You don’t have enough Analytics Translators, here’s why that’s a problem

I often get asked the question ‘Why do AI projects fail?’ As a data science consultant, I’ve seen a variety of organizations struggle to make AI work for them. Here are some of the challenges I often hear:

• “We have trouble understanding what AI is or why we should invest in it”
• “We don’t know how to set up data science teams and what skills are needed”
• “We can’t seem to find the right use cases that bring real value”
• “Our projects get stuck in the experimental phase; we’re not able to scale them up”

As with most technology projects, the problem is not in the tech itself, but rather in understanding how it is best used to drive value. For that, you need to be working on the right use cases and organize your teams such that they can deliver data products efficiently.

What we have found is that successful AI initiatives share at least one common factor: they were driven by someone linking the technical expertise of data scientists and data engineers to domain and operational expertise embedded in the business. Some examples:

• A major food and beverage player in the Americas asked us to help create a demand forecast for their warehouses. Success was at least 50% determined by finding someone on the ground who could gather requirements, free up IT resources, as well as evaluate and (sometimes literally) translate model output to business end users.
• A Dutch Telco looked to recommend better treatments to its call center agents who work every day to solve customers’ issues with digital TV and internet. We could not have built anything without having a champion with domain expertise who knew who to convince to prioritize the project and develop a wider vision on the potential of AI-powered decision engines.
• One of our airline clients wanted to develop a passenger forecasting system for supply chain management. Crucial to success was a product owner who liaised between business and data science teams, took ownership of the product, managed project timelines, and ensured the final product was deployed smoothly to production in line with the meticulously mapped-out requirements.

The Harvard Business Review calls these people Analytics Translators, and estimates that demand for them will grow to up to a quarter of your workforce.

With enough investment, you can get the best data scientists, engineers and architects to build impressive software applications for you, using the most cutting-edge open source tooling currently available. But without translator skills as part of the DNA of your organization their talents will not be utilized in the best way possible.

Of course, you can hire a lead data scientist who has years of machine learning experience under their belt in order to see at a glance if a use case has potential and what is needed to build the right software. But these people are scarce, expensive, and ultimately do not have the time to scope out your entire organization.

Real transformation to a data-driven organization requires that anyone in a position to recognize and exploit opportunities for analytics have had at least some degree of translation training. This change of course starts with the front runners of your transformation: the CIO, advanced analytics program managers and full-time product owners of AI teams. But eventually it should spillover and extend to analysts, consultants, architects, mid-level management, etc.

So what are the set of activities we expect translators to perform, what skills do your employees need to master?

• Identify use cases — articulate the advanced analytics proposition and identify use case opportunities. Translate business problems into advanced analytics solutions.
• Manage the data product life cycle — establish and manage the life cycle for data science products (Fig. 1), agile processes, teams and IT infrastructure for the experimentation and industrialization phases.
• Challenge and communicate results — measure and challenge data science results and communicate them to business stakeholders, using a basic understanding of data science and data engineering techniques.

Fig 1. The data product life cycle, characterized by three distinct phases: ideation, experimentation and industrialization

With this blend of data strategy, agile project management and technical knowledge, translators are able to anticipate the most prevalent showstoppers in AI projects, and identify impactful ways forward.

Of course, we do not expect translators to be technical experts themselves. That’s why they excel when flanked by a lead data scientist and/or data engineer to help quickly assess use case feasibility, coach the scientists and engineers who are doing the work and create the technology roadmaps that bring an organization towards maturity.

Stay tuned for more blogs on data translation and strategy.

BigData Republic is hosting an analytics translator training 24–26 June. For more information: see here. In addition, we offer lead data scientist and lead data engineer roles.