AI STRATEGY & IMPLEMENTATION

My Mission

Of the three options identified above, I'm a strong proponent of the "implementing it with caution" approach.  My mission is to engage with clients who agree, and who view me as a resource to help them navigate the myriad of options, and to successfully implement AI projects that will have a positive impact on the bottom line.

I hope you find the suggestions below to be helpful.

TYING IT ALL TOGETHER

Assemble the Right Team

Failure to assemble the right team to help evaluate and monitor proposed projects is, based on my experience, one of the leading causes contributing to the "nine out of ten" failure rate of AI projects. So what is the right team? I strongly recommend the team include, in addition to the the obvious members, product management and data science, representatives from other groups you might not have considered.

Specifically:

• Engineering
• Data operations
• Security
• Legal

The typical response from leadership (and the data scientists) when I share this proposal is "Legal, security? Why? Won't they just gum up the works and slow things down?" The short answer is, no, this team will speed things up.

Engineering needs to be at the table from day one. It's sadly common that, after a  substantial investment has been made to develop a minimal viable product, engineering is tasked with productization and reports that it's going to take way more resources and way more time than anticipated.

Data operations needs to understand the production data needs and participate in refining the delivery road map. Here too, it's sadly common that, after a  substantial investment has been made to develop a minimal viable product we learn that while the test data was easy to get, delivery of production data is going to take way more resources and way more time than anticipated

As for security and legal, it's virtually certain, or it should be, that they will get involved at some point. I've seen projects delayed by months or cancelled altogether because of security and legal roadblocks.

The challenge here is communication between members of the group who are new to the AI table and the core data science team. Given my background in business, data science, and engineering, I can help. With a one-day engagement when the team first meets I can serve as moderator and help create a foundation for effective and productive communication among the team members.

Please invest a few minutes to read this article I wrote on the subject.

Banish PowerPoint! Embrace Prototypes!

The challenge: Your data science team is presenting a proposal for a new AI-based service. In my experience, this is accomplished all too often, either by someone from the product or data science teams, with a PowerPoint presentation. Let's say you like the concept and marketing gives it an enthusiastic thumbs up so you green-light it and it's added to the road map. You may have just approved one of those nine out of ten projects that never make it from development into production. Prototyping can be an effective way to dodge this bullet.

In the context of AI, prototyping is a method for exposing functionality in a way that lets you and other internal stakeholders have hands-on access during development to the proposed application via your phone or web browser. If done well, prototyping can deliver these valuable benefits and more:

• It provides you and other stakeholders with an opportunity to test the application provide feedback.
• It's a great tool for collaboration between data science, engineering, security and product management
• It's iterative; ideas can be tested, and incrementally improved upon or discarded.

One of the services I provide, and something I enjoy very much doing, is collaborating with data scientists and engineers to produce these prototypes. I have the tools and experience to quickly incorporate changes to the prototypes in response to feedback. My background means I can also be very effective collaborating with and facilitating communication between data science, engineering, product management and marketing to address how to incorporate their feedback in a new iteration of the prototype.

And, finally, it's about the bottom line. Your internal teams can produce these types of prototypes, but probably not without significant cost. Data scientists and engineers are expensive. Do you want to pay for throwaway prototypes when the teams could be working on products destined for production? Furthermore, it's likely engineering support will be required even for prototypes because most data science teams do not have the expertise required for the cloud deployment work necessary to make the prototypes available to users. In addition to being expensive, engineers are usually busy. So in addition to cost, turnaround time becomes an issue.

Train Your Sales Team

Are you developing a new product or adding a feature to an existing product? Do you have a sales team or others within the organization who speak directly to prospects or existing customers? If so then ignore this last item at your peril.

The several risks associated with a poorly trained sales team:

• They may exaggerate, ex. "This model is 99% accurate".
• They may unintentionally mislead, ex. "It's one of those ChatGPT systems." It's not.
• They may undersell due to discomfort answering questions.

A good example of the last item in the list above is the challenge auto dealers are experiencing with EV auto sales. The issue has been identified as poorly trained sales staff. They just aren't comfortable selling a product that they don't understand.

Earlier in my career I was responsible for developing sales training programs for a large insurance company. That experience, combined with my ten-plus years as a data scientist, provides a good foundation for me to help you prepare your sales team to become effective and persuasive communicators of the features and beneifts of your new AI product.

ABOUT