pillars (data and infrastructure), and the size of the company. If the company has a high-level data analytic function, an advanced IT and data infrastructure, and an agile way of managing and implementing projects, it should certainly think primarily about the "build" option. To add, if it is a large multinational corporation, the possibility to "acquire", i.e. the acquisition of a company that deals with AI in the industry, should also be considered. If the company is mature enough in these two AI pillars, but is relatively small, it is worth considering whether it has the capacity to build its own AI models or whether it can benefit more from implementing existing AI solutions. In the case of a company not generating, nor recording, a lot of data and/or does not have a robust data and IT infrastructure, it should consider the "buy" option. In the case of a large corporation, the "build" option also comes into consideration, but with an external help.
The "Buy" strategy especially may seem very easy to execute. When a company identifies an area or a process to automate and starts looking at what the market has to offer, they will find that there is a plethora of AI softwares and applications on the market, for the exact purpose they have chosen. They also don't even appear to be too expensive. But the biggest risk of the "buy" strategy is loosing money. The opportunities to spend money on ineffective or underutilized AI apps are unfortunately endless. It is never a certain that such an investment will be effective for the company, i.e. that the benefits will be higher than costs. If the benefit of the purchased AI application is zero, any cost to purchase and implement it is a net loss. That's why no AI pillar should be neglected and even a small AI projects should be set up from the start for future modifications and iterations. Another mistake is setting the goal of AI implementation solely as a cost saving (e.g. through replacing human labour with automation). Early stages of AI implementation are mostly an investment, rather than cost cutting. The appropriate goal is to IMPROVE performance through a collaborative approach of cooperation between people and technology. Even when it comes to your first "buy vs build" decision, you need to think about the long run. This decision is not the only one or a one-off one, but the beginning of a transformational process for our company and our industry.
