Trying to keep up in AI? Be SMART with data
IFLR is part of Legal Benchmarking Limited, 4 Bouverie Street, London, EC4Y 8AX
Copyright © Legal Benchmarking Limited and its affiliated companies 2024

Accessibility | Terms of Use | Privacy Policy | Modern Slavery Statement

Trying to keep up in AI? Be SMART with data

Sponsored by


Salvador Abecasis de Carvalho of Morais Leitão, Galvão Teles, Soares da Silva & Associados presents several dos and don’ts as companies ponder how to create value using AI and avoid standing still

Access to information through data has been evolving at a very high pace. However, with the introduction of AI and generative AI, the jump has been bigger than ever, and we can now find AI in everyday tools and, indeed, everywhere.

The inevitable presence of AI seems to be a bright light in the long-tail challenge associated with data. Large amounts of data, complex toolsets, machine learning operations out of reach, and limited returns on investment are not constraints anymore, as AI presents a possibility to find value in use cases that was not available before. Know your customer procedures (automated with an ID), retail (where smarter recommendations have boosted customer segmentation), or even manufacturing (smart anomaly detection from bugs) are use cases where the landscape has changed, and simplification has emerged.

Companies have multiple chances and paths to enter this new world, which seems to be here to stay. Nevertheless, what is a good strategy? How is it possible to create value from AI? What do we need to secure prior to this shift? The purpose of this article is to delve deeper into these questions and try to understand why AI is as good as the data we have.

Creating value with AI

According to McKinsey & Company (2023), the generative AI impact is set to boost the economy by an impressive $4.4 trillion annually, based on its capability to unleash a new wave of productivity.

This forecast is based on the crucial assumption that we, the data owners, are seated at the front row, which makes us responsible not only for achieving AI’s maximum potential but also adapting our mentalities. We can take the wheel if we have in mind the following six connected steps.

1 Data needs to be aligned with the business strategy

Data and business should have a common language. We need to understand what business problems we have and try to solve them with planned data and analytics. We should only be leveraging the data necessary to solve the existential business problems. This alignment has a positive effect on building relationships with the company leadership, and since this is not an overnight or cost-free process, such support is crucial.

2 Companies need a data-driven culture and ecosystem

This is a major step towards enabling your company’s growth, efficiency, productivity, innovation, and competitive edge. It relates to the ability of the entire company, from bottom to top, to optimise processes and make better decisions with data. However, this is a step that needs to start at the top end, with top managers who set an expectation that decisions must be anchored in data. The analytics and data within the company need to:

  • Be transparent;

  • Be accessible for all (whether we provide employees with training or just use it on their behalf); and

  • Have robust and simple proofs of concept rather than fancy and frail ones.

3 Data governance is core

Data governance is one of the main pillars of a data organisation, and crucial to get our data ‘ready for business’, as it encompasses everything we do to ensure that data is secure, private, available, and accurate. Good data governance allows us to go from a data anarchy to a data democracy where data is reusable, data usage increases, there is a better lead time, there is the ability to scale data use cases (answering business challenges by leveraging data) with less effort, and the company is in a place where data is the common good and is communicated as a brand, but it comes at a cost of organisational and management changes.

4 Data management

Data teams may be spending 80% of their time finding, cleaning, and organising the data, and 20% performing the analysis. Along with other data-related issues – such as inconsistent data, data illiteracy, and constraints to change – this requires better data management and an improvement of data skills.

Data management consists of all practices that deliver, control, protect, and enhance the value of data and information assets throughout their lifetime and they must be cross-functional with all business and IT units, and be guided by industry-standard frameworks (i.e., DAMA) and data maturity models (i.e., follow a life cycle of data–modelling–deployment and orchestration–monitoring). Do not forget that the success of AI models is dependent on the availability of trusted and timely data.

5 Data privacy

According to Harvard Business Review (2020), companies tend to be data rich and information poor, meaning companies still struggle to incorporate data into their business strategy. As they try to do so, data privacy is shaped by two major pillars: how is data collected and how is data used.

In previous years, many data protection laws have been enacted, such as the EU’s General Data Protection Regulation in 2018, and 137 out of 194 countries have legislation to secure the protection and security of data. However, a company’s response to a data breach can result in a positive or negative impact on its reputation, market, and money.

On this note, Target and Home Depot have suffered data breaches; however, since Home Depot had a high-quality response, the home improvement retail corporation was able to solve the situation in a matter of days, whereas Target took weeks to solve its issue, resulting in profit loses, executive losses, and a negative bump in the stock market. A good data privacy policy enhances a company’s strategy, organisation, management, process, and technology.

6 Ethical AI and data

Due to what is at stake and the associated risks, companies need to bring the ethical side of AI and data use to the table. Having internal approval for all AI projects, a well-defined data governance programme, safety filters, transparency in terms of trends and challenges, approaching each case as a different case, and adding the human touch/sensibility to give context are the soft spots to become more ethical in this line of work.

Implementing an AI strategy

Anything that draws attention and involves the possibility of making money is tempting and therefore people tend to rush in without assessing how to solve the puzzle in front of them.

AI brings this exact feeling, and a recent study carried out by Gartner (2023) showed that 70% of organisations are already in an exploratory mode with regard to AI.

However, we must be aware that this does not happen overnight, and it might need a big cultural change. Also, new technologies such as large language models (LLMs) do not always combine with design thinking processes for new product ideas. We shouldn’t ask “How can we integrate LLMs in our product?” but instead “What problem are we solving?” This way we can avoid flaws due to lack of clarity or loss of connection to value creation.

We should seek ground solutions for tangible products and adopt the concept of emerging thinking instead. This type of approach focuses on understanding the technology first (for example, what is the natural language? Where does it collect information from? What types of formats can smooth the processes?) and lets us know we should only connect the dots for relevant problems if we can at some point find one of the following trees:

  • Exploit – for example, Talabat, a giant food delivery tech company that introduced a chatbot within its app and adds the required ingredients for a particular recipe to the shopping cart, thus improving customer retention and reducing the thinking process;

  • Expand – for example, Khan Academy, an online teaching platform that helps teachers by explaining teaching methods and designing lesson plans, allowing them to improve their skills and innovating the classes; and

  • Create – for example, Genei, a summarisation and research tool that enables users to extract insights and information from any web page or document, which is groundbreaking with regard to the management of information.

After the initial approach of how to think, we then need to be aware that AI is not a bed of roses, and recognise the existing vulnerabilities and complexity. Therefore, we need to:

  • Understand attack risks (i.e., prompt leak, prompt Injection);

  • Design risk management strategies (i.e., predefined responses, rate limits, limit context retention);

  • Raise awareness within the tech teams (i.e., train staff about attacks, stay informed about best practices and emerging treats);

  • See integration as a multi-step process (bring different teams together, train the models various times, prove value prior to long steps, account for uncertainty, real-time learning);

  • Realise that solid data infrastructure, data governance, and data quality are not a nice to have but a must have, so that data can be trusted, and the flow can be easily adjusted;

  • Build on a good and solid experimentation platform, build trust, clear ways of work and processes, foster coordination between teams and assignments, engage in continuous experimentation (A/B testing), and align people and their incentives); and

  • Explore open-source possibilities (i.e., Netflix is 100% open source).

Your AI strategy must be resilient, and you should bear in mind the following:

  • Continuous metric-driven experimentation – never lose focus of enticing your customers, and create and test new metrics that can resolve problems and help your company to go the extra mile, thus creating a competitive advantage (for example, Netflix has a much lower churn rate when compared with its competitors because it acts on a metric-driven mindset, by creating a repository for watching later, sending regular reminders, and having a rich range of content);

  • Innovation and product development – in a constantly changing world, it is important to innovate and keep up with evolution, and AI can help us to do that (for example, Rescale, a cloud-based platform for computational engineering and R&D, is leveraging AI to improve design and testing processes to build better products and pioneer innovations);

  • Automation and efficiency – one of the main goals of a company is to make money, and to do so, revenue should increase and costs should reduce, which can only be done if there is an increase in efficiency and automation (for example, Klarna – a buy now, pay later service – has developed an AI assistant powered by OpenAI that is capable of doing the work of 700 full-time agents, speaking 35 languages, communicating 24/7, and handling a third of its customer service chats, and is expected to drive $40 million profit by the end of 2024).

Final thoughts: be SMART or risk inertia

The fear of standing still is real, because regardless of how stable the business may seem, you must keep moving forward. Challenging the status quo and taking risks are essential. Otherwise, instead of progress, you risk falling behind the competition, developing inefficient processes, facing rising costs, having limited decision-making capabilities, and missing out on opportunities for innovation.

AI is only good if your data is also good. The opportunity and the value of entering this world is within reach; we just need to be SMART (be specific, save or make money, be accurate, ensure business relevance, and have a timeline in mind) about it.

Gift this article