While South African organisations rush to deploy artificial intelligence, many are finding their efforts slowed by a lack of basic data readiness.
This is according to Daniel Acton, chief technology officer at Accelera Digital Group (ADG), who adds that the sophisticated use cases promised by AI are impossible without a stable foundation of high-quality data.
At its core, says Acton, “machine learning uses data to make predictions. To illustrate with a simple example: predicting the price of a property if the only attribute that mattered was its size could be fairly straightforward. If a 100m² property costs R100 000 and a 200m² property costs R200 000, a machine can easily predict the cost of a 400m² unit (as could a human).
“However, real-world business problems are rarely that simple. When you add variables like location, amenities, and security features, the complexity quickly makes it infeasible for a human and creates a need for machine learning models.
“Whether using simple regression or generative AI, the prediction is only as good as the data upon which the model is trained. If the data is unavailable or of poor quality, the value of using machine learning reduces greatly,” he states.
The triad of data readiness: silos, authority, and quality
Most organisations currently struggle with three specific barriers to data readiness for AI and machine learning (ML) use-cases. The first is data silos, where data sprawl across enterprise systems (such as CRM, warehouse management, and legacy systems) requires integration which can become brittle and expensive to maintain.
The second is authority. Acton cites a common scenario in banking where a customer holds a personal account and vehicle finance at the same institution, yet their details may differ across these systems.
“As a customer, it often feels like those systems belong to different banks,” he says. “Organisations struggle to define (and enforce) which system holds the ‘golden record’, and that could create complexity in adhering to compliance mandates like FICA.”
The third barrier is data quality, referring to the accuracy and completeness of information. As systems duplicate data, values conflict, and without a single source of truth, it can be complicated to know which value is correct.
Five priorities for the C-suite
To move from data chaos to data readiness for AI, Acton suggests executives prioritise the following five areas under a ‘data foundations’ strategy.
- Treat data as a product – CxOs must shift internal mindsets so that data is viewed as a product. Each team, product or project that produces data must own that data and ensure it is available for other systems to use. This moves ownership away from a vague IT responsibility to a specific business deliverable.
- Unify structured and unstructured sources – Valuable business insights are not just found in rows and columns. They exist in forms, documents, and images. Leaders should implement a data strategy that accommodates both, typically through a data lakehouse architecture that handles batch and real-time sources.
- Define governance and security upfront – Before building models, organisations must analyse their data sources to understand governance needs. This includes defining how fresh the data needs to be, what archiving, data lifecycle or data retention is required, and especially how Personally Identifiable Information (PII) is treated to ensure regulatory compliance.
- Monitor integration performance – Building the data integration pipelines is not enough. Businesses need to implement integrations (ETL or ELT) and actively monitor the performance of these data flows. If the pipeline breaks, the AI stops learning.
- Enforce accountability via KPIs – Organisations should implement Key Performance Indicators (KPIs) around data quality and availability to ensure producers take ownership. “If we want to be data-driven, data must be the concern of everyone,” says Acton. “While a central data team or Centre of Excellence can act as, and enable, engineers and stewards, the quality of the data is the responsibility of the team that generates it.”
Continuous improvement
Acton emphasises that data readiness is not a one-off project. Once the foundation is set, it requires continuous monitoring of quality, freshness, and integration health.
“When a new team or system is introduced, it must align with the data strategy immediately. By establishing these foundations, businesses can ensure their AI investments deliver genuine insight rather than expensive guesswork,” concludes Acton.