Key Steps to Implementing AI in Your Enterprise

James Hoshor | May 1, 2018 | In the News Mobile Strategy

In a recent episode of our Device Squad podcast: AI in the Enterprise, my colleagues Steve Brykman, Michael Golub and I discussed everything from AI’s evolution to its value in the enterprise to potential hurdles companies face implementing AI. We covered a lot and it is definitely worth a listen! However, in this blog post I want to elaborate more on the key steps to implementing enterprise AI, including adoption hurdles, and recommendations on how organizations should prepare for implementing AI-specifically given what we’ve learned helping clients with their AI initiatives recently.

As discussed in our podcast, plenty of organizations across multiple industries have already invested heavily in AI and are reaping its benefits, while many others have just begun the journey. If you’re one of those organizations that has yet to dip its toe in the water, you’re in luck. It’s still very early in the Enterprise AI story, and now you can learn from the challenges and lessons of others.

But before we share the key steps to take when implementing AI, let’s take a moment to look at some of the challenges:

Lack of Data Access. Data is the ingredient most critical to AI readiness. Nearly every organization has tons data across disparate systems. The challenge is ensuring you can access your data at a granular level. For example, if you are a manufacturing company with years of data spread across different systems (e.g. ERP, MRP, CRM, etc.) you will need to be able to access this data-and more importantly the right data-with granularity down to the daily transactional data level (e.g. SKU locations, orders, customer information, etc.) in order to leverage AI successfully.

So, step number one: be sure your data collection and storage mechanisms are easily accessible and can support highly granular data.

Lack of Infrastructure. Not surprisingly, Machine Learning (ML) and Deep Learning (DL) require serious processing power. Most companies turn to cloud computing and massively-parallel processing (MPP) systems to solve this challenge, but these are short-term solutions. Here’s why: as data volumes continue to grow, and ML and DL drive the automated creation of increasingly complex algorithms, the bottleneck will continue to slow progress.

So, in addition to making sure you have the right data, you need to be sure the data is ready for AI algorithms to digest, and that you have the ability to process the algorithms quickly.

However, we should also note that GPU—not CPU—advancements are what’s behind ML’s recent proliferation. Utilizing different processing architectures that scale more easily than traditional servers. Not surprisingly, Google and FB are both investing in custom GPUs to optimize performance for this specific type of process (Google calls them TPUs or “TensorFlow Processing Units”).

Lack of Talent. Until now, only a handful of organizations were willing to invest in developing the skills necessary to implement ML-given the relatively few AI use cases in the enterprise. But with the explosion of interest in the last few years, all this has changed. Data science courses on AI development have become prevalent and are generally over-subscribed.

Now that we’ve highlighted the three primary challenges organizations have faced when it comes to implementing AI, here are the key steps we recommend for implementing AI in your enterprise. In short, don’t be caught lacking!

  1. Develop an AI Strategy. Before implementing AI, it’s important to have a well-defined strategy in place that clearly defines the drivers you want to achieve with AI (e.g. improve customer service engagements), and identifies your target users (e.g. customers, sales reps, internal office personnel). The next step of the Propelics Kickstart approach describes some possible use cases and prioritize them based on their projected impact on key business drivers as well as on organizational readiness, implementation complexity, and technological constraints.

    Now that you have a clear direction, and have defined and prioritized your AI use cases, you need to determine your organizational readiness to support AI. This requires assessing and understanding the maturity of your people, processes and technologies necessary to realize your top AI use cases. Doing so lets you identify any gaps and determine what’s required to achieve the level of maturity to support your AI initiatives-both current and long-term. We recommend developing an AI Readiness Roadmap that defines the tactical actionable initiatives required to fill the identified gaps.
  2. Determine your AI Readiness. Review the AI Readiness Roadmap and develop a plan for implementing the short-term AI solutions enabling you to achieve a quick win.

    Based on our experience, the areas that typically call for immediate focus are data collection and storage, data quality, security, and an examination of existing business processes that will be affected in order to support your AI use cases.
  3. Perform a Proof-of-Concept. We recommend performing a Proof-of-Concept-using your top AI use cases-that demonstrates how the AI solution will behave under real-world conditions. This provides a true test to determine the usability of the prototype and provide immediate feedback, identifying any shortcomings in user experience, data access and accuracy.

    Based on the POC, next you should create a detailed approach and plan for implementation, including the cost estimates, timelines, and high-level work breakdown required to take the POC to the next step and begin building the ‘real deal’ app.

In summary, before your organization can start leveraging the value of AI, make sure you’re ready to overcome the challenges of implementing AI—lack of IT and security infrastructure, lack of talent and AI knowledge, and the biggest barrier-accessing the data itself.

If you haven’t given AI much thought, now is a great time to start. Many industry-leading companies have successfully implemented and deployed AI and mobile solutions-by first developing strategies to leverage these emerging technologies, and ensuring they were aligned with business drivers. Are you fascinated by all that Machine Learning can do for your business, but don’t know where to start? Our Emerging Technologies Kickstart may be just the boost your business has been looking for. To let Propelics help your organization develop a strategy aligned with your business drivers and guaranteed for success, simply reach out. We’d love to get you started.

James Hoshor

James is a Senior Mobile Strategist & Solutions Architect for Propelics. He has over 20 years experience in executive leadership, strategic planning, marketing and business development in information technology. For the past 10 years James has worked with many clients across multiple industries, including financial services, insurance, retail and manufacturing, approach mobile strategically to identify and deliver solutions that result in market differentiating solutions and great user experiences.

More Posts

Follow Me:
Twitter