How Data Mesh Can Unlock Value Through Hyperconnectivity

We’ve heard that data mesh represents a paradigm shift in analytical data management. On the surface, data mesh refers to a decentralised, distributed approach to enterprise data management; a holistic concept that sees different datasets as distributed products, orientated around domains. 

Nevertheless, perennial questions remain as to how organisations can successfully establish clear data connection points across dispersed business units, finding a balance between allocating governance responsibilities at the mesh and domain level. 

In a recent Virtual Town Hall, we sat down with leading CDOs in the UK & Ireland region to explore what data mesh entails beyond conceptualisation – exploring tangible, real-world instances. 

Leading this discussion, Andy Mott, EMEA Head of Partner Solutions Architecture & Data Mesh Lead at Starburst and Sanjeevan Bala, Group Chief Data & AI Officer at ITV discussed some of the nuances and challenges encountered while implementing data mesh across the organisation. 
 

An Alternative to the Single Source of Truth

Principally, we saw that data mesh represents a move away from the traditional ‘single source of truth’ model. Where the single source model approach favours the aggregation of dispersed data into a singular, monolithic location, data mesh breaks this down into distinct, interpretable reference points corresponding to the organisation’s various business domains. 

These individual sources of truth are owned by the various business units, who have absolute autonomy over the ingestion and transformation of their data sets. 

This hints toward one of the fundamental pillars of data mesh: domain-driven data ownership and architecture. Domain ownership empowers teams to perform their data activities separately, foregoing the singular, centralised Data team altogether. 

The upshot is that teams with the greatest depth of knowledge and experience around a particular source of data will ultimately decide how this data set will be leveraged. This, in turn, helps to eliminate a bottleneck between centralised data producers and analysts who are far less knowledgeable about how data should be applied. 

As one could imagine, data mesh requires a broader cultural shift within the organisation, reorganising data skills in accordance with the various business domains. Herein lies the first key challenge: how can an organisation clearly define who these domain owners are, and which data sources they should possess autonomy over? 
 

Challenge 1: Defining Domains

As we have seen, in order to empower business domains fully, it is crucial for organisations to clearly define distinct categories, but how? One approach favoured by many organisations is to map them in accordance with prima facie intuition. It seems logical that if a particular business unit tends to deal with specific sets of data, then they should be the ones to have ownership and autonomy over its data sources.

At a cursory glance, this looks like an effective strategy – and this is true in many cases. However, the efficacy of this approach appears less obvious in cases where the margins of data ownership are blurred, and multiple teams can justifiably stake their claim to a particular data source. 

Take the example of ‘the customer’ – to which business domain should this data source be allocated?

One answer might be the Marketing team. What about the Sales and Content teams? Surely, by following the same line of prima facie reasoning, all these business units can provide reasonable grounds for why they should be the domain owners over this data source.

Indeed, these all seem like logical choices, and this is the challenge at hand. By asking which team should possess autonomy over a particular data set, you can end up with a plethora of distinct answers, where you ideally wanted one.
 

Challenge 2: Improving Standards of Data Literacy 

Nevertheless, let us assume you are able to successfully define and operationalise your various business units as owners over their respective domains – what next? We have seen that those with the greatest depth of knowledge and experience around a given data source can ultimately decide how it will be leveraged. 

However, there is a crucial distinction between having knowledge around a particular source of data and being able to successfully interpret and execute on it. This is where we arrive at the second key challenge: how to improve standards of data literacy across the domain owners themselves.

Broadly speaking, the long-established paradigm of analytical data management as the sole priority of the centralised data team has resulted in significant deficiencies in data literacy across organisations. Here, business units existing outside the traditional walls of the centralised data team lack fundamental experience in interpreting and utilising raw data pools.

On this point, Mott and Bala refer to the purported transformation of a "data-driven" organisation to one which is "data-enabled." Here, the former refers to an organisation which has successfully aggregated an enormous amount of data but does not know what to do with it. By contrast, the latter "data-enabled" organisation draws directly from these aggregated data pools to inform their best practices and business decisions. 

Within the context of a data mesh model, organisations must therefore strive to improve data literacy across the domain leaders themselves – only then will they possess full autonomy over the ingestion and transformation of the data sets.

However, this is no easy task. For stakeholders who have been at the business for many years, the prospect of working directly with raw data sets might seem alien and impractical, preferring the traditional ‘single source of truth’ model – because as the old adage goes, “if it ain't broke, don't fix it.”

Therefore, to facilitate the broader transition to a “data-enabled” organisation, it is crucial to develop novel learning pathways, providing foundational data training and upskilling the dispersed business units. Crucially, the aim here must not only be to improve data literacy, but also to incentivise the learning process itself. To take full advantage of the data mesh model, business units must want to learn how they can best utilise their newfound autonomy as domain leaders. 
 

Closing Thoughts – Redundancy of the CDO?

As we have seen, the road towards data mesh is not without its obstacles. Defining and empowering domain leaders presents a myriad of unique challenges which cannot be overlooked or underestimated. Nevertheless, despite these challenges, the data mesh model opens an intriguing conversation about the future of the Chief Data Officer role.

With the decentralisation of authority over data sets – and the concomitant emergence of autonomous domain leaders – the need for a Chief Data Officer no longer appears self-evident. Closing the conversation, Mott and Bala speculate the imminent redundancy of the CDO role. 

While some might view this through a pessimistic lens, Mott and Bala argue that this is the very purpose of the CDO – to design a data governance structure such that they are no longer required for it to operate effectively. 

In this view, as the CDO’s remit evolves towards one which implements and oversees change, the role itself may re-emerge under the banner of Chief Transformation or Chief Growth Officer. Only time will tell.

by CDOs, for CDOs

Join the conversation with peers in your local CDO community.