In this episode Bobby Lehew, Chief AI Officer at Commscue, introduces Ask Skewcast, a free AI‑powered question‑answering tool that lets users query the entire archive of the SKU Cast podcast (over 350 episodes spanning nine years). Bobby demos the tool, showing how it retrieves and cites specific episodes in response to user questions, such as how distributors structure their sales teams. He then brings on Ritika “Ritz” Shukla, the show’s executive producer and content‑marketing manager, to explain how the system was built. They discuss the massive data‑preparation effort required to turn raw audio into searchable text—transcribing, cleaning guest names and industry terminology, adding metadata, and timestamping each segment. The conversation moves to the technical architecture: a Retrieval‑Augmented Generation (RAG) pipeline that fetches the most relevant transcript chunks and feeds them to Claude for answer generation. Early attempts to index all company content proved too slow and produced generic answers, leading the team to constrain the knowledge base to SKU Cast episodes, which turned a limitation into a feature by grounding every answer in a specific podcast episode. They also describe the production workflow that leverages Descript for AI‑assisted editing and Claude for transcript cleanup and show‑note generation. The episode shares early user insights gathered from anonymized queries, revealing two dominant topic buckets—growth (prospecting, landing larger accounts) and team management (structuring sales teams, compensation). These trends inform both product improvements and future episode planning. Finally, Ritz outlines the roadmap: continuous model fine‑tuning, expanding the tool to other content assets like the Backpack newsletter and AI promo briefs, and iterating based on user feedback.
Chapters
Well, it's been 9 years of the Skewcast, 350 episodes, and somewhere in that archive, one of our guests has already answered a question about business you're asking right now. The trouble has always been finding it. On today's brief show, we're going to share a little bit more about an AI tool we built to fix that. It's called Ask Skewcast. It's live now at commonskew.com/askskewcast. And welcome to the Skewcast.
Thank you.
The podcast for innovators and maverick thinkers in the promotional product space. My name is Bobby Lehew. I'm the Chief AI Officer at Commscue, and I'm glad you're here. Ask SKU Cast is a free tool that allows you to ask a question to the entirety of our guests at once, all of our SKU Cast guests at once, whether it's growing your business, supplier relationships, hiring, building a team, AI. Answers come from what our guests said, and it cites the episode so you can go hear the whole conversation. You know how some folks love a behind the scenes or how we built it? Well, we're gonna do that today, demo for you how it works and tell you why and how we built it. But first, a huge thank you to all our guests for sharing their wisdom. This kind of experience is only possible because each and every one of you have been generous with your time and your advice. So this is not only a great resource, but it's a testament to their generosity and to your generosity. For today's episode, I'll play you a few examples or one primary example of how Ask Q Cats works. I'll ask it a question and let you hear how I, or rather how AI Bobby responds. It's about to get real meta. Here's one example. I ask Ask Skucast, how are distributors structuring their teams for selling? Here's the reply.
It really depends on the size and sophistication of the distributor. I've talked to distributors everywhere from 5 people to 500, and there's no single playbook. What I've seen work is this. The distributors who are scaling intentionally build out layers of leadership first. Caleb Gilbertson at Imprint Engine talks about the Delegate and Elevate framework. You look at what you're doing today that you don't like doing, and that's the first thing you delegate. Over time, you build an executive team. He went from doing everything himself to having a full C-suite. And that happened over a decade of deliberate moves. On the sales side specifically, Tom Havens described scaling to 8 sales teams with 15 designers backing them up. That's a structure where you've got specialized roles, salespeople in the field, design support behind them, and the infrastructure to make both effective. The other thing that matters is how you connect your team to your systems. Team-based management, where people work in collaborative units rather than as isolated contributors, seems to make a huge difference as you grow. When you've got visibility across what the whole team is doing, You can scale faster and catch problems earlier. But honestly, the structure that works depends on who you're selling to, how you're selling, and what your margin structure looks like. A distributor selling to local businesses needs a different team than one chasing national accounts.
Now that's just one example, but I thought I'd invite Ritz, who's the executive producer of the SKU Cast, on for a brief minute to talk about the how. Ritika Shukla is our content marketing manager at CommsCue and the executive producer of this show, which means she's the reason these episodes exist at all. And she's the architect behind Ask SKUcast, along with Joelle on our team who edits every one of our episodes. They all make me sound a lot smarter than I am. Ritz is Toronto-based, an Ivy MBA, and before any of the content work, she trained as a computer engineer from circuits to storytelling, as she puts it, which turns out to be exactly the right background for a project like this one. Here's my chat with Ritz. Well, Ritz, welcome to SKUcast, which is a really funny thing to say to the person who produces it, but welcome.
Great to be here, Bobby. Yeah, it is a little weird being on this side of the mic.
For listeners who only ever hear the finished episode, and what they don't know is you and I obsess about AI, all things AI, and we've been working as colleagues really closely regarding AI for a while. Before we get into Ask SKUcast, can you talk about what other AI tools we've used for producing the show?
Yeah, absolutely. I'd say the big one is really Descript, and for those who don't know, What that is, it's a video editor which has these amazing AI features baked into it. And the one that we live in is text-based editing where you can essentially cross out a line from the transcripts and it's just gone from the audio. It's just so easy. Anyone can use it. You don't have to be a video editor to really produce anything at all. Although we have a great video producer that's Joelle, and she ends up taking the first pass on all of our episodes. And once that is locked in Descript, that's when Claude comes in. So we've set up these skills and projects in Claude that end up cleaning up the transcripts, setting up the show notes for us, also getting some of those customer quotes ready for socials.
Having wrangled podcasts in the past, you and I know just how complicated some of the audio engineering can be. Just there's so many details and nuance that now are a click of a button. And so that's part of the reason why we got really excited with AI early on is because we saw it working in audio. But anyways, let's get into how Ask Skucast got made. Now we had 9 years of conversation basically sitting in a library of audio files. So what happened to those files before any of it could answer a question?
Yeah, that's a great question, Bobby. You know, people assume AI is gonna be the hard bit. The hard bit is really that data cleanup. It's like you mentioned, 9 years, more than 350 episodes that we've produced, and all of them sitting in an audio, something that the machines cannot read. So I guess the first part is gonna be the transcript, getting our hands on all of those transcripts, which was a bit tedious, but it's also, it wasn't the hardest bit, the cleaning up those transcripts. As you can imagine, the transcripts had all of our guests' names that were wrong, the industry terms, they were all mangled. And once we were able to ingest it to our tool, they all got stamped with the same date. So as far as the tool goes, all of these conversations happened in the last year. So really the big part was a lot of cleanup, a lot of adding more context to all of our transcripts, who the guest is, what that episode is about, so that the retrieval could grab onto something. And yeah, the unglamorous truth is that data work was most of what the project was about.
Yeah, it's kind of wild to think because now transcriptions are much cleaner than they used to be. We're talking 9 years ago, which is just eons ago in terms of AI, you know. So for anyone listening who's sitting on a pile of their own content and sort of wondering whether they could do something like this now, this could be something internally. They may be sitting on piles of content internally or externally. Maybe they've got blog posts, things like that. What did this show you, this whole process of going through it? What did it show you in terms of what's possible with AI?
Anyone can build this now, and genuinely anyone. There are no excuses left. As you know, I am an engineer, but I also wanna say that that was almost a lifetime ago, and I did not need any of that knowledge to work on this particular project. The tools are really off the shelf, and it sounds a little cliché when I say that the ceiling that used to be the technology and skillset is no longer there, but it's also so true. The new ceiling is going to be your content and your patience. So really, if you are sitting on years of episodes or articles or webinars, you just, you've already done the hard bit. Building the tool is really going to be the easiest side of it. One caveat though, so the difference between just building a fun experimental tool versus a tool that people are going to come back to, it's going to be the boring stuff. Like I mentioned, the cleanup, like tighten up that scope.
Yeah.
testing it out with questions until they stop being embarrassing. You know, like, we've been through that. And some of the days, some of the days, the model, it just acts up. It goes haywire. It's gonna come back with all sorts of nonsense for no reason whatsoever. We all have those days. It's gonna happen. Go for a walk, take the day off, come back to it the next day, and it's gonna be fine. Yeah.
Yeah, I think one of the fun things that you and I explored through this, 'cause we were testing it with lots of questions, Is how well in the end, how well it was answering— the AI was answering the questions. That was really cool because it's interesting to train a tool to draw down on the right conversations is probably the one of the hardest parts. Our original concept was going to draw on all our content, but we sort of drew a line in the sand that would only answer questions from the Skewcast episodes. Why did that— why did that line in the sand matter?
Yeah, so this is where I'm gonna geek out a little bit about the tools. So these tools, all of them work the same way underneath. You ask them a question and it goes and fetches the most relevant chunk of content. This is what is essentially called a RAG, that's called a Retrieval Augmented Generation. And all of this fetching and all of the responses, how good or how bad they are, they all come down to what it's fetching from. And like you said, in our case, it was huge and it was mixed. We were looking at all of our Skewcast episodes. We were looking at our newsletters, the Backpack, all of the issues we've done, all of the articles that we have produced. And that's where we hit the wall, like to get a genuinely useful response because the data was so much and it was unstructured data. I remember initially it took us 40, 50 seconds to get a useful response and No one has that amount of time in today's era. And when we tuned it up for more speed, we remember getting really generic, useless responses. And that's also when we're like, this is not working at all. I guess that's when we pivoted. That's where we did what a reasonable person should do. If you're across a problem that's too big, you make the problem smaller. So we pivoted to SKUcast only. And the fun bit is that The constraint ended up becoming a feature where, you know, now every answer is linking back to specific episodes where you can go back, you can listen to them. It's grounded in that reality. Yeah.
AskUCast gives us anonymous reports of topics people are asking. So we only know names if people submit names, which you don't have to. It's anonymous. But what kinds of topics are we seeing surface to the top right now? And I actually want to interject this thought. AskUCast is a tool we dropped, and it's just the beginning. Because we're learning what our distributor customers and the market is struggling with, whether it's building a team or driving sales and growth. But what kinds of topics are you seeing surface right now and, and why is that important?
Yeah, but first to reiterate, like it is all anonymized, so we do not know who these conversations are being held with. That is very important distinction to make here. But getting into topics, I think, The 2 big buckets that we're getting is one is all about growth. So some of the questions will be, how do I prospect? How do I land bigger accounts? And how do I get past this revenue ceiling that I've been stuck at?
Yeah.
And the other big one is all about teams, mostly about how do I structure a sales team? How do I pay people? How do I lead them? Those sides. So these are the 2 big buckets. We are getting a few platform-specific questions that are more for the support side, which we understand why that could happen. But just to let everyone know, this is the wrong tool for that. It's been trained on StewCast conversations, and that's all the value that it's bringing is like, bring it the stuff that you would ask from a mentor.
Yeah.
And yeah, I think the second bit of the question was why this really matters.
Yeah.
It's really a live read of where the industry is at. So it's doing 2 things for us in the moment. It is showing us where the tool is weak at so that we know what to fix and make it more useful for our users. And the second one is just like showing us some of the topics that they're asking again and again. So if there's a topic that we haven't covered, it ends up getting us our guest draft ready for the upcoming episodes.
We'll keep shaping this tool as we go along. Ritz, thank you for all the work that you've done on this. It's been phenomenal. And it's funny, you and I have been sitting on this for actually a couple months. We could have released this a while back, We held on. It was good because we were able to make some final tweaks and improvements, but what's next on your list to improve?
Yeah, I mean, Ask Newcastle is definitely one. It's not one and done. We have to keep iterating on it, keep training on it. Then second is like building more of it. I know we've discussed if people find this useful, then bringing this to other aspects of like maybe getting the backpack. Backpack would be a really good choice, like getting all of those trends in and maybe AI promo brief after that. So again, it's, it's just, there's a lot that we can build out here. It just boils down to the usefulness. Like, what are people really gonna be using it for? Great.
Ritz, thanks again for being on your own show, but that was cool. What a— congrats to you for, for this tool, and we are excited to roll it out.
