Pablo Yusta
CEO
AI Consortium
In a context where Artificial Intelligence (AI) no longer needs evangelists and even the Government of Spain has decided to invest massively in this technology, there arises both an opportunity and a simultaneous challenge for businesses and public administrations in the country. The opportunity lies in the potential to exponentially improve productivity, while the challenge focuses on how to adopt this technology effectively before the competition capitalizes on its potential.
Foundational AI models, such as GPT, Gemini, and ALIA, are prominent examples of what can be achieved with substantial investments and intensive training. These models have been trained on vast datasets to develop advanced capabilities in analysis and natural language processing, positioning them at the forefront of AI technology. However, transitioning these capabilities from theoretical models to practical applications is not trivial, and integrating these models into the daily operations of companies and governmental entities faces significant obstacles.
The main issue is not the capability of these models but the accessibility and utility of these systems in real business or administrative environments. The standard chat interface, similar to what ChatGPT uses, although useful in general consumer contexts, fails to unlock the full potential of these models for specific business or governmental applications. Organizations need solutions that integrate seamlessly into their existing systems and are capable of adapting to the specific needs of each department or service.
We have all experienced the challenge of implementing an ERP (enterprise resource planning) or other business software in our organizations. We know it is a tedious process, and no matter how hard we try, in the end, many of our business processes have to adapt to the software rather than the software adapting to us.
One of the proposed solutions to transfer the potential of foundational AI models to specific business processes has been through these business software systems, by installing AI "copilots" that "assist" us in interacting with them. While this approach has some logic, since AI models are technologies with outstanding capabilities and business software are solutions designed to help employees work faster and with fewer errors, it suffers from a serious flaw: the capabilities of AI far exceed what can be achieved by confining it within a rigid system like business software.
As Peter Drucker aptly pointed out, "There is nothing so useless as doing efficiently that which should not be done at all." Limiting the unlimited potential of AI to the confines of traditional business software is a task doomed to failure, as these platforms were not conceived to fully leverage the capabilities of this revolutionary technology.
Figure 1. Smart robot with outstanding capabilities tasked with organizing an overflowing file, illustrating the futility of efficiently performing unnecessary tasks.
Source: image generated with Dall-E 3.
Think of AI as your personal oracle or an all-knowing advisor, expert even in the most obscure processes of your business. Your initial encounters with this technology may not have been entirely satisfactory, and you might not fully trust it yet. However, AI has the gift of learning quickly, very quickly. It learns exponentially. Therefore, I suggest you give it a chance.
If you could see AI as an oracle, wouldn’t you use it to thoroughly understand every corner of your business? To evaluate the efficiency levels of processes? Wouldn’t you be interested in improving decision-making with its help?
Foundational AI models know very little about your specific company. They are trained with some publicly available information from the internet, but nothing more. With that limited "dataset", AI will not be able to unlock its potential within your organization. It might make some generic tasks more efficient or integrate with your business software, but certainly, it will not be able to unleash its true transformative power.
For this reason, it is necessary to train the foundational model with relevant data from your company. It is important that this "training" be carried out by an expert you trust, as there are multiple techniques and the outcome can vary drastically. If you have tried to train an AI model with your data before and the results were unsatisfactory, it is highly likely that the training was not done properly.
Once you have a foundational model trained and connected with your data, you will be able to “converse” with your company. How is the marketing campaign for product X going? What is the absenteeism rate? Could you project the business figures until the end of the year? The possibilities are endless.
As a final piece of advice, I recommend having a long-term vision and patience. AI will surprise you, but needs time to learn. It is and will be important to invest in it.
Share the power of AI with your employees. It is the most advisable way to leverage the potential increase in performance. Having multiple human minds using AI daily will bring continuous positive synergies to your company.
The key question is: how do I do it? Some might say it should be done through training courses or similar. But in my opinion, the best way to learn how to use AI is… by using it! For this reason, I suggest you provide your employees with direct access to that AI model customized for your company.
But, will they know how to use it? Will they know what questions to ask? Will they know how to get the most out of it? Certainly, if we do not guide them, it will be difficult. The optimal way for your employees to take advantage of AI is to customize the use cases to their specific roles and job positions. For example: a worker in the marketing department will have Generative AI tools related to copywriting, advertising, social media content, etc.; a worker in the compliance department will have tools to draft contracts, interpret laws, validate operations, etc.; a worker in human resources will have tools to draft job offers, review resumes, organize team-building activities, etc.
It is very important to commit to the employees that their generative AI tools are fully customized to their specific use cases. Generative AI has exceptional and outstanding capabilities. Therefore, it makes no sense not to take full advantage of them and opt not to customize AI to the highest level of detail. In this way, even two workers within the same department will have their AI tools customized according to their unique needs.
And the million-dollar question: is this very costly? The answer is no. If the AI platform is well-designed, creating a customized tool should be accessible for anyone, even those without programming knowledge. In less than 5 minutes, a new AI tool should be ready to be used by employees.
The data of your company should always remain under your strict control. It is a common practice for foundational AI models to "learn" from the data shared with them to become increasingly intelligent. In fact, if we read the fine print of the free versions of ChatGPT and similar tools, we will find the message that your data will be used to train the model.
When you go to implement any solution that involves an AI model, whatever it may be, ask to be explained and guaranteed what will happen with your company's data. It is essential that the data does not go outside without your supervision and consent.
It is also crucial to have control of the data within the company. For example, if you have trained an AI model with your company's salary database, employees in the human resources department should be able to interact with this information, but the rest of the workers should not.
Ensuring all this is controlled is not technically complex, but it requires establishing very clear guidelines from business management to your AI solutions provider.
The generic foundational AI models currently available are not the optimal solution for your business. While technically impressive, their value will be limited by not being customized to your specific organization.
Confining AI within your current business software is a mistake that restricts its true potential. These traditional systems were not designed to fully leverage AI capabilities, and keeping AI captive within them would be like trying to fit a giant into a box that is too small.
Instead, invest in training a customized foundational AI model with your organization's relevant data. This involves a careful process of selecting and preparing the most valuable data sets, then feeding this vital information to the AI. In this way, the model will learn to deeply understand the complexities and specifics of your business.
However, as you open the doors of your company to AI, it is crucial to protect and strictly control the access and use of your corporate data. Data is the most valuable asset of any organization and must be treated with the utmost care and respect for privacy and security. Establishing strong policies and controls is essential to ensure that information is not leaked, shared inappropriately, or used without your explicit consent.
Finally, it is crucial to have patience and a long-term vision. The effective integration of Generative AI into your business is not an instant process, but a journey that requires time, perseverance, and sustained focus. However, as you progress on this path, you will benefit from two powerful converging trends. The first is that, as the AI learns and draws from more of your organization's data, the understanding of your company and its processes will become increasingly deep and precise. The second trend is that the foundational AI models themselves will continue to evolve and enhance their capabilities exponentially.
There will come a point where complex instructions like "calculate and send the Model of the Corporate Tax to the Tax Authority" will become routine and simple tasks for your customized AI system. That will be the moment when traditional business software becomes obsolete, ushering in a new era where Generative AI, fueled by your data and tailored to your specific processes, becomes the engine driving efficiency and innovation in your organization.
Figure 2. Business person analyzing and making decisions in their company with the invaluable assistance of an AI acting as their assistant.
Source: image generated with Dall-E 3.
The budget dedicated to integrating generative AI will depend on many factors specific to your business, to which I do not have full access and would not dare to answer generically. The more advanced your technological requirements are, the higher the associated cost will be. However, some indicative ranges can be provided.
The cost of training an AI model with your company's data and deploying a platform for both management and employees to use it appropriately would range from €10,000 for smaller companies to over €100,000 for larger and more complex organizations.
It is important to highlight that the Government of Spain, through European Funds, is investing in various programs to facilitate the integration of AI into operations for small and medium-sized enterprises (SMEs). Here is an indicative list of the aid they can access:
These aids are very interesting, as, in most cases, they allow SMEs to take their first steps in integrating AI either for free or at a very reduced cost, greatly facilitating the process.
While these aids are a valuable opportunity, it is crucial that all companies, regardless of size, allocate a portion of their budget to adapting AI. In the near future, all companies will be AI companies.
The emergence of foundational AI models, like GPT, Gemini, or Claude, has marked an unprecedented revolution in the world of technology. We are faced with a tool of outstanding capabilities and constant evolution, whose potential we still do not fully understand.
However, there is a substantial gap between the possibilities these models offer and the actual ability of organizations and their employees to fully leverage them. To bridge this gap, the most effective solution consists of two fundamental steps:
Other attempted solutions, such as chat interfaces with generic AI models or the integration of AI copilots into existing ERP systems, have proven to be insufficient and limited. The method described above is not only superior at present but will also be in the future. As the model becomes more familiar with your company and as AI itself continues to enhance its capabilities, there will come a time when this AI deployed in your organization will be able to autonomously perform any type of complex task.
As Peter Drucker aptly pointed out, "There is nothing so useless as doing efficiently that which should not be done at all." Limiting the unlimited potential of AI to the confines of traditional business software is a task doomed to failure, as these platforms were not conceived to fully leverage the capabilities of this revolutionary technology.