Since their conceptualisation five years ago, data meshes have yet to achieve the outright transformation of data management that many had forecast. That is not the fault of the concept or its philosophical underpinnings, but an issue of human nature, argues one of the technology’s proponents.
Data Virtuality’s Tammie Coles believes that the key ingredient to any successful data mesh implementation is getting the human element right. From boardroom buy-in to data team acceptance, the pain points for financial institutions that still struggle with their transformation can be traced back to change management policies, Coles tells Data Management Insight (DMI).
“A lot of clients that already have initiated something on data mesh realise that the idea of it and the reality of it are very difficult to bring together,” says Coles, who heads the company’s global sales team. “Data culture can kill a data strategy as well as bring it to life; you need to get the people within the organisation open for change and open to new things. As obvious as that is, it remains a big stumbling block.”
The data mesh concept was crystalised in 2019 by former ThoughtWorks technology director Zhamak Dehghani as a solution to increasingly complex data management challenges, especially as enterprises ingest ever-greater volumes of information from outside sources.
Decentralised Specialisation
Dehghani, who has since gone on to found and head Nextdata, envisaged a mesh as a decentralised framework in which data is treated as a product owned within corporate domains but shared across an enterprise through a self-service setup under a federated governance structure.
Along with other proponents, she sees meshes as a means of making data more accessible, available, discoverable, secure and interoperable. With so much data within modern systems, this federated architecture should enable teams to query data and to translate insights into faster times-to-value.
Coles, however, says that many firms have failed to grasp the preconditions necessary to make a mesh work.
“I would argue that the ones who say they tried to put data mesh in place and haven’t succeeded predominantly have not done so because of the [corporate] political thing,” she says. “It might excite a couple of people in an organisation, but then other people might be too afraid of change.”
Ill Defined
Another reason for failure has been an inability to fully understand the concept. Coles says she is aware of companies that have tried to implement a mesh-based strategy only to find that they had interpreted the principles incorrectly. Dehghani also suggested as much, arguing that companies had redefined “federated governance” to mean giving non-technical people access to all data.
Managing change is difficult and Coles sympathises with companies that embark on such a programme. Any data management transformation can be long and arduous, she explains with reference to the experience of Commerzbank in relation to a separate data-related project. The German bank managed to get the project in place within a year, but it took another six months to get the technical implementation of the project right.
She likens the governance change necessary to implement a data mesh to the transition of a country from a monarchy to a democracy; when users are first given autonomy over how they use and manage their data, “they are lost” and unsure of how to behave, how to explore and how to make best use of their data.
Similarly, proposals to decentralise control over data are often rejected by those who command access to it.
“If you are the monarch you’re not likely to want to give up your power,” she says. “They think ‘there’s no way I’m handing it over’.”
Still Necessary
The charm of a data mesh, Coles says, is that it doesn’t require organisations to “rip out” the technologies and products that they already use. Instead, a mesh can be moulded to sit above tech stacks. This makes them a valuable proposition for bringing more effective data use to organisations without the expense of rearchitecting their infrastructural setups.
Making that easier now, she argues, is artificial intelligence. Machine learning (ML) and large language models (LMMs) can do more than transform the data monarchy, she argues – it will instigate a full-on revolution.
“In future, I believe that AI will enable the less technical to take full advantage of the data – thereby challenging the data monarch,” she says. “This revolution will have an impact on our day’s data management concepts, including data mesh and data fabric.”
In the meantime, data meshes have a valuable role to play for organisations if they establish the interdepartmental connections to make it work before they embark on implementation.
“It’s about awareness, about picking people up and generally pointing out what the advantages are, what the company wants to achieve and how it benefits most of the people,” she says. “I don’t think that gets clearly communicated. There must be trust in the companies to do that, really, and not necessarily lead people to believe that changing to a data mesh will mean they will have to do more work.”
Subscribe to our newsletter
