Artificial Intelligence and the EMS Industry
AI can improve productivity across EMS operations, but effective governance is critical for managing information security, ethical use and output quality.
Artificial intelligence (AI) is a game changer for increasing productivity and organizational oversight. However, as it is deployed more widely across organizations, the downsides are also becoming visible. The three biggest downsides I’ve seen are the increase in information exposure risks, ethical challenges and failure to thoroughly review the quality of output.
The biggest driver of information exposure in AI use is the lack of AI governance policies. The number of AI tools grows every day. Tell your organization to adopt AI tools as quickly as possible, and people will grab what they know, or they think they know. AI models learn by analyzing information shared by the companies using them, and in a mandated “increase productivity through AI” environment, employees may experiment with tools that don’t restrict shared information to the company. Having clear training on information sharing, which internal tools represent a secure environment and which information can be shared, helps limit information exposure. A central AI governance point that approves tools and continuously assesses risks is also an important check and balance. While it is nice to think that tools advertised as secure are secure, AI models are making headlines, rewriting their rules every day.
Information that may not be appropriate to share in any model includes:
- Detailed customer contact lists, particularly if associated with specific projects;
- Customer documentation or IP covered by an NDA;
- Sensitive employee information.
A central point of AI governance, along with corporate policies on appropriate AI use, is also critical to ensuring ethical AI use. From a sales standpoint, AI tools can profile potential customers to the nth degree, learning far more than they are likely to share. A deep dive in this pool will help you figure out leverage points, interests, behavior patterns, networks and other information that can help build the relationship. The challenge becomes: how much is too much? I’ve seen prospects get angry simply finding out that their email opens, clicks and pass-along activity is tracked. Sharing the fact that you know personal information that a prospect doesn’t believe they are sharing publicly can end a deal. More importantly, gathering information a person hasn’t agreed to share may break privacy laws in some geographies. Those laws are continuing to tighten, so corporate governance needs to anticipate future legal constraints as well as current laws.
Outcome quality is another potential downside. I use AI extensively as a search tool and do my best to refine my prompts to get accurate information. That said, I often find the information that comes back in a search is questionable when I refine the prompt. One humorous example I’ve had was when a client questioned data I’d written in a report about EMS quote-to-win rates. I’ve developed my assumptions about quote-to-win metrics over 45 years of studying them.
The client wanted me to document this by citing outside sources. I did an AI web search and found multiple sources with the same data, which satisfied the client. The problem? I’ve been a prolific speaker and writer. The wording on most of those citations told me the information had likely been pulled from articles I’d written or my book. So I satisfied the client by citing sources that quoted my past work without attribution. I’ve seen worse examples in marketing. A salesperson used AI to write a whitepaper and carefully attributed key sections to appropriate sources.
The problem? All the attributions led to competitors’ websites. AI is great for writing report summaries on action item completion or activities. However, unedited, those reports tend to be repetitive and bland, which can reflect poorly on the person generating them. Similarly, when AI writes marketing emails, it can be pretty obvious, especially if the tool customizes information it finds about the recipient as an introductory icebreaker. This is another area where a customer can be offended either by the information shared or by the fact that the information the AI tool thought was appropriate shows a total lack of understanding of that business. What works well with consumers doesn’t necessarily work well in business-to-business communication when you delegate it entirely to an AI tool.
So, where is AI really good?
EMS is controlled chaos with more information flow than most team members can fully absorb. AI tools can take exception-based reporting to the next level, helping teams start each week with shortage reports aligned with production schedules. Capacity planning can be enhanced. AI tools can help cut programming time on non-recurring engineering activities such as test programming. Virtually any information or trend analysis task can be sped up with AI tools, as long as the information shared fits within corporate governance guidelines. Industry 4.0 lines incorporate a mix of machine learning and AI when fully leveraged.
AI can offer significant benefits to organizations that adopt it with careful planning. What are appropriate uses? How secure is information shared with the proposed tool? Will uses stay aligned with trends in privacy laws? Where is information flow outpacing our team’s ability to analyze it? What trend analyses should we be doing to improve productivity, quality, or return on investment? How do we balance AI assistance with overdependence on AI? Is the output accurate and well written? Companies and associated users who ask these questions as they adopt AI will come out ahead of those who don’t.End of article content
Susan Mucha is president of Powell-Mucha Consulting Inc. (powell-muchaconsulting.com), a consulting firm providing strategic planning, training and market positioning support to EMS companies and author of Find It. Book It. Grow It. A Robust Process for Account Acquisition in Electronics Manufacturing Services. She can be reached at smucha@powell-muchaconsulting.com.