Data and analytics leaders are investing heavily in analytics, machine learning, data science, AI and other related technologies. The belief is that such technologies can drive better decision making and thus business outcomes. However, there is a real gap between this belief and the notion of what a decision really is. The data and Analytics gap between building models and getting the most out of them is a “missing link.”. McKinsey reported that less than 10 percent of Analytic Models that are developed actually make it to production where they can deliver ROI.
We already know that building a test bed for analytical models using the latest technology is straightforward. However, when it comes to actually placing these models into production so that we can start reaping the benefits, that’s a totally different story. In fact, we can see models that are built and trained in couple of weeks will take months to be put into production, if at all!
Putting a model into production is not a matter of copy and pasting the Python, R scripts… to other machines, servers or environments. It requires managing integration points, adding behaviour to real-time systems and processes and transforming the input and output of algorithms in order to be useful in production.
On top of that, it also requires the skills and ability to understand business decisions and their impact. It involves integrating the models into operational decisions and combining them with domain expertise rules.
The data and Analytics Gap is not only about the technology, but, it is about the mindset as well.
Productionizing of Data Scientists Models
There is no point in building a data and analytics algorithm that does not improve or has no impact on the operational decisions of an organisation. The end goal of these models is to improve the quality of an organisation’s operational decisions and also to enable organisations to make batter and smarter decisions. But if we don’t know what these decisions are and how they impact the current situation of an organisation, how can we improve and re-engineer them after all?
Either improving or radically re-engineering these require an understanding of the operational decisions involved in those scenarios in the first place, as well as understanding and modeling the current situation in order to be able to measure the impact of new ones. Also of course, this means putting the the operational decisions into practice by building business rules around them. And certainly ensuring ease of use of the models with domain experts rather than just data scientists.
The data and Analytics Gap that exists at this juncture is due to the fact that companies do not invest nearly enough in understanding, modeling and executing operational decisions in a systematic and consistent way
Model Deployment and Management
For companies, trying to implement data and analytics efforts is a big challenge due to the fact that the management and deployment of models is a hand-off to IT. Because IT systems are complex, they are designed in certain ways and tested and deployed within environments that are not necessarily ready to intervene. Deployment and management of data and analytics models requires a “link” between the models and IT systems and processes. Therefore, the models must be versioned, tested with different and multiple data sources, and be measured based on business decision KPIs. Also, to use them effectively, they need to become part of the automation landscape of the enterprise and be supported and integrated by the IT team.
The data and Analytics Gap at this stage exists because companies do not utilise right tooling to put the models in practice as part of the real-time and day-to-day operation of the business.
Putting the Data and Analytics Results into Action
The last piece of the data and analytics gap exists due to the fact that the results will be left unactioned on the dashboard or other systems as a suggestion or recommendation. Untraced, unmeasured and unactioned.
The data and Analytics Gap at this point is due to the fact that those algorithms are not the arms and hands to do the actual job.
The overall integration and orchestration should actually put them into action. Taking actions based on the results is not always straightforward. That’s why they are left out of the picture. Yet that’s the last mile that must be taken in order to ensure they are having an effect on the organisation. This requires data integration, system and process orchestration, decision robotics to interact with applications, as well as user-interfaces (UI) and sometimes human intervention. Yet the data and analytics job is not done until actions are taken.
There are several developments required in order to realize the full potential of these analytical models, and in so doing give them a viable and integral role in day-to-day decision-making processes.
- Understanding and modelling the current decisions within workflows and systems
- Understanding how the decisions are measured and how they impact and support business values
- Integrating multiple versions of the data and analytics models into different operational decisions
- Simple and flexible integration of data and analytics models into real-time workflows and production systems
- Utilising data and analytics models as part of operational decisions and modelling business rules based on these
- Separating the models from data to allow different and multiple data sources to flow through the model for different scenarios
- An overall end-to-end model that not only take cares of data and system integration as well as decision execution, but also takes actions based on results
It is critical that any model developed by data scientists is presented to IT in a way that is consistent with accepted best practices for technological production processes in a scalable way. This is the only way that even the best models can become integrated into the real time workflows that are pertinent to every day decision-making in a business environment.