AI-Powered Supply Chain: From BOM Scrubs to Lead Time Automation

Discover how artificial intelligence is revolutionizing supply chain management in electronics manufacturing. Everett Frank, CEO of DigiSource, shares practical strategies for implementing AI tools like ChatGPT and custom GPTs to automate critical processes from BOM scrubs to real-time lead time monitoring. Learn how to integrate Octopart APIs, create intelligent AI actions, and stay ahead in an industry where AI adoption separates leaders from followers.

This conversation explores the shift from traditional supply chain management to AI-driven automation, covering everything from cross-reference generation to predictive supply chain resilience. Whether you're an electronics professional, supply chain manager, or industry leader, discover actionable insights for leveraging AI to transform your operations and competitive advantage.

Resources from this episode:

• Connect with Everett: https://www.linkedin.com/in/everettfrank/
• X-refs: https://www.x-refs.com/index.html
• DigiSource: https://thedigisource.com/

Listen to the Episode

Watch the Episode

Transcript

James: Hi everyone, this is James from the Ctrl+Listen Podcast, brought to you by Octopart. Today I have a special guest for you. This is Everett Frank, CEO of DigiSource. Thank you so much for coming on the show, great to have you.

Everett: Thanks very much, James. It's terrific to be here and I'm honored to be a part of it.

James: Thank you very much. Always happy to have guests who have a lot of expertise in specific areas, which is what we’d like to talk about today. Starting off, would you like to run us through DigiSource’s story and your background in the industry?

Everett: Yeah. I’ve been in the industry 30-plus years. I’ve held senior management positions in distribution and spent a long time in the EMS industry. I’ve done consulting roles for OEMs, so I’ve been on both sides of the industry. I’ve managed sales, operations, sheet metal, cables, PCB, PCB assembly—so a diverse background.

At the end of 2019, going into 2020, we started doing some other things and spent a couple of years really doing brokering. When that subsided in 2022, we pivoted back into software, and that’s when ChatGPT hit the scene. I looked at that and thought, “That’s the most amazing thing I’ve ever seen.”

I’ve really spent the last two and a half years developing different applications in our industry for AI. The most recent thing, which we’re rolling out this month, is X-refs, which provides cross-references. It’s really the first one that’s free: you can just put in your part number, you don’t have to subscribe to anything. That’s built on top of the Octopart dataset. We’ve been in partnership with Octopart for many years, and the latest one is X-refs.

James: Okay, fantastic. So obviously today we want to talk a lot about AI. Maybe we start with: what do you see as the role of AI in the workplace?

Everett: I encourage people to think of AI as an assistant. That’s the way to put it in your mind. The way I talk about the future now is: everyone is a manager. Everyone is an AI manager.

It used to be we said managers hire, motivate, and train in support of strategic objectives—that basic idea was popularized by Peter Drucker in the ’60s or ’70s. Today I think everyone needs to be thinking about: how do I select, prompt, and iterate AI assistants? Think about AI as an assistant. Everyone should be doing that.

James: We had a guest on the podcast a little while ago who talked about something very similar, where AI is becoming the new standard, and if you don’t learn it you will be left behind. But the question is: how do you stand out in that space? How do you make sure that you’re doing more than the average person as it becomes a base skill?

Everett: First we’re sort of speaking about individual roles—what do people do? I don’t know that you need to stand out yet. You need to get comfortable and get involved. The number of people who are not engaging and not thinking about “How can I…”—it’s 90% of the people I talk to. The way you can stand out is: start using AI. Start using ChatGPT. At the moment, that still makes you a standout.

James: So it requires a level of proactivity, is what you’re saying.

Everett: Yeah, and that’s a great way to put it, James. You have to be proactive. You have to be a person who moves forward. If you wait for the company to tell you or provide training or send you to something, now you’re in the “too late” category.

If you start to think about it: if your company is training you on how to do something that does your job—put those dots together. You need to be ahead of the curve. There’s such an exciting and tremendous opportunity for the people who do that.

A simple example in our industry is lead times. Being proactive on lead times. One thing that’s long been talked about is: “I wish I knew, if there was an earthquake in Japan, what was affected—what of my parts are affected?” You can do that now. You can get that news immediately. You can connect that news to your part list and say, “All the parts in Japan, I’m now moving their lead time to zero,” which will cause them in the ERP to move forward.

It’s thinking about things like that: how can I take my tasks and objectives and execute them better?

I think it was Spotify or Shopify that recently announced that no manager could have a hiring requisition approved unless they can demonstrate that AI cannot do that job.

James: Interesting, okay.

Everett: Think of the statement that’s making.

James: And obviously, everyone in the industry has different views on what that means. Some people see it as, “This is progress.” Some say, “This is stealing people’s jobs.” There’s always that dual mindset in this space with any new technology.

Everett: Yeah. I think the thing, though, is that’s an irrelevant question. It doesn’t matter. It’s happening. You can debate whether it could or should. It is.

Right now, just in the last few weeks, if you’re watching the news at all, companies like GM and JPMorgan are explicitly saying, “We are going to let go of 10 or 20% of our workforce, at least. We are converting these roles to AI.” A year ago, even six months ago, they were shy about saying that. Right now, this is the now.

But I encourage people not to think of it as a negative. It’s a positive. Everyone’s a manager now. You just need to be a manager faster and better than the people you’re competing with in your role. It’s an amazing opportunity if you look at it that way.

James: And the thing is, not every AI is suited to every role. If you’re struggling with one software, you shouldn’t give up—you should try to find something else that works better for your particular role.

Everett: Oh, absolutely. I kind of categorize it this way: AI is really good at summarization and at data transformation. It’s really good at those two areas, which have a whole world beneath them.

That couples with the main limitation of AI right now: it can only reason one or two levels deep. If you ask it, “Does A equal B?” and it first needs to go ask, “Does B equal C? Does C equal A?” or it needs to ask something about D, E, or F—at about that third level it starts to fall apart.

Right now you can’t ask AI, “What’s the least expensive way to buy this tape-and-reel component?” The reason is: you can look at the price, but it needs to know, “Is this part number on tape and reel? How many parts are on the reel?” If the reel is 1000 and you want to buy 1200, is the best answer to buy 2000, or 1000 plus 200 bulk, or 200 on cut tape? It can’t do that chain of reasoning reliably yet.

But you can do that by breaking it into questions that AI can answer in order.

James: That makes sense. And that brings me to what I wanted to ask you next: the importance of prompts. I think that’s where a lot of the skill comes in with using AI—being someone who’s mastered writing prompts effectively.

Everett: Yeah. You have to learn what’s possible, and the tools are different. Grok and ChatGPT have different strengths. I think over time those differences will be irrelevant for the big models, but a better example is prompting Midjourney for an image versus prompting ChatGPT for a lead time or a price.

You have to learn what’s possible. When I first used Midjourney, I didn’t know I could give it very specific professional settings—f-stop, aperture, lighting, white balance. You can give very specific instructions. If you don’t know that, if you haven’t worked with it, you don’t know how to do it.

James: I think most people use it on a surface level: “Make me an image about this,” and they give a one-line description of what they want to see. Then they say, “It’s not really what I want, but that’s good enough.”

Everett: Yeah. Which brings me to another important point: learn what it can do. Don’t focus on what it can’t do.

It’s like if you take a hammer and say, “This hammer is terrible at putting in a screw.” You don’t throw the hammer away; you go get a screwdriver. So I don’t have much patience for people who say, “It can’t do this, it can’t do that.” That may be so, but it can do a lot of things. Focus on what it can do.

James: And there’s an area I know you and I spoke about before this call that’s really important: custom GPTs. Can you explain why those are important and how they’re helpful?

Everett: Yeah. It goes back to this idea of “select, prompt, and iterate.” As tools get better, instead of “hire, train, motivate,” you’ll “select, prompt, iterate.”

On the prompting side, you can get quite involved in describing what you want, how to do what you want, across any field. Take writing as an example: you can write in different styles, voices, for different purposes—all expressed in the prompt.

If you want to do something repeatedly, a great way is to create custom GPTs in ChatGPT. That lets you take the prompting knowledge you’ve learned and memorialize it in a tool, so next time you don’t have to write the whole prompt again—it’s already set.

For perspective, I now have about a hundred custom GPTs. There are just different specific things I want them to do. They’re like little assistants.

They’ve also added scheduling. You can create a custom GPT that checks lead times and then tell it, “At 8:00 AM every morning, check all the lead times,” and then execute the next steps.

The really powerful thing is when you learn to create “actions” inside a custom GPT. Actions connect to external data. So you could use Octopart. We’ve written an article about using the Octopart API inside ChatGPT. You can query Octopart via ChatGPT and use conversational language—“What’s the lead time? Who has stock?”—and it will know to pull information from the Octopart API.

James: Interesting. And I believe, correct me if I’m wrong, but within custom GPTs you can train them on specific information. So, say Octopart was running something and you could feed it everything from The Pulse, and have it draw on that knowledge base so when it answers, it’s educated from that particular source.

Everett: Yeah. There’s an additional feature in custom GPTs called Knowledge. In Knowledge, you give it static information. You might put in standards or writing guidelines—the AP Style Guide, custom industry dictionaries, IPC-610, etc.

Then you can say, “Is this part solderable under these conditions?” It can go get information on the part from Octopart, then apply IPC-610 to say whether the temperature range is acceptable. You can go crazy with that.

James: That’s impressive. It’s come a long way in a very short time.

Everett: It really has. But it goes back to what you said at the beginning: it’s up to the individual to think, “What do I need to accomplish, and how can I do this better and faster using these tools?”

James: Yeah, you’re right. It comes down to the user. It’s very functional, but not if you don’t know how to use it.

Everett: Exactly.

James: Going back to APIs for a second, how crucial are APIs to the collection and analysis of data? Within the context of the supply chain, how helpful is that?

Everett: They’re a huge deal. It depends on what you’re doing, but most things in supply chain involve static and dynamic data. Static data would be parametric data. There’s a fair amount of parametric data available directly in ChatGPT, but it’s limited. When you pull it from the Octopart API, you have a much richer field of information.

Via the API, you can also get the datasheet link. So you can pull the datasheet and include it in your conversation about what you want to know. That’s static data.

Dynamic data would be real-time pricing, real-time stock, etc. All those things you can connect via Octopart. If you’re set up, Octopart can be connected to your distributors, so your bespoke prices and discounts with DigiKey, Mouser, etc., can all be present. You can be chatting with that via your custom GPT.

Then, taking the idea of data transformation: it can transform between structures. As a buyer, you commonly get a “buy actions” report—it says what to buy, cancel, push/pull. You could take that report, paste it into your custom GPT, hit go, and it will query Octopart for who has inventory and come back to tell you where all the parts are. Maybe not in one heartbeat, but in three or four.

James: Wow. That’s crazy. It’s really progressing.

Everett: Yeah. There are so many things you can do in supply chain, particularly around data transformation, where you can set up your next actions by linking data that way.

James: Do you think there’s a lot of missed opportunity in the supply chain space to operate in this kind of way?

Everett: In most companies, yes—100% missed opportunity right now. I hope I’m wrong, but I haven’t seen much to make me think otherwise.

Especially on the EMS side, we are notoriously five years behind the curve. We’re not early adopters. There are a few companies out there who are, but in general we’re behind.

Take the simple idea of a BOM scrub. Everybody talks about it; it’s the first thing you need to do when a new BOM comes in. You can do so much with a BOM scrub in ChatGPT, especially if you connect it to Octopart. It’s like: why is this even a question anymore? It’s so doable.

So yes, the racetrack ahead is wide open for people who want to embrace it and learn.

James: Good to know. Speaking of knowing, I’ve heard a term floating around—“MCPs.” Would you like to talk us through what those are and what they’re useful for?

Everett: MCP stands for Model Context Protocol. When you’re going out and getting additional information—like using an API to bring in information from Octopart—that information, in LLM parlance, is “context.” It’s additional context the model can act upon.

Model Context Protocol is a way for LLMs to exchange information between each other. It’s a more capable form of API. An API is very strict: you must ask in a certain way and it returns data in a very structured way.

With MCP, you have a more flexible way. Instead of saying “What is this?” metaphorically you can say “What do you think about that?” MCP gives a way for LLMs to exchange richer context.

Right now MCP is ahead in the race for how we evolve beyond APIs. It’s the next big thing in knowledge exchange.

James: Interesting. It’s a term I’ve started hearing recently, so I imagine it’s something we’ll hear a lot more about.

Everett: Yeah. It’s a backend thing, so it might take time before end users care, but it’s easier than APIs. It might be the first thing users start to connect to as more data providers create an MCP “opening” to their data.

James: This is unrelated to supply chain, but I had a question about something that happened recently in the media that’s very on-topic: data crawling with AI. We’ve had issues with people crawling things they shouldn’t without agreements in place. Do you see this as an ongoing issue? Do you see any solution?

Everett: I can tell you it’s a problem. The fundamental thing is: people training AIs want more and more knowledge to train their models. So you have robots going out and collecting information.

As a website owner, it’s a huge issue because they’re starting to flood traffic. They’re so constant you’re constantly having to deal with it—at the server level and above. For our stuff, we’re using Nginx and have to make all kinds of Nginx considerations. Then it goes through Cloudflare, which has its own things.

We have pages we’re trying to block from AI robots that we’re inadvertently blocking from Google. From a website owner’s side, that’s a problem.

More broadly, it’s dominating internet traffic right now—all these bots. And then there’s the ethics of what information is being used.

In our case with X-refs, we take data from Octopart and do additional things to create our cross-references. We organically figure out what the cross-references are. Do I want those cross-references exposed in a way that ChatGPT can just suck them up? That’s a metaphor for almost anything: any new information you create, you generally don’t want vacuumed up for free.

James: And the other issue I’ve seen discussed is that some people play by the rules and pay to access data like this, while others just come along and do it anyway without any agreement.

Everett: That’s a bit different—that’s web crawling. It’s related, but that’s more individual applications going to specific websites to gather information programmatically. Rather than using the Octopart API, they go to Octopart pages and scrape the data. That’s web crawling.

That’s a whole other problem. There are AI-driven tools that are very good at crawling, and data owners are using AI to be very good at blocking crawlers. It’s a war.

For companies like us, we’re not doing those things, but we still get caught up in various protections, so it creates problems.

James: Last question about AI before we change topics slightly: if someone wants to get up to speed with using it, what are the best ways for them to engage with it?

Everett: The steps are: start prompting. Go into ChatGPT. The first thing I did was ask it to write a story. If that’s what you want to do, ask it to write a story. But ask it to do things.

If you don’t know if ChatGPT can do something, just ask it: “Can you do that?” If you don’t know how to do something—like creating a custom GPT—ask ChatGPT to teach you. It will.

If you want to connect to the Octopart API, you have to create an action that uses an OpenAPI schema, which is quite detailed. I use it a lot and have no idea how to write OpenAPI by hand—but ChatGPT does. You just have to say, “I want it to do this,” and it will help you construct it.

This is what makes it such a crazy world. When we were learning Excel, we couldn’t ask Excel how to make formulas. Now, when we want to use ChatGPT, if we don’t know how to do something, we just ask it.

James: That’s crazy. You can ask it how to use itself. It’s a very unique period in human evolution that we have something like this available to everyone.

Everett: Yeah. I’m reading a book called “Nexus” by the author of “Sapiens,” who talks about this kind of thing. He’s talking about exactly what you mentioned: this is an unprecedented change in information and how it’s exchanged.

Who knows what’s going to happen, but something is happening.

James: For me, personally, it’s like: we had the democratization of information when the internet became widely available. This is the next step in that process. It’s a whole other level of democratization of information.

Everett: Yeah. I talk to writers or artists who instinctively don’t like it—“This can’t do as well as I can,” etc. But it’s not only democratizing information; it’s democratizing skill.

I can envision a sculpture and think “wow,” but I can’t make it. I don’t have the skill. I don’t have the skill to draw certain images or make a video of a cat riding a zebra—but I can imagine it. That’s the big thing: democratizing skill.

Now people can do things they couldn’t before. You’ve got a million examples of 18- or 19-year-olds who’ve never had a job, sitting in their bedroom making software that they sell for a hundred thousand dollars a few months later. It’s not changing the whole world, but it’s real democratization, and it’s non-geographic.

You can be in any part of the world and participate. It levels the playing field where people didn’t traditionally have access to resources.

James: Definitely.

Everett: I think the guy who just sold Scale AI to Google for billions is now running Google DeepMind. Scale AI was hiring people all over the globe to add to databases—relatively simple work, but at huge scale.

There are all kinds of possibilities.

James: So before we wrap up, I wanted to talk about a few topics unrelated to AI, because you have a lot of knowledge in the supply chain space. How have you seen things change in supply chain post-COVID?

Everett: The most talked-about thing is supply chain resilience, and that COVID created an awareness of vulnerabilities around resilience. It created awareness; I’m not entirely sure how much has actually changed.

It did lead to greater comfort with the broker world—with so-called unauthorized distribution. Prior to COVID, working with brokers was basically verboten. Now there’s more acceptance and more understanding of different types of brokers, credibility, inspection, etc. That’s more accepted now.

Honestly, I worry about how much we really learned. We are doing some things now at a bigger level, like with rare-earth minerals—restarting Mountain Pass in California, things like that. I’m not sure the CHIPS Act is a significant contributor to adding semiconductor capacity in the US. It’s positive, but…

I don’t know if I have a great answer.

James: I’ve definitely heard a lot of people talk about the shift from “just in time” to “just in case.” There’s more caution in how things are sourced and stockpiled. People realized, “I can’t just rely on one factory in a region if trade is cut off or manufacturing stops.”

I feel like that was a big wake-up call. Prior to that, we didn’t really have any events in the last few decades that slowed the entire globe down—manufacturing and supply chain.

Everett: Yeah. Amazing and incredible. I’m not sure the world would handle another COVID the same way. I think the world will react with a little more intelligence about the downstream effects of completely shutting down society.

James: It was very knee-jerk because we didn’t know what was going to happen.

Everett: Yeah, we didn’t know. Now we’ve got more information, and there’s been a lot of advancement in medical tech since then to detect and treat viruses. Before, it was more theoretical—“this could happen.” Now it’s, “It could happen again at any point. How do we deal with it?”

James: Exactly. So the last thing I had to ask was: if people want to learn more about DigiSource, see your offerings and resources, what’s the best place to do that?

Everett: Our website is thedigisource.com. And we’ve just launched x-refs.com—that’s x-refs.com. Those are the best places to see what we’re doing. We also write on LinkedIn and other places.

James: And you are personally on LinkedIn as well, correct?

Everett: Sure, yeah.

James: Great. Well, thank you so much for coming on the show. It’s been fascinating talking with you. I’m sure people are going to go straight to ChatGPT and try a bunch of new things.

Everett: I hope so.

James: Very exciting. Thanks, Everett, appreciate it.

Everett: Thanks much, James.

James: For anyone listening, thank you for tuning in, and come back next time for another guest.