Blog

From November 13 to 15, 2024, IBERAMIA 2024, the XVIII Ibero-American Congress on Artificial Intelligence took place in Montevideo (Uruguay), organized by the Faculty of Engineering of the University of the Republic.

IBERAMIA is the main international conference where the Ibero-American AI community meets to share research results and experiences with Artificial Intelligence researchers from all over the world. Yesterday I had the honor of giving a presentation on the “Industry” track of IBERAMIA, the content of which I will try to reproduce in this article.

The title of my talk was “GenAI: useful business casuistry“, but it could have been “Why do only 15% of GenAI’s projects succeed?” or “How do you make your GenAI projects really improve your key processes and produce the expected ROI and Payback?“, but our Marketing team would never have allowed us to use such lengthy and descriptive titles. In any case, it is clear that the objective is to analyze, based on recent experiences, the main recommendations so that GenAI is not just another fad in your organization, but that it really adds value, optimizing processes, and helping in attracting customers, reducing costs and developing new products, all in perfect alignment with your strategy.

AI/GenAI boom and confusion

It is impossible to ignore the unprecedented impact that AI/GenAI are having globally on all aspects of the economy, society, and personal life. Let’s see some numbers released by prestigious specialized companies and market analysts in force at the beginning of this year:

• AI adoption has increased 2.5x since 2017 (Gartner).
• 80% of organizations will have incorporated AI into their processes by 2026 (PWC).
• AI is the greatest economic opportunity in history, expected to contribute USD 16 trillion to global GDP by 2030 (McKinsey).
• AI produces productivity increases of more than 45% (McKinsey).

Almost two years ago, the announcement of ChatGPT started the boom of GenAI, popularizing it at levels comparable to those of the Google search engine. But the immediate enthusiasm contrasted with the lack of real support for GenAI’s initiatives, leading some analysts to warn of exaggerated market optimism regarding artificial intelligence, especially due to the huge investments in data center infrastructure, which after more than a year were not having their corresponding return on investment (ROI).

Some specialized journalists went as far as to comment casually that the boom of the GenAI was transforming the CIOs in relation to the GenAI into something similar to teenagers in relation to sex, because “everyone talks; everyone wants to do it; but very few know much” about the subject.

The difficulty in identifying use cases truly aligned with strategy, capable of generating the expected ROI, led some analysts from Goldman Sachs and Sequoia to announce the existence of some evidence of an AI bubble, which the collective memory quickly associated with the Internet bubble of a few years ago.

In a recent CNBC interview on 07/08/2024, Roger McNamee, CEO of venture capital firm Elevation Partners and an experienced technology investor, said that CAPEX in AI is being made based on projections that are difficult to assess and with a high chance of not being confirmed. “The United States loves financial manias,” said the executive of Elevation Partners, one of the first firms to put money into Facebook.

Roger McNamee also made other “provocative” statements:

·       “There is evidence of the formation of a bubble in artificial intelligence, and, if that were the case, sooner or later there will be an adjustment.“.

·       “When companies like Sequoia and Goldman Sachs tell me there’s a bubble, I pay attention.“.

Goldman Sachs analysts quickly “picked up the gauntlet” rushing to rectify the apocalyptic vision by emphasizing the differences with the Internet bubble, starring startups of recent existence and very few resources, contrasting with the “financial backs” of Big Tech and Hyper Scalers, protagonists and main investors in the infrastructure required to host LLM’s, large consumers of storage capacity and expensive graphics processors.

In any case, the rapid evolution of the technological offer aggravated the confusion, confronting decision-makers with supposed dilemmas about a wide range of possibilities in terms of the type of model and platform to be used. Firstly, the questionable high cost of the large foundational models and LLM’s (Large Language Models), gave rise to the emergence of SLM’s (Small Language Models), of smaller size and cost, although covering most of the scope of the LLM’s. This created the possibility (need?) to opt for an LLM or an SLM.

You can also opt for implementation in the cloud, or on servers or in your own data center, on premises. In the case of using an LLM, the option will have to be cloud, since the large foundational models and LLMs are only available in SaaS mode in the cloud. SLMs, on the other hand, can be considered for an on-premises implementation due to their smaller size.

Whether it is an LLM or an SLM, to be used in the cloud or on premises, it will also be possible to opt for the use of proprietary models (in general Big Tech) as well as open, open-source models. The most widespread benchmarks report increasingly similar performances between one option and another, and it is advisable to conduct a more detailed technical analysis considering the types of functionalities most used in each case to make this decision.

Another possible dilemma is to opt for probabilistic models or deterministic models. A real and recent case of a Quanam Client  worked with us in a POC using an LLM (probabilistic model) and the Users of the Client’s Business Area detected that, at different times, when faced with the same question, the answers generated in natural language did not coincide exactly with each other with exactly the same words, although the conceptual answer in all cases was the same. Curiously, this factor, typical of any probabilistic model (as it also happens with human beings) led these Users to ask us to evaluate using a deterministic model, based on a decision tree, which would always produce exactly the same answer to the same question including the same words.

In our experience, without prejudice to analyze the characteristics of each case, it is very common for these supposed dilemmas LLM vs. SLM, cloud vs. on-premises, proprietary vs. open source, probabilistic model vs. deterministic model, are in fact false dilemmas, being possible to opt for hybrid combinations of these characteristics, making functionalities from any of the aforementioned origins collaborate.

Tips for success

Although there are no exact formulas for the success of innovative initiatives using GenAI, our rich experience over the past two years has allowed us to identify some determinants of success or failure. Here we briefly summarize five areas to consider:

1.      A key trilogy for success: working with a truly specialized Consulting firm that helps define the Platform Manufacturer and the LLM/SLM/Foundation Model to be used, as well as to define together with the Users of the Business Areas, the use cases to be implemented and their form and objectives.

2.      Perhaps the most important thing is to define use-cases that really add value to business processes: optimizing the core business, aligned with the overall strategy and data strategies, capable of producing the expected ROI and executable in timeframes and with a budget adequate to the organization’s pocket.

Stanford University recently listed the main areas of effective application of GenAI:

a. Contact Center: 26%

b.      Personalization: 23%

c.      Customer Acquisition: 22%

d.      Improvement of AI-based products/services: 22%

e.      Creation of AI-based products/services: 19%

Our experience shows the special relevance of working as a team with the Client’s user areas, even after the use case has been chosen. For example, at the IMM (Municipal Government of the city of Montevideo) we originally thought of developing a chatbot to answer questions from specialized users (architects and qualified technicians) about the regulatory provisions related to Construction Permits and, while interacting with IMM’s users we realized that in addition to the answers in natural language it was essential to provide, in a hierarchical manner of relevance, the applicable articles of the Municipal Digest.

3.      Defining the most appropriate architecture for the use case is important, solving the dilemmas noted above: LLM or SLM, probabilistic or deterministic models (or both), “on-premises” or “cloud“, proprietary or “open source” (or both).

4.      The fourth tip is to integrate different technologies, collaborating with each other in the same solution or use case, that is, not to limit yourself to the use of AI-GenAI technology, considering other advanced technologies, such as Statistical Models of Predictive Analysis, BPM (Business Process Management) and RPA (Robot Process Automation) especially in cases of Intelligent process automation. Preferably taking advantage of technological products already existing in the Client.

We cite the case of a project implemented by Quanam with and for the National Directorate of Industries of Uruguay (DNI)with the aim of reviewing the feasibility of incorporating artificial intelligence into existing processes. The process of importing auto parts from Brazil was selected, within the framework of the Economic Complementarity Agreement in force, considering the number of procedures carried out annually, in the vicinity of 3,000 instances.

The main benefits for DNI are that the solution allows you to speed up the validation of applications and free up human resources. To this end, BPM processes were integrated taking advantage of the APIA BPM tool that the DNI already had, with a BOT developed with the Microsoft Power Automate Desktop RPA tool that the DNI also had, and functionalities of the Microsoft Power Automate Flow and Microsoft AI Builder tools of Power Platform.

5. Finally, our fifth tip is to take a risk-based approach to your innovative initiatives, assessing and mitigating specific risks associated with GenAI projects, such as data privacy breaches, model biases, hallucinations and inaccuracies, and infringement of intellectual property provisions.

We recommend incorporating into your compliance and ethics policies, standards and guidelines specific aspects related to the responsible use of AI/GenAI-based services and solutions, and hiring the services of consulting firms that, like Quanam, also have this ethical profile and use good practices of transparency, ethics, integrity and compliance on a daily basis.

More about the Consulting firm

We emphasize again the importance of caring about the key trilogy for success, starting with a Consulting firm with deep knowledge from the technological and methodological point of view in relation to AI/GenAI projects, but also with genuine experience in such projects. Hiring a Consulting firm with these credentials will help the Client to make informed choices about the platform(s) and technologies to be used, choosing the Manufacturer(s) and interacting with Client’s Users, including the IT&C and business areas staff  involved, to choose and “shape” use cases truly adapted to the requirements, needs and possibilities of the Client, thus significantly increasing the probability of success of the implementation, achieving its objectives and producing the expected return on investment.

Quanam is an “agnostic” consulting firm: without prejudice to advising which platforms, services and IT&C tools best suit each case, we “do not marry” any brand, we are open to all and have experience with most of them. We commonly use the main foundational models / LLM’s, in alphabetical order:

· Bedrock from Amazon
· Gemini from Google
· Granite & watsonx.ai from IBM
· Llama from Meta
· Mistral LLM by Mistral AI
· OpenAI ChatGPT from Microsoft

In Uruguay we have the privilege of having two world-class laboratories, both installed on the LATU (Technological Laboratory of Uruguay) campus: the Microsoft Co-Innovation Lab and the IBM Build Lab at the Uruguay Innovation Hub. These are laboratories with world-class resources, and that, although they are in Montevideo, were created to conduct projects from all the countries of the region, from the Rio Grande (Mexico) to Patagonia. Quanam’s knowledge and experience are at the service of private companies and public institutions in Uruguay and Latin America to help them identify projects implementing use-cases really useful to the organization’s purpose, integrating with them and with the selected Manufacturer(s) the appropriate trilogy to guarantee the success of their innovative projects.

Quanam has been working on AI for more than fifteen years, and in the last two years we have led GenAI’s initiatives in Uruguay and in the region, having created numerous successful projects, among which we mention the following:

· OUC – Orlando Utilities Commission (water and electricity): Genera IA, GenAI application/product created by Quanam to automatically generate specifications / RFPs / RFQs in purchasing processes. Numerous Uruguayan public institutions are evaluating this product.

· BROU (Banco de la República Oriental del Uruguay): Development and implementation of Virtual Assistants for the bank’s Internal Processes.

· IMC (Municipal Government of Canelones): Development and implementation of a Virtual Assistant for information and traffic procedures in the Department of Canelones, Uruguay.

· ANTEL (National Telecommunications Administration): Development and implementation of Virtual Assistants for Account Executives and for Marketing Survey Analysis of the main telephony and telecommunications operator in Uruguay.

· Hospital de Clínicas: Development and implementation of Summaries of Medical Records for ITC (Intensive Treatment Center) patients of the largest University Hospital in Uruguay, optimizing patient care time.

· Xn: Development and implementation of Virtual Assistant for access to knowledge bases including videos/documents. “X a la n” is a consultancy specializing in strategy, management, leadership, and organizational and personal development.

José C. Nordmann
Chief Compliance Officer en Quanam