Like often with innovative technology, it generates massive interest and attention, as well as a degree of FOMO. This is also the case with GenAI and LLMs. For instance, many companies have developed a narrative and an expectation that the strategy is to "create data products" and "utilize AI to the full extent"…

These main points, when viewed in isolation, make sense and are something most companies should strive for, but they come with a cost and at times, a somewhat backward approach...

Today, most CXO's responsible for data and IT are facing two closely connected areas of the data agenda:

• Area A - The backbone essentials
  Data architecture, data infrastructure, data management, data governance, BI and reporting, self-service, and data operations
  
  

• Area B - Advanced analytics
  Data science, statistical modelling, Machine Learning, Deep Learning, GenAI etc.

Both areas are important – and in most scenarios key – in the value creation with data. However, there is a natural flow and dependency from A to B. The required foundation in A is essential to succeed with B. In other terms: Without data, no machine learning or AI. It is as simple as that.

Yet, a narrative has escalated focusing primarily on the buzz of B while taking all elements of A for granted. The problem with this narrative is that it neglects the fundamentals and puts many data leaders, data teams, decision makes and data enthusiasts in a difficult spot as boards, management, business stakeholders, end-users, media – well, most of us to be honest – find the new and fancy AI innovation more fascinating, leaving out the path and prerequisites to get there.

AI sells the tickets; Platform Engineering and Data Engineering make it possible to execute upon and succeed with.

As companies increasingly strive towards “utilizing AI”, “become data-driven” or “create data products” the dependencies and barriers between A and B becomes even more visible:

• No strategic direction on data
  No wind is favorable if the direction is unknown. And if a data strategy exists it is often not followed, communicated, anchored, of sufficient quality or a realistic functional version of the technology and business requirements. A two-page document in the archives is not a strategy.
  
  

• Architecture and platform are lacking
  High ambitions with data are hard to realize on legacy systems. Certain architectural pieces are required to support analytical needs, scale, and be operational.
  
  

• Use cases aren't really AI use-cases
  Focus is way too often on which tool to use rather than value creation. Identify the challenge first - then the tool. This also applies to GenAI versus ML, Deep Learning, etc.
  
  

• Low data quality, lack of data management and governance
  It's quite simple. Poor data, poor insights. Data quality is a shared responsibility and not just a technical exercise.

  
  

  The essence of the barriers highlighted above is that there are some basic needs that must be in place within companies to “utilize AI” from some actual ROI. Solid data products or AI models only become a reality when both basic and associated needs are addressed. From A to B: