Enterprise AI

This article is intended for decision makers who have a mission to enable advance analytic capabilities (Artificial Intelligence and Machine Learning) for the enterprise. It aims to bring the focus back to the basic frameworks, using real world experiences from an earlier implementation of an analytic organization. The narratives and conclusions in this article are backed by experience and real world examples, although much of the academic research and peer views will support the key conclusions.

Let’s get some of the basic concepts and terminology ironed out to avoid any confusion as to what we’re talking about in these articles.

Please be mindful, there are many definitions and changes to these domains and the only constant in the tool/process section is the change.

As the title suggested we’re going to focus on Advance Analytics in these series of articles, unfolding the people, process and cultural barriers to enterprise wide adoption of AI in particular. But I understand, some of the readers might be wondering if AI is real, some might be wondering if this is another bubble about to burst with no business value, so let’s take a moment and help those readers as well.

If you fall into the camp that doesn’t believe in AI’s transformational role, I suggest you get yourself a toy like google home or a little robot like cozmo. Spend some time with google home and you’ll soon find yourself talking to a little box which will intelligently answer your questions or carry a conversation and when it can’t it will politely say so. Google home is a few years old but it already knows almost everything that is published on internet, just say “Hey google, good morning” and see how it intelligently responds and tells you about traffic and weather condition and plays the latest new!!! Imagine google home capabilities in 5 years. Cozmo is a bad robot, he will recognize your face, and find it’s way around obstacles and do things you may not approve off.

So to those unbelievers out there, AI is real and here to stay; the world around us will adopt it to automate tasks, recognize objects, increase safety, productivity, and augment human knowledge. If you don’t embrace this technology and harness its transformative power, your business suffer.

But how do we prepare, where do we start, how do we execute and how can we measure the value?

These key questions should be developed and answered in your AI strategy and roadmap.  I don’t mean an exhaustive, and expensive strategy exercise led by consultants; No! Just answer some key questions and accept you might not know the correct answer and explore the outcomes. It’s a change of attitude.

There are many choices to be made based on your specific organization characteristics. The idea here is to make calculated decisions fast, otherwise you will be too late and the ecosystem will change rapidly rendering your well thought decisions obsolete.

These key questions should be developed and answered in your AI strategy and roadmap.  I don’t mean an exhaustive, and expensive strategy exercise led by consultants; No! Just answer some key questions and accept you might not know the correct answer and explore the outcomes. It’s a change of attitude.

There are many choices to be made based on your specific organization characteristics. The idea here is to make calculated decisions fast, otherwise you will be too late and the ecosystem will change rapidly rendering your well thought decisions obsolete.

For this article we will briefly describe the nature and scope of the questions you need to ask in order to inform your strategy.

Use cases: The first and most important starting point is identifying what AI/ML can do for your organization. There are several universal use cases that applies to any mid-size organization such as chat bots, RPA’s, sentiment analysis, content analysis, etc. If you are in an industry that is a lagger, you have some work to do to find appropriate use cases that are executable. The key for selecting use cases is to be agile, and build the basic building blocks needed to develop the capabilities and show the art of possible. Developing the use cases is crucial in your long term success and aligning the organization.

Funding & Investment: The questions in this category deal with how you want to justify and fund these AI initiatives. Unless you are part of the C suite or have unlimited discretionary expense budget (a unicorn), you need to work through the established processes to justify your initiative. However when your organization is prioritizing projects that keep the lights on, you have little to no voice at the table to justify your expenditure unless you have a good business case and peer support. So do not underestimate the effort that must go into developing the business case and the support network.

Organization and operating model: The questions in this category deal with how you will organize the resources and design the operating model for success. Would you create a group in Information Technology, corporate functions or business units? Would you have a centralized, decentralized or a hybrid team with matrix reporting relationships. How do you intake demand and process it.  The key here is to ensure people feel they are included and their voice is heard while keeping a centralized authority to quickly resolve various operating issues. This is particularly a hot topic because it influences how people react and feel towards the various initiative, so we will dedicate more articles to this element in future.

Infrastructure: Chances are you most likely don’t have an existing infrastructure to fully support AI/ML use cases, so where do you build one? Are you going to build it in your data centres or in cloud? how do you integrate and extend the data pipeline required to enable the AI/ML engine and apps? If you have a cloud strategy that defines a set of criteria for selecting a cloud v.s. on-prem solution then you might have a starting point but complexities of standing up an AI/ML platform is far beyond what an enterprise cloud strategy might address. For example, depending on your use cases you have to consider what are the most common (high velocity) datasources you need to connect to. If your use case deals with sensors and SCADA data, you might have to consider physical constraint of your network capacity and ability to communicate. Chances are you will end up with multiple clouds and on-prem solutions and a mix of communication infrastructure, so be mindful of these restrictions when considering the use cases.

Platform: There are several major platforms for AI or ML from Amazon, Microsoft and Google to name a few and you must consider and understand the differences between these platforms. The vendors platform philosophy, knowledge of your industry, innovation agenda, scalability, integration and cost model are a few key consideration. Depending on the use cases you want to subscribe to a platform that is thriving and has a well supported ecosystem, hence avoid blackbox and closed environments at all cost.

People: Bridging the knowledge gap and managing change are classic challenges. To fill the knowledge gap are you going to hire full time or part time resources? What skills do you need and how many? How would you manage people perception of AI/ML as it replaces their daily tasks? This is a very important topic and you must take time to understand the skills and talent required to both implement and sustain AI/ML capabilities and manage the changes. We will do more on this topic in future articles.

Governance: If you are eager to develop AI/ML capabilities, chances are you vary of over architecting your governance structure and you are right about that. However, you need to consider who will steward the outcomes and how these capabilities are governed within the organization to ensure the AI/ML capabilities become part of your organizational DNA. The trick is to be flexible and remember governance is not the place to start rather an element to consider in your design.

Ethics: This is a very sensitive and somewhat neglected aspect of AI/ML implications. Do you know what decisions the AI/ML engine is making on behalf of the organization? Does it line up with your organizational values? Are you potentially discriminating or using the data in ways it was not intended to be used. Like governance you may not need to start here but you should consider the implications of your AI/ML initiative, more so because you will be feeding the AI/ML engine with lots of data to develop the necessary features and models.

Tools: Every AI/ML engine needs good quality data, which is a rare commodity in some cases. What tools are you going to use to capture the necessary data, store it, transform it and make it ready for AI/ML. The traditional Extraction, Transformation and Loading (ETL) tools will fail you miserably so you must look for modern toolsets that use AI/ML in order to facilitate the data discovery and integration process. Further, the new toolsets allow the business users to curate the data which is a critical capability if you have a hybrid operating model and essential to scale your data team quickly. Once data is cleansed and ready you want to store it and govern it in a place that is easy to find so future use cases can take advantage of the existing data assets.

The list above is not exhaustive but touches on some of the basic questions for consideration. In summary, like other initiative you must go back to the basics, identify the business value generation opportunity, go through a process and deliver results. The difference is, the applications of AI/ML may not be well know for your industry but it shouldn’t mean you wait for someone else to go first. The key is to embrace the learning potential, have an open mind, build a collaborative network and work through small cases to prove out the value while keeping the big picture in mind.