Loading...
You are here:  Home  >  Data Education  >  BI / Data Science News, Articles, & Education  >  BI / Data Science Blogs  >  Current Article

Managing Business Leadership Expectations Around the AI Hype

By   /  October 8, 2018  /  No Comments

Click to learn more about author Oksana Sokolovsky and Rohit Mahajan.

Like many technologies, AI is now in a full-scale hype cycle, both in the industry and the societal/consumer mainstream. While it’s nice to have broad interest and appreciation in what we technologists and Data Scientists are doing with AI, having the spotlight on it can also be disruptive to our work. It can lead to expectations among the C-Suite that are out of sync with what we practitioners – or indeed the technology – are able to deliver today.

Managing these expectations is as critical to the success of your AI projects as is any technological consideration. Here are a few expectations that might emerge as a result of the current hype cycle, and how to manage them in a reasonable way.

What AI is, and What it isn’t
AI has been around for decades, but what was only theoretical before is feasibly delivered by the technology today. Still, that doesn’t mean Data Science in a business context can do anything and everything. Building Machine Learning Models won’t produce a HAL 9000, or even an Alexa, nor is that the intent. But when non-technologists hear about AI they may reasonably think in those terms.

It’s important to mentor your leadership around the general parameters of what Machine Learning and Deep Learning do, and do not, entail. Don’t just explain what’s not in scope. Provide positive examples of what can be done, ranging from the simple cases like upsell recommendations and cyber security anomaly detection, to more complex use cases like human activity recognition, that can bring great business value but will require more work. Give management a sense of what’s on the spectrum, and what is not.

Speed of Delivery
The processes of algorithm selection and experimentation are still largely ad hoc, both in terms of their execution and the logging and sharing of the results. Tools to productionalize models, then monitor their accuracy and efficacy in production, are only now emerging, and they’ve got a long way to go.

Sometimes executives may think technology is a fix all, and underestimate the efforts required to produce deliverables. It will help to educate leadership on these “input costs” and their complexity. This needs to be done through proactive outreach and not merely cited as justification when complaints roll in.

Power of the Technology
As Data Science professionals, we know that many Machine Learning and Deep Learning algorithms can be incredibly powerful. We also know they’re imperfect. Models will not be 100 percent accurate. Predictions need to be corroborated with human intelligence and audited by subject matter experts.

In short, Machine Learning and Deep Learning propel us and enable predictive scenarios that would otherwise be out of reach, but they are not just black boxes that output perfect answers on their own. The potential of AI is huge, but when that potential is communicated, the caveats should be articulated as well. Otherwise the possibilities of AI may be conflated with what the Data Science team are actually able to deliver, potentially setting the team up for failure.

Pervasiveness of Deployment
Building Machine Learning, and Deep Learning Models isn’t just time consuming – it’s also opportunistic. If a given area of the business has data that captures its operations and outcomes well, and is relatively clean, authoritative and reliable, then building predictive models to help that business area is feasible. Those prerequisites are non-trivial, though, and only some business areas or units will be able to meet them. Even among the group of business units that do, patience will be required for reasons we’ve already discussed.

All of this combines to limit the areas of the business to which AI can be applied and deployed, and the speed at which that deployment can happen. Those limits will relax as Data Management and AI tooling improve, but those gains won’t be immediate. The Data Science stakeholders must communicate all this to management as well.

To Cloud or Not to Cloud
Will you do your AI work in the Public Cloud? For certain AI workloads, especially training Deep Learning Models, where significant GPU infrastructure may be required, working in the Public Cloud may be desirable. But, for reasons of data gravity, regulatory constraints and more, the Public Cloud may be a no-go. Some platforms, like Apache Spark-based Databricks, are available exclusively in the Cloud, so a decision on this subject may dictate platform and tools, and not just infrastructure.

Training Deep Learning models in the Cloud can be fast and cost effective, as massive GPU infrastructure can be provisioned to minimize training time and then immediately deprovisioned to minimize cost. But for many enterprise AI applications, affordable, on-premises infrastructure may well be sufficient for model training.

The far-reduced operational expenses involved in using on-prem hardware may outweigh the pain of garnering the capital outlay involved in acquiring it. Cluster management software can automate the provisioning and deprovisioning of the infrastructure in a cloud-like fashion, allowing sharing of resources and enabling appropriate chargebacks.

The Public Cloud decision is one that must be made jointly, between the Data Science team, management and, most likely, IT. The decision must be made carefully, and the tradeoffs must be fully inventoried. This must be handled thoughtfully, precisely and graciously, so that explaining the tradeoffs isn’t mistaken for articulating excuses.

Closing Thoughts
Most technologies will capture the imagination of visionary, thoughtful leaders, and that can lead to extrapolations that inflate expectations and lead to disappointment down the line. This goes double for AI, because it is in its hype cycle and because it impacts consumers and society at large, rather than just companies and commerce.

Outreach, mentoring and healthy caution can help to manage expectations and avoid hyperbole. Ultimately, helping business leaders understand both the power and the limits of AI will benefit the Data Science team, executive leadership and customers. It will also lay the foundation for further AI efforts that will garner even more impactful business results.

About the author

Oksana Sokolovsky is an ex Wall Street executive turned entrepreneur; an experienced CEO who achieved early stage acquisition. Sokolovsky is passionate about developing disruptive technology. Her technology expertise combined with business acumen, allows her to bring a unique perspective to developing innovative products, commercializing them, and taking them to market. She is a technologist with experience running large IT departments within leading global Financial Services firms, establishing and transforming technology functions, and leading global high performing teams. During her 20-year technology career, Sokolovsky has held a number of senior roles at JPMorgan Chase, Morgan Stanley, and Deutsche Bank, as well as United Health Care, Instinet, and Barnes and Noble.com. Most recently, Sokolovsky built disruptive data discovery technology, which was acquired by Centrica's Io-Tahoe. Rohit Mahajan is an ex Wall Street executive turned entrepreneur. Mahajan is passionate about developing disruptive technology for data discovery using machine learning. He is an experienced technologist with a proven track record of implementing global solutions at financial institutions for devops, testing, security and data center transformation. He has developed technical solutions for global implementation and led major technology initiatives to drive transformation. In his 20 year technology career, Mahajan has held a number of senior roles at Dun and Bradstreet, Morgan Stanley, and Deutsche Bank. Most recently, Mahajan built disruptive data discovery technology, which was acquired by Centrica's Io-Tahoe.

You might also like...

Data Governance and Data Stewardship Drive Successful Glossaries and Dictionaries

Read More →