Are you considering data-driven solutions to improve or create new business outcomes, and/or business models? Do you need to consider how such a solution would fit within your infrastructure? As I architect solutions at Arundo, I’m often asked to address similar concerns that our customers have early in the engagement. Here I’ll present a few of these concerns along with my thoughts.
I want data-driven insights, where do I start?
I often get this question from business leaders with profit and loss responsibilities who aren't involved in day-to-day equipment maintenance. I say you should focus on the specific business objectives you can meet through data-driven initiatives. Starting with a business objective and formulating a hypothesis will help drive an ROI-focused discussion. This discussion may include conducting a use case workshop with your subject matter experts and maintenance leads. In this workshop, you should build hypotheses, determine expected ROI, and run data availability/feasibility analyses.
You should engage in a problem with a measurable financial impact and gain comfort by remaining engaged with the technical experts driving the data-driven initiatives. By building confidence in the process, you can take advantage of quick wins and build momentum for additional use cases. Add points on a digital scoreboard to tackle higher value use cases, which can require greater investment to deliver.
Data availability and quality often appear as roadblocks to data-driven initiatives. However, you should approach these concerns in a prioritized use case driven approach. Acknowledging the fact that, in general, data availability and data quality can be challenging. Dig deeper into identifying and addressing it in a methodical fashion, and prioritize your larger data infrastructure initiatives based on data-driven use case requirements. Moreover, some data science techniques do not always require pre-existing historical data and failure information.
Should I build or buy?
Another common question I get is if you should build or buy the solution to deliver data-driven initiatives at scale across the enterprise. There's no definitive answer. It depends on where you are in your larger enterprise architecture strategy. Which of these IT strategies will allow your business achieving its goals at a faster pace? Which strategy will enable the following benefits across your enterprise:
- reduced IT costs
- increased IT responsiveness and management satisfaction
- improved risk management, and
- better strategic business outcomes?
You should weigh your response heavily on speed-to-value. Platform-as-a-Service (PaaS) / Infrastructure-as-a-Service (IaaS) components are key in time to market (speed) when building a new product. They may allow you to focus on your key strategic objectives, instead of diverting resources and attention to building IT capabilities where they don’t align. In addition, you should carefully scope a curated use case and deliver speed-to-value through rapidly prototyping engagements with key business units. This "bottom-up" approach can help build momentum towards a wider discussion and approach to technology and methods in the rest of your organization.
How do I integrate data-driven solutions in a sustainable way?
This is another question I often get. My answer is that you need to be clear on how the output from the data-driven solution is "consumed" by your end-users, and how you should roll out the solution to actually get end-users to take actions from the insights. Make sure the solution is generating value for your business on a consistent basis.
No matter how accurate and timely a prediction, analysis, or insight is, you need to think about how to drive action from the output in the best way. For example, being able to quickly line up a root-cause analysis workflow following an alert, is that the most appropriate next step? In addition, how do you build trust in what the solution is telling you within your organization? This is especially critical for data-driven methods because often, the solution is modeling complex systems, where the insights aren’t always easily explainable.
The solution must incorporate both the technical integration into the customer’s operational technology (OT) / information technology (IT) infrastructure and the human factor.
The technical integration will answer questions such as:
- Will the output be integrated into existing plant operations and infrastructure, and if so, where? or,
- Will it be separate, for example, in the form of a new operator display, along with the necessary output to guide the end-user for root cause analysis?
The human factor will answer questions helping explain the data-driven model’s output. The resulting model has to be interpretable to the end-user. Interpretability builds trust in the output of the model in order to take any required action. The model, must, for example, be able to explain the main factors that can affect the insights returned, and also the patterns learned by the model.
You need to ask and address these questions early, both from a solution design point of view, and also to ensure that any new patterns of behavior to take action from are fully realized.
You should empower individuals within your organization to champion and "own" the new solution. Where appropriate, include subject matter experts into the discussions. Finally, provide timely technical training, both on the technical element, and on model interpretability, as appropriate.
There are many elements and risks to manage at every stage for a data-driven solution’s successful deployment. Ensure that your organization’s subject matter experts and business leaders are involved early and throughout the initiative to help it generate traction within the organization, value for the stakeholders, and ultimately the business outcomes you are looking for as an increasingly important part of your enterprise architecture strategy.