Making decisions with data – for machine learning success, follow the lessons learned in embedded analytics

Bringing together data, analytics and automation, machine learning can help improve your chances of success and make everyone more productive.

Machine learning has been touted as a huge boon for businesses. Research by Infosys found that companies anticipate a 39 per cent boost to their revenues on average by 2020, from their investments in Artificial Intelligence and Machine Learning. 

The prospects for machine learning involve adding smarter processes into applications that are used every day, in an effort to reduce the amount of human intervention required to help individuals make better and faster decisions. Bringing together data, analytics and automation, machine learning can help improve your chances of success and make everyone more productive. At least, that is the theory. 

In the real world, it will take a while for these applications to be developed and deployed. However, both the will and the investment to make this happen are growing. According to a Forrester survey, there will be more than a 300 per cent increase in investment in cognitive computing in 2017, compared with 2016. Companies are investing in areas such as the Internet of Things to create more data that can then be analysed and consumed alongside data from their business applications.

For the business to see benefit from machine learning, IT teams have to focus on what existing employees can use data for, and how automation capabilities can wrap more value around that data over time. Currently, providing more insight around a customer segment is the first step for analytics. However, machine learning can be applied to that customer segment to see what purchases were made, what timescales were involved and what results might be expected. Using this learning activity, analytics can not only provide some insight, but also prescribe next steps for employees to take.

Lessons to learn

In the example of sales, this might involve pitching specific products or making certain offers that have a greater chance of success. In logistics and supply chain, it might involve structuring deals so that the customer gets deliveries quicker and the business can manage inventories and assets more efficiently over time. These incremental improvements can then be balanced against any wider digital transformation programmes in which the business can utilise machine learning over time.

Now, this market is still very new. While machine learning tools and frameworks exist, they have not been implemented widely, and the number of people with the right skills is low. However, there are other approaches that can help. For IT teams that want to take advantage of their data and machine learning together, lessons can be learned from embedded BI and analytics implementations. 

Embedded BI refers to how companies can put analytics into their business and then provide either applications or services that are based on the results. These analytics tools can be used to differentiate a company’s products or services against competitors, and they can enable that company to make more money based on the data it holds. With embedded analytics, companies have to implement for customers quickly and with specific goals in mind. 

How does embedding analytics work in practice?

Embedded analytics projects are different compared to more traditional BI or analytics implementations. Rather than looking at internal customers and their requirements, embedded analytics services are aimed at external customers that will normally have less knowledge of how to work with data. What they do want is a service that can help them quickly, provide more insight than they can get on their own, and that they can use without training.

For the team implementing a project like this, the fundamental aims should be to provide tools that can explain their approaches as they go and avoid introducing problems. Alongside this, the project has to solve real business pain points so that the service provides value from day one. For those looking at machine learning implementations, these same aims should be in place from the start.


Let’s look at an example – providing data and analytics tools on a data set such as travel and expenses. For HR staff, sorting through all the data to spot patterns is both a chore and a technical challenge. Providing the tools to automate this process is a good value-added service opportunity for the application provider. By building dashboards and analytic tools that can be shared within the application, the vendor can offer something that is more useful to the business team.

However, just making existing data look prettier is not the foundation for a successful long-term service offering. Instead, those analytics tools have to be rich enough that people can ask their own questions and see how the results are put together. This can then aid collaboration for the HR team with other departments that would be interested in the results.

The opportunity for machine learning takes this further. By looking at the data sets over time and what kinds of results are most valued, machine learning systems can be trained to look for patterns that should be useful to the customer and may be more difficult for an individual analyst to identify. These systems can then provide updates automatically when specific patterns are spotted.

This automation stage can extend embedded projects and make staff more efficient. However, one important element is how that data is shared. If the results are simple reports based on PDFs or spreadsheets, then the ability to share how the insight is generated is reduced. Embedding analytics into a product can make it easier to bring others directly into the data and the results, so they can interact with the data by themselves. 

In some respects, this follows on from how we all learn mathematics at school. Even if we can get the right result, it’s just as important to show how you achieved the result so that others can follow your thought processes. For embedded analytics, showing how results were reached can be useful for all those who might need them. When machine learning is involved as well, the ability to see how a result is achieved with all its history and lineage is going to be more important than ever.

Designing for the business

Supporting the future growth of machine learning can take some lessons from embedded analytics projects. The most important of these is how to execute projects based on specific business value, rather than for potential insights that might come in the future.

As an example, many companies have made the move to adopt big data technologies such as Hadoop. Storing data at scale is far easier today compared to the past, so saving information “just in case” it might prove valuable in the future is an understandable reaction. However, this approach ignores how difficult it is to get value out of truly huge data sets that are not suitably structured or available back to the business. 

Relying on data scientists to find insights buried in this morass of data is therefore not a guarantee that those insights will be found, let alone that they will justify the cost of technology or staff time. The same is also true of machine learning – just implementing new technologies so that the data can be analysed is not enough. The risk with this approach is that expensive staff time is dedicated to trying to create value from nothing, rather than looking at how to optimise value around current approaches.

Instead, it is worth looking at how the business works today and where automation can help improve productivity. This incremental gain can deliver a result more quickly and provide a greater return on investment. 

At the same time, automation frees data scientists to work on bigger issues, and think of how data can be used to solve those issues. This may lead to complete changes in processes over time, but it does not get in the way of getting some quick wins through automation. The most important consideration is that this is not a question of “either/or” when it comes to approach around machine learning, but both.

Machine learning has a huge amount of potential. However, it has to be understood in context. It won’t solve all business problems, and it won’t suddenly turn poor ideas into good ones overnight. What it can do is enable faster and more accurate decision-making and help create more opportunities for success. Using the lessons of embedded analytics projects, companies can start planning how best to make use of machine learning in their own ways.

Pedro Arellano, Vice President, Product Strategy at Birst
Image Credit: Pitney Bowes Software