The rise of augmented analytics
Augmented analytics is one of the trends set by Gartner this year. In an organisation, it acts as a bridge between data analytics and business intelligence. Augmented analytics helps to speed up the decision-making process, thus improving the transparency of the business value chain. The use of augmented analytics is best for industries that want to automate their supply chain, end to end. I highly recommend companies that are already savvy in automation to think about adapting the concepts of augmented analytics. It would add value to the business in long run.
2019 Technology Trend:
Using augmented analytics,
- Automates the processes between machine learning (ML) models with business intelligence in a company
- Increases transparency by minimising the knowledge gap between technical and business people
- Enhances productivity by delivering advanced insights in real-time from operators to executives of an organisation
- Improves agility by interacting data scientists and citizen data scientists to make quick decisions dynamically
- Enables business teams to generate queries, explore data, and receive and act on insights in natural language (voice or text) via mobile devices or personal assistants (AI enabled)
The evolution of augmented analytics
Any organisation is capable of driving the concepts of augmented analytics. However, the starting point depends on where it currently stands. The maturity can be described as follows (Gartner):
The role of citizen data scientists
With augmented analytics the role of “citizen data scientist” becomes popular and plays a key part. They may be seen as translators who connect the business and technical worlds. They have knowledge of business acumen and data analytics. At present, data scientists are capable of handling machine learning algorithms. However, they have to communicate to business teams to ascertain whether the outcomes from machine learning models are appropriate for the business.
Data scientists love data. They used to work on data without always having a clear hypothesis of what problems they could solve using the given data. However, this mindset is now changing. I have noticed that data scientists have started to define an initial hypothesis. Then based on that, they leverage data to verify those predefined hypotheses. This provides quick wins for businesses. Citizen data scientists can engage with data scientists earlier in the process and help them to improve the decision-making process.
Business case
Let’s see how we can use augmented analytics in predictive maintenance (Pdm) cases. Predictive maintenance is used to minimise the unplanned downtimes of machinery equipment. It can also help you to run your assets more efficiently and to minimise the breakage and labour costs to fix them. The business case may be divided into two parts.
- Data science – activities carried out by data scientists
- The decision making process – decisions involved with outcomes of machine learning models which are decided by business teams
Data science
Imagine a case where we have to develop a predictive model to minimise the unplanned downtimes of machines on a factory floor. We need to:
1) Identify the equipment that contributes to the production loss. Take for example, a conveyor belt in a production line. If the conveyor belt goes unexpectedly down, you will lose production.
2) Determine the maintenance costs for each equipment category (PM01 – preventive maintenance, PM02 – breakdown maintenance, PM03 – general maintenance etc.) in the factory.
3) Prioritise which equipment is needed to run the predictive maintenance based on the impact and the relevance to the business.
Impact: It is a good idea work in conjunction with the business teams, as they know what is most relevant from the business point of view in terms of impact ($ value). As an example, running a Pdm model on conveyor belts is more valuable to the business than the fans on the factory floor.
Relevance: This is measured by data availability. It can be measured in terms of granularity, quality and accuracy of your available data.
Predictive maintenance score matrix:
4. Select the right equipment to implement predictive maintenance based on the predictive maintenance score matrix. As an example, if you have several conveyor belts on your factory floor choose the one which has the highest impact and valid data.
5. Find the reasons behind the failure of the belts (root cause analysis). For a conveyor belt, it could be due to many reasons such as vibration, load variation, cleaning performance etc.
6. Develop the machine learning models as a proof of concept (POC) or as a pilot project.
7. Integrate machine learning models on the factory floor to obtain real-time insights.
8. Finally, scale up these models to other conveyor belts in the factory.
Predictive maintenance modelling flow
Modelling
- Develop a ML model using algorithms such as regressions, random forest, NN or DL etc
- Develop an expert model which includes rules to optimise ML models based on the opinions from experts
- Develop the optimised model which combines the expert and ML models and then fine-tune them
Modelling activities:
The typical flow of predictive maintenance modelling is explained above.
Decision making
Consider the case where your ML models predict that one of your machines is going to fail in the next few days. You need to make sure that you have enough spare parts and resources to replace it. You have to involve different stakeholders in order to make decisions. Sometimes it is time consuming. If your organisation is very hierarchical, it could become even more time consuming. With the power of augmented analytics, we are able to automate this process. Here are some of the decisions we need to make with the people who are involved with the process, along with some complications.
Stakeholder engagement:
Operators:
Identify the location of the equipment (which floor, which area etc)
Check the root cause and the running condition of the machine, which is predicted by ML models to become troublesome
Make sure you have spare parts if necessary
Check the inventory
Allocate resources (manpower)
Inform upstream and downstream owners of the production line of planned downtimes of the machine
Ensure there is no effect on production during the downtime
Business team
Check the budget for a new machine
Allocate additional personal
Make sure that the planned downtime of the machine causes minimum impact on production (revenue)
Complications
There are no additional spare parts available in the inventory
There is no additional team to run the maintenance; you need to hire subcontractors
There is no budget for a new machine or no manpower to fix the machine
Time taken to approve the budget
The big picture
The figure above shows the use of augmented analytics which apply in a predictive maintenance case. However, it is not an exhaustive list. It is used to deliver the message at a high-level.
The use of augmented analytics bridges the gap between data science and business decision making, which can be fully automated. The other advantage is to use SMS or email to alert corresponding owners in the value chain. I have seen it becoming very popular in finance, retail, telco and manufacturing sectors. Concepts of augmented analytics are emerging in sectors like the automobile industry, insurance, logistics, and pharmaceuticals. However, mining, oil and gas and agriculture are still lagging behind.
We run augmented analytics readiness assessments of an organisation.
If you want to know more, feel free to reach out Dr Kash Sirinanda, founder of Elite Futurists.
Reference
https://www.gartner.com/smarterwithgartner/gartner-top-10-strategic-technology-trends-for-2019/
