11 Steps to Build an AI Roadmap for Automotive Manufacturers and Suppliers
In talking with some clients, prospective clients, and other colleagues during the past few weeks I discovered that there is still a lot of confusion around designing a Generative AI roadmap for large, enterprise organizations — especially in the Automotive Industry. Equally confusing, perhaps, was the terminology around Artificial Intelligence.
I would like to help. When mapping out a Generative AI Roadmap, you will need:
1. Problems to Fix and Overall Goals
2. Supporting Organizations/People
3. Experimental Use Cases/POCs/Pilots
4. Data Readiness including ownership, licensing, quality, and availability
5. Ecosystem Partnerships and Collaboration — Expertise and experience needed
6. Technologies and Techniques
7. Data Security, Privacy, and Compliance
8. Integrations and workflow needs
9. Training Needed
10. Success Criteria and KPIs Identified
11. Strategies to Iterate, Modify, Expand
1. Problems to Fix and Overall Goals: They say that when you’re a hammer, everything looks like a nail. Instead of starting with the hammer, you can look for nails — and then evaluate if the nail can be pounded down with the AI hammer.
2. Supporting Organizations/People — AI is a powerful tool and even though it’s been around for a long time, it’s important to have as many people as possible in your camp to give AI a go. The good news is that if AI is a hammer that can pound in other leaders’ nails, they’ll be much more likely to support your efforts and build upon them.
3. Experimental Use Cases/POCs/Pilots — There’s risk in doing nothing. This may seem obvious, but sometimes coming up with the simplest, easiest, most straightforward experimental use case — where you can gain some early wins — is a great way to avoid analysis paralysis. Of course, there’s risk in using powerful tools — but there’s also risk in doing nothing and letting competitors get ahead.
4. Data Readiness including ownership, licensing, quality, and availability — Because data is at the heart of all AI, it’s critical to identify who owns the data needed to train the AI (if applicable), how that data was originally acquired (was it acquired ethically?) How about the quality of the data? We all know the phrase “garbage in, garbage out” but quality takes on a new urgency when that data is used to train an AI model. Last but not least, is the data tricky to acquire or is it easy to commandeer?
5. Ecosystem Partnerships and Collaboration — Expertise and experience needed — The need to evaluate and vet your external and internal partners cannot be understated. I’ve lost track of how many times I’ve talked to a potential client who didn’t properly vet a vendor or partner — and they almost always say the same thing: “ We started this project three years ago and we’re still trying to get it to work and roll in out…” No one wants that.
6. Technologies and Techniques — As strange as it sounds, figuring out which AI technologies to employ and what techniques to use is one of the easier questions to answer. If you’ve managed to locate great supporting organizations and awesome internal and/or external partners, they will be able to help you identify just the right technologies and techniques.
7. Data Security, Privacy, and Compliance — We’ve probably all heard about cases in the news where an individual uses a third-party AI technology to get some help at work and all of a sudden, that third-party using the data to train its AI. It’s of the utmost importance to ensure that an organization’s data is secure and stays secure. Are potential partners regularly going through rigorous penetration and vulnerability tests? Will they share the results from their latest report? Have they ever had a critical or high security incident? How did they handle it? It’s important to note that no system is 100% immune from hacks or security incidents, but it’s important to know that every precaution is being taken to prevent security or privacy incidents.
8. Integrations and workflow needs — How will people on the job use this powerful new tool to solve their problems, become more efficient, and propel the company to a better future? What other systems does it need to integrate with? How will it “appear” for them, right in the flow of work? In Humaxa’s experience, workers don’t want to have to download “yet another app” — they want the AI to be ready to help, right in Microsoft Teams or Slack. Others may need different integrations. These touch points are important to identify ahead of time, even if they are built later.
9. Training Needed — Everyone experimenting with AI at the enterprise level will have different levels of comfort with using the technology, but empowering workers with the training and personal use cases that will enable them to get the most out of the technology will help with adoption and impact.
10. Success Criteria and KPIs Identified — Speaking of impact, defining success criteria ahead of any experiment with AI will help refine future tweaks and iterations, ultimately driving success. What metrics can you measure before and after your AI experiment that will demonstrate the impact? Cold, hard numbers are a great way to measure success. Your partners should be able to help you define several KPIs if you ask ahead of time.
11. Strategies to Iterate, Modify, Expand — It’s extremely rare that an AI experiment will succeed in every way, right out of the gate. Starting small and getting some early data will allow for iteration, trying new approaches, and tweaking AI training models. In the current environment of continuous improvement, experimentation with AI will most certainly evolve, improving bit by bit. This will ultimately lead everyone on an impressive journey, with tangible results, and a plan.
Good luck in your AI endeavors and please let me know if you have questions or need help.
Carolyn Peer
CEO/Co-founder Humaxa
carolyn.peer@humaxa.com
