Are you using advanced analytics, AI, or machine learning to power your business performance? If not…perhaps you should be! But where do you begin?
That was the focus of our webinar on 19 August.
And that’s what we’ll focus on in this blog where we’ll share two use cases with wide applicability that can add value to your data:
1.USE CASE #1 – MATCHING RESOURCES TO DEMAND USING TIME-SERIES FORECASTING
2. USE CASE #2 – SEGMENTING CUSTOMERS USING UNSUPERVISED MACHINE LEARNING
Cafes, restaurants, shops, libraries, call-centres, emergency services and more, all have one thing in common – they need to ensure they have the right staff available, at the right time, to meet demand. Think of staff rotas and matching staff levels to customer levels. Getting it wrong, and having too few staff, means poor service levels & missed customer opportunities. Alternatively, bringing in too many staff can result in a higher cost, and lower profit, than expected. Balancing supply and demand is a common problem that challenges many organisations across the world.
So how can data help? You can use data analytics to drill down into the day-by-day characteristics of demand (eg day of the week, pay-day, bank holiday, seasonality) in order to improve the day-to-day resource allocation. We recommend an application of modern time-series techniques.
Think of a rota where you want to be able to allocate shifts up to two weeks ahead…
Firstly, you can use a time-series model trained on daily historical data. This can allow for:
- Trends: discounting longer-term trends, and keeping short-term trends
- Known past/future influences: day-of-the-week or month, bank holidays, school holidays, Christmas, etc
- Known past & unknown future influences: for instance, weather is a ‘signal’ in the past that becomes ‘noise’ in the future
Using this approach, you can reasonably forecast trends with a level of uncertainty.
This is a useful start. But quantifying future uncertainty is crucial for decision support.
So secondly, you can build on the time-series model using decision support to minimize negative consequences. We can use statistical decision theory to make optimal recommendations according to assessment of the consequences of getting the supply wrong. For example, it could be far worse to have too few staff in a busy shift, than to have too many staff in a quiet one.
If our forecast includes uncertainty, we can factor this asymmetry of consequences into the choice of shift size to make a smarter, business-focused decision on staff rotas.
What to find out how to do this? Watch our webinar recording or contact hello@data-cubed.co.uk
By helping organisations to better match supply to demand, you can ultimately improve efficiency, staff welfare, service levels and profit.
Most organisations have many users, customers, or clients and they don’t want to issue a one-size-fits-all approach to them. That’s where customer segmentation comes in – grouping similar types of customers together. Getting this wrong means missing out on valuable customer insight. Plus the potential issuance of non-relevant communications and services to customers causing frustration and potentially missing out on commercial opportunities.
It’s tricky to segment customers with irregular records, such as purchases: the standard methods of cluster analysis (which can also be very useful) don’t apply. Clustering methods on bipartite graphs can address this problem, and potentially give you new insights into what types of customer you have.
Suppose your supermarket has two types of customer:
a. Buys mainly food and toiletries
b. Buys mainly toiletries and booze & fags
If you knew this and organized your records, then they might look like this – with well-organised customer groupings.
But, in practice, they’re going to look a lot more like this – with no clear organisation of customers.
By using machine learning, partitioning a bipartite graph can reorder your customer records like this – to create well-organised customer groupings.
What to find out how to do this in practice? Watch our webinar recording or contact hello@data-cubed.co.uk
This approach will give you new insights into the types of customer you have and the capability to deliver targeted, relevant comms and services.
1. Become a leaner, stronger organisation
2. Get ahead of the competition and create a commercial advantage
3. Make or save money
Want to know more about these two use cases? Check out our accompanying webinar here or contact us at hello@data-cubed.co.uk
Helen is a data fanatic. Born in the corporate world, she honed her skills at global corporate giants, including AXA, before setting up Data3, the UK data lab. Helen and her team help businesses get closer to their customers, make smarter business decisions and make & save money using data.
Are you using advanced analytics, AI, or machine learning to power your business performance? If not…perhaps you should be! But where do you begin?
That was the focus of our webinar on 19 August.
And that’s what we’ll focus on in this blog where we’ll share two use cases with wide applicability that can add value to your data:
1.USE CASE #1 – MATCHING RESOURCES TO DEMAND USING TIME-SERIES FORECASTING
2. USE CASE #2 – SEGMENTING CUSTOMERS USING UNSUPERVISED MACHINE LEARNING
Cafes, restaurants, shops, libraries, call-centres, emergency services and more, all have one thing in common – they need to ensure they have the right staff available, at the right time, to meet demand. Think of staff rotas and matching staff levels to customer levels. Getting it wrong, and having too few staff, means poor service levels & missed customer opportunities. Alternatively, bringing in too many staff can result in a higher cost, and lower profit, than expected. Balancing supply and demand is a common problem that challenges many organisations across the world.
So how can data help? You can use data analytics to drill down into the day-by-day characteristics of demand (eg day of the week, pay-day, bank holiday, seasonality) in order to improve the day-to-day resource allocation. We recommend an application of modern time-series techniques.
Think of a rota where you want to be able to allocate shifts up to two weeks ahead…
Firstly, you can use a time-series model trained on daily historical data. This can allow for:
- Trends: discounting longer-term trends, and keeping short-term trends
- Known past/future influences: day-of-the-week or month, bank holidays, school holidays, Christmas, etc
- Known past & unknown future influences: for instance, weather is a ‘signal’ in the past that becomes ‘noise’ in the future
Using this approach, you can reasonably forecast trends with a level of uncertainty.
This is a useful start. But quantifying future uncertainty is crucial for decision support.
So secondly, you can build on the time-series model using decision support to minimize negative consequences. We can use statistical decision theory to make optimal recommendations according to assessment of the consequences of getting the supply wrong. For example, it could be far worse to have too few staff in a busy shift, than to have too many staff in a quiet one.
If our forecast includes uncertainty, we can factor this asymmetry of consequences into the choice of shift size to make a smarter, business-focused decision on staff rotas.
What to find out how to do this? Watch our webinar recording or contact hello@data-cubed.co.uk
By helping organisations to better match supply to demand, you can ultimately improve efficiency, staff welfare, service levels and profit.
Most organisations have many users, customers, or clients and they don’t want to issue a one-size-fits-all approach to them. That’s where customer segmentation comes in – grouping similar types of customers together. Getting this wrong means missing out on valuable customer insight. Plus the potential issuance of non-relevant communications and services to customers causing frustration and potentially missing out on commercial opportunities.
It’s tricky to segment customers with irregular records, such as purchases: the standard methods of cluster analysis (which can also be very useful) don’t apply. Clustering methods on bipartite graphs can address this problem, and potentially give you new insights into what types of customer you have.
Suppose your supermarket has two types of customer:
a. Buys mainly food and toiletries
b. Buys mainly toiletries and booze & fags
If you knew this and organized your records, then they might look like this – with well-organised customer groupings.
But, in practice, they’re going to look a lot more like this – with no clear organisation of customers.
By using machine learning, partitioning a bipartite graph can reorder your customer records like this – to create well-organised customer groupings.
What to find out how to do this in practice? Watch our webinar recording or contact hello@data-cubed.co.uk
This approach will give you new insights into the types of customer you have and the capability to deliver targeted, relevant comms and services.
1. Become a leaner, stronger organisation
2. Get ahead of the competition and create a commercial advantage
3. Make or save money
Want to know more about these two use cases? Check out our accompanying webinar here or contact us at hello@datacubed.nz
Helen is a data fanatic. Born in the corporate world, she honed her skills at global corporate giants, including AXA, before setting up Data3, the UK data lab. Helen and her team help businesses get closer to their customers, make smarter business decisions and make & save money using data.
Jonty is a consulting statistician who was until recently Professor of Statistical Science and Head of Statistics at the University of Bristol. He has worked at all levels, from charities and social enterprises up to large PLCs, and with several public sector organisations. His teaching and research cover both the theory and the practice of statistics and machine learning.