The 5 (data) foundations of AI success

AI is already shaping business and society. Organisations are turning to AI for task automation, improved data analysis, better customer relations, and increased efficiencies across all operational functions, helping them to progress from digital to intelligent organisations.

Advances in computing power and technology are making AI much easier and cheaper to use. This will encourage accurate data aggregation that, in some cases, could save lives. For example, working with Unilever, we co-created the COVID-19 Awareness and Situational Intelligence (CASI) predictive intelligence tool, combining our expertise in AI, machine learning, data analytics, and operational resilience to accurately predict COVID-19 trends at global, regional, and site level.

However, organisations sometimes feel stuck, struggling to move from an initial experimentation phase into production, perhaps abandoning AI plans altogether. But why? Is AI not the right solution, or were the right foundations not in place? Let’s unpack the data foundations needed for AI success, which also provide a high-level sense check for AI readiness.

Data strategy 

It’s a common mistake to jump on the bandwagon without knowing where you want the wagon to go. To get the most value from AI, take hold of the reigns and steer towards clearly defined objectives. Do you want to streamline your internal processes? Do you want to understand your customers on a much deeper level? Do you want to enhance critical decision-making? Once you have defined which aspects of your organisation would benefit from AI, ask, “Is my AI strategy aligned with my overall digital and data strategy?”, and if not, what needs to change in your digital and data strategy to allow AI success?

An AI Studio could provide a fertile environment for addressing these questions. We’re working with clients across multiple industries to embed agile AI Studios, which combine digital and strategy expertise to build capabilities and policies specifically for AI. This approach enables clients to set up their AI strategy, identify use cases, assess maturity, and get started with experimentation.

Data preparation

Preparing your data isn’t just about collection. It’s about organising, structuring, and storing up-to-date data so it can be easily accessed and understood by humans and algorithms alike. AI is only as good as the data it is fed – in other words, poor input, poor output. Start by carefully designing DataOps models including processes, roles and responsibilities, technology, and ways of working. Outline governance structures and architectures to enable transition, enhancing data management, governance, and data quality while also reducing duplication and improving data democratisation.

To maximise success, collect, cleanse, curate, and organise data appropriately, stripping out errors and inconsistencies. To promote integrity, implement systems that generate clean, updated data, and store data in a controlled data management solution. If this sounds daunting, consider investing in data specialists or upskilling employees to make data work normal – not novel.

Governance 

A data governance framework is essential for compliant data is handling in line with regulations and policies. A good data governance framework includes rules and standards that enable consistent, secure access to data across the whole organisation. The framework must ensure data is collected, managed, and used securely and ethically to promote transparency and accountability. You will need also to create explicit guidelines for ethical AI to avoid conscious and unconscious bias, protect privacy, and sanity-check against possible unintended consequences. This also highlights the importance of tempering AI with human sense and sensitivity.

Scalability

AI algorithms are fuelled by data – and lots of it. To support future scalability and manage growing data volumes, embed a robust IT infrastructure able to handle large amounts of data, alongside flexible data storage and processing capabilities that can accommodate varying data requirements. Design and deliver cloud-native Machine Learning Operations (MLOps) architecture and maximise the capabilities of your chosen cloud partner (or partners). Ensure you architect IT environments in a rapidly scalable way to support fast data processing without incurring long waiting times for AI insights.

Feedback

Keep track of progress, establishing benchmarks and KPIs to measure and evaluate performance over time. An AI value tree can provide a useful tool to help organise and prioritise the different dimensions and outcomes that your AI strategy aims to achieve. An AI value tree clearly visualises the hierarchy of objectives and the relationships between them. Based on this understanding of what works and what can be improved, you can start to experiment with new cases and opportunities.

AI is not a fix-all solution. While its potential is phenomenal, it relies on the right conditions – and the right data – to thrive. Closing the gap between the excitement that AI brings and the need for foundational data work is crucial to ensure the ethical, successful development of any organisation’s AI future.