Charles Wright

Head of Technology Solutions

  • Data
  • RPA
  • AI
  • Robotics
  • Machine Learning

23 Jul, 2021

Preparing for your first Machine Learning Use Case

Machine learning promises significant cost savings, efficiencies, and new services. Yet it remains something of a ‘leap into the unknown’ for many organisations looking to take their first step.

Gartner found that 85% of machine learning projects will fail, a stat that is predicted to continue. This should not come as a surprise, Not only is the subject area complex, but implementing machine learning demands a variety of other technologies and skills aside from the raw code.

Often the media's depiction of artificial intelligence and machine learning is simplified, focusing only on the task at hand - overlooking the essential factors which need to sit around the tech for it to work effectively. Bearing this in mind, what is the first step in the journey?

Building a team

The biggest reason that machine learning projects fail to get off the ground is the ability to identify and articulate the requirements. Ensure you have the right people available to identify a full breadth of use cases, to prioritise which to move forward with, and then gaining buy-in for those ideas from leadership.

Identifying use cases requires a balance of knowledge between understanding:

  • the organisation, how it operates and where tasks performed by machine learning could apply
  • the wider components that make up a Data or AI Strategy
  • the technologies used across an end-to-end machine learning project
  • machine learning itself, the feasibility of applying machine learning successfully to potential use cases and what algorithms are best

Often organisations will aim to fill this requirement via appointing a position such as a Head of Data Science. However, skill sets are often concentrated in one or two of the above facets, leading to shallow discovery and planning. Additionally, these individuals tend to have limited experience performing this type of discovery activity, as it is infrequently requested by organisations (but frequently needed!).

Therefore, a common option is often to look for temporary resourcing to perform the discovery and selection of use cases. Contractors or consultants that have direct experience performing this type of task for multiple clients can pull the required breadth of information out of an organisation’s employees extremely efficiently, build a solid business case and then gain buy-in from executives.

Here I explore four areas of human capital required for machine learning use-case discovery in more depth:

Understanding the organisation

Organisations and industries universally present opportunities for certain tasks best addressed by machine learning: sales forecasting, customer segmentation and predictive maintenance are often prominent examples, among many others. Turning these tasks into use cases is not just limited to the technology. Identifying everything around the task such as data quality or the organisations capability to be able to use the outputs of machine learning is essential.

However, as organisations grow, they build functions and business units which do not confirm to a standard value chain model. They often grow these new areas in silos, isolating knowledge, tools, and data. Without central governance, choices are made that do not consider wider needs and processes are designed without the full power that the wider organisation can bring. This frequently provides many opportunities for smaller and unique use cases of automation and analytics. You must understand the potential of machine learning demonstrated in the wider market and map this to the data ecosystem and opportunities within an organisation.

Data & AI Strategy

The reality is that machine learning implementation should consider a variety of factors including:

  • Technology & Processes
  • Operating Model & People
  • Governance, Ethics & Bias
  • Security

If you fail to address any one of these components following implementation projects may fail. For example, if I do not consider the cyber security implications of moving data into the cloud for processing, or of the solution as a whole, the machine learning implementation may never be signed off in the first place. Similarly, if I do not consider an operating model, I will not be able to maintain an implementation and so it goes on.

Proficiency in Technology

Machine learning is not just Python code, a machine learning solution is closer to a Cloud data platform. Typically, while the machine learning model itself is being developed on data extracts, there is a large amount of work in parallel to build automated data pipelines, sucking data into cloud storage and automating the processing to prepare it to feed a model. The model should be hosted on cost and performance optimised infrastructure and the results output into a BI solution or web interface to allow for visualisation of the results and performance monitoring.

All these technologies must be clearly understood and planned for before embarking, not only to ensure baseline maturity is there to enable implementation. Failing to consider the wider technology ecosystem bloats implementation timelines causing challenges with data integration, planning, and reacting on the fly to new information and difficult governance.

Machine Learning Expertise

Research in machine learning, in particular Computer Vision and Natural Language Processing is rapidly progressing. Six months can leave you significantly behind the curve, applying outdated algorithms with significant differences in results.

Algorithm selection does not just define the performance of machine learning, it also determines demand for data, infrastructure requirements for training and inference, maintenance, and agility. Something about staying abreast, latest understanding...

How do I confirm the team has the right skills?

Whether you are looking for one person or a full team it is imperative that you evidence capabilities to consider the organisation’s needs, its strategy, the wider technology ecosystem, and the breadth of machine learning experience. Covering each of these areas should provide confidence that your organisations Machine Learning programme will have a solid foundation to begin.

Charles Wright

Head of Technology Solutions

Chaucer's AI specialist delivering data strategies and capabilities for Fortune 500 organisations. He is passionate about driving data led digital transformation to enable organisations to realise the benefits of machine learning and holds both an MBA and MA in Educational Leadership and Management.

Blog 23 Jul, 2021

Data Strategy, Data Science & Analytics, Data Visualisation, Data, Data Ethics

Five risks of getting ‘return to work’ wrong and how to avoid them

23 Jul, 2021

Elodie De Fontenay

Insight Partner – Data & AI

Blog 06 Jul, 2021

Privacy, Cyber Security, Data, Ransomware, Information Security

Handling a ransomware data extortion attack

06 Jul, 2021