Retail, in data we must trust!
Consumer behaviour has changed — Machine Learning can help.
The world is a different place now compared to this time last year; the level and speed of change experienced in the last 11 months was unprecedented.
Within the retail sector, the physical shopping experience has changed, potentially forever; with customer numbers limited per shop, the requirement to wear masks while in-store, the need to maintain a distance of 2m between other customers, and the sharp rise in cashless payments. The online experience has dramatically changed as well, mainly driven by new consumer behaviours e.g. bulk buying of ‘essential’ household items and increased ‘spread ordering’ (multiple purchases of the same item in different sizes or colours).
These have given rise to new challenges and issues within retail businesses; decreased footfall, reduced ROI of physical retail outlets, stock-outs and supply issues, inaccurate demand forecasting, cash flow forecasting inaccuracies and costly increases in logistics, to name a few. Businesses are unable to rely on their tried and tested ways of doing things and are having to turn to data and machine learning to overcome these challenges by increasing the level of data-driven decision making in their business.
For many, this has accelerated a journey they were already on; however, for others, it has meant starting something new. In both instances, this can be daunting. Business executives and leaders are keen to see a return on investment with solutions delivering operational improvements day in day out, and so selecting the right use case/opportunity to focus on is key.
Choosing the right opportunities to pursue is key.
AI has many applications. However, it is important to pursue those with the optimal mix of value and viability, balancing tactical quick wins with longer-term strategic capability development. Some key AI use cases that DataSparQ are discussing with their customers in the retail sector include:
- Reducing stock-outs, waste and improving availability/optimising working capital through improved demand forecasting
- Reducing costs and improving customer satisfaction by optimising vehicle fleets, delivery routing and final mile logistics
- Increasing basket size by improving product recommendations to make more personalised customer recommendations
- Improving customer retention by identifying customers with a high propensity to churn and recommending retention strategies/actions
The size of the prize/return on investment can be significant. For example, we have seen multi-million pound annual benefits being delivered through fleet and routing optimisation, and 10x ROI delivered through improvements in customer retention, as a result of successfully operationalising AI.
In order to identify and select the right opportunities, you should start by analysing the end-to-end business value chain to uncover opportunities and identify solutions that will maximise ROI. Then you can prioritise solutions by exploring and estimating value vs delivery viability. This is exactly what we do in our SparQshops, a collaborative process that identifies high-value applications of data and AI and helps kick start your AI journey. Why not access DataSparQ’s AI starter pack to help get you started. We also encourage our clients to consider 5 key AI qualifying questions.
- How will you work faster, cheaper or better?
- What could 1,000 interns achieve?
- What would you ask a crystal ball?
- What's the cost of getting it wrong?
- Who is going to enjoy using it?
You can find more information here: Could AI solve your business problem?
Getting the solution operational and maintaining it effectively is key to maximising value.
As well as coming up with the right idea it is equally important to focus on successfully operationalising the solution so that it will continually deliver value every day. In our experience there are typically three reasons why great AI ideas don’t make it into production:
- Data quality/availability — more often than not the underlying quality of an organisations data is poor and/or there isn’t the right level of senior sponsorship to ensure the wider business coordinates access to the required data sources
- Building in silos — technology solutions can rarely be deployed without impacting people and process, and this is no different for AI solutions. Organisations often overlook the change needed within the wider business (operations, roles and processes) to ensure AI solutions can be integrated and adopted successfully
- High maintenance costs/model degradation — data scientists, while experts in developing models, often lack the engineering expertise to build a reliable, resilient and scalable model that remains accurate and continues to run cost-effectively
For more information on why Data Science initiatives fail read our blog: 3 reasons you’re probably stuck in pilot purgatory
As many organisations have experienced recently, unforeseen events, like the current global pandemic, expose AI and machine learning solutions that have not been re-trained to take account of new trading conditions. One of our engineers explores this further here: Why your ML model is broken.
One way to overcome this is to build automated retraining architectures that pick up change points and include or discard certain historical training data as a result. However, autoscaling architectures can be hard to implement, and often rack up very high costs, as the underlying cloud infrastructure is not used efficiently. It is therefore important to have a well-configured, demand responsive setup which requires a combination of expert engineering skills and data science.
That is what we bring here at DataSparQ; data science and data engineering expertise to design, build, deploy and run operational AI solutions that deliver value every day, long after they have gone live.
If you are exploring how AI can benefit your organisation or looking to accelerate in-flight/stalling AI initiatives then get in touch. We would be happy to share our learnings and see if we can help.
