- Augmented Analytics
- Big Data Analytics
- Business Analytics
- Data Governance & Data Fabric Architecture
- Data-Driven Strategy & Transformation
- Performance Management & Improvement
- About us
Advanced analytics (i.e. Artificial Intelligence and Machine Learning) is one of those areas that retailers know that they can’t ignore. Some are on their second or third generation of deployment, others are still getting to grips with their first and yet more are wondering where and how to start.
Dec 12, 2019
But “doing” advanced analytics and gaining advantage from it are not one and the same thing. A number of retailers have deployments in place that are delivering positive results but this is an area that needs to constantly evolve and move seamlessly from prototype to production on an ongoing basis.
A newcomer to the topic could be forgiven for thinking that advanced analytics starts and finishes with writing code to create the best model to, for example, match customers to products they will like, thereby ensuring competitive advantage. Today, however, organisations of all sizes have low-cost access to the software and infrastructure required to perform what would have been extremely specialised analysis on expensive equipment just a few years ago. Cloud platform providers have re-written the commercial rules around access to computing power and the vast majority of the algorithms required are readily available and royalty-free. In addition to this, there is a plethora of books and online material that anybody with a modest amount of technical knowledge can use to utilise advanced analytics models.
Why then, do we frequently hear from our clients that they are not achieving the benefits that they expect from their analytics projects?
The answer is that success in this area requires more than algorithms. More than 80% of the time that a data scientist spends on a successful project has nothing to do with an algorithm: it involves liaising with business domain stakeholders; gathering, transforming and processing data; planning how a prototype project can be deployed in production; ensuring it enables actions to be taken that will have a positive and tangible impact upon the business and determining how that production model can be monitored for effectiveness, then updated as circumstances change.
Many of these are far from trivial tasks.
Above all, the process starts with the key stakeholder defining the business outcomes required, and the value of those outcomes. AI and ML can be applied to a large number of business problems but they are not, themselves, capable of determining what problem they should solve; or whether a problem is a priority for the business.
Why determine the value of the outcome? Because advanced analytics has the capacity to consume significant resources on an ongoing basis and it is often the case that “good enough” yields a better business return than “better”. Like any project, the benefits have to justify the effort.
The key takeaway here is that AI and ML projects should not be considered as solely technical exercises but rather joint undertakings by various business and technical functions in order to ensure success. Success should also be measured in terms of the measurable impact upon the business and for this to be realised, projects need to successfully transition from the lab into real-world production environments - often requiring a different set of skills to the original development.
Steve Crosson Smith
SDG Group UK