The Right Tool for the Right Job: How AI Fits into Sales Training

Introduction

Nolan: Hello everyone, and welcome to the Talent Equation Podcast, sponsored by Infopro Learning. As always, I’m your host, Nolan Hout. Today I’m talking with David O’Neal, who started his career as a seventh-grade science teacher. Not exactly the path you’d think would lead to sales training, but that classroom instinct for how people actually learn led him right into the business world, where he currently leads sales training for a major life sciences company.

He’s spent years figuring out how to make sales reps genuinely better at their job, not just checking a box that says they did the thing. Today we’re talking with David about something he’s been wrestling with in real time: where AI belongs in the sales training process, and where he’s decided, on purpose, to leave it out. David, welcome to the podcast.

David: Thank you, Nolan. Happy to be here and get into this conversation. There’s quite a bit of conversation going on around AI, how to use it, and where to use it. Looking forward to it.

Where AI Already Shows Up in Sales Training

Nolan: Let’s start there. Where is AI already showing up in how you approach sales training today? Where do you stand in this journey?

David: I’ve had an in-and-out relationship with AI for a long time. After my teaching career, I sold K-12 education software, and I actually sold an AI-enabled education tool back in 2010. So, I’ve been having the AI conversation for coming up on fifteen years now.

When I transitioned from K-12 education software into the life sciences industry in 2019, there was already chatter around AI-enabled coaching platforms, doing analysis on video calls, facial response, that kind of thing. But the way I see it, AI is just another tool. AI itself is not a solution. At some point your phone became a tool. Before that, it was a laptop. Before that, a typewriter. This is just the next evolution of technology.

When I was a teacher, I remember a workshop where they passed around an article from the late 1800s about how pencils were going to ruin education because students needed to use chalk and slate. So here we are. We’ve just got another tool in our toolbox, and everyone’s trying to figure out how to use it.

I’m dabbling, experimenting a lot. I think, like most things, when there’s a shiny new toy, people just ask how do I use the toy, instead of going back to the foundation of learning design: what’s my need, what problem am I trying to solve, and then applying AI as a tool to solve that problem. AI isn’t always going to be the right tool. For a very nuanced training topic, that’s probably not the place for AI. Where I try to guide my team is toward table-stakes work, things that are repetitive, common, or consistent. Can we streamline that? Can we make it more efficient?

Learning the Tool vs. Solving the Business Problem

Nolan: It’s one of those things that requires knowledge of the tool itself. When you first got a computer, you had to know what the computer did before you could just tell it to solve a problem. But then we typically oversteer, and we start thinking the computer, or AI, solves every problem.

David: Right.

Nolan: But then you have to go back to what business problem you’re actually solving. At the end of the day, how am I adding value to the organization? Usually that means I can do my job faster, better, or cheaper. So, the question becomes: can AI do this job cheaper? Sure. But is it better?

David: Not necessarily.

Nolan: Right. So, what do you actually need? Something cheap and quick, or real value?

David: For me, the whole conversation around AI for fifteen years has been this: proper deployment of AI shouldn’t change the volume of work you’re doing, and it shouldn’t give you more free time. Those are the two things it shouldn’t do. What it should do is allow you to tap into the unique human capacity you have to elevate the quality of the work.

Using AI for Question Authoring with SMEs

David: If you focus on improving quality, part of that comes with time, but we take shortcuts all the time. The main thing I’m using AI for today is partnering with my subject matter expert partners on question authoring. Humans, by nature, are not good at writing good questions. It’s a distinct skill set that most people don’t carry, unless you’re a psychometrician. It’s tiring and fatiguing. People are okay at writing a question that proves someone knows something, but bad at writing good distractors, or explaining why an answer is right or wrong.

So, our primary use case today is using AI to help our subject matter experts write better questions. Instead of them writing the questions, AI generates a question set that they then evaluate and review, which is really the job of a subject matter expert. It’s not to write, it’s to evaluate.

Quality vs. Quantity: The Real AI Trade-Off

Nolan: Let’s stick on that. If your job was to write ten test questions in four hours, and now you can do it in two hours, do you not think that means you could produce more test questions?

David: Is it better for me to produce more, or is it better to make sure those ten are written at a higher quality? The structure of the question, the structure of the correct answer, the distractors, the explanation delivered when someone answers right or wrong. I can probably learn more from ten well thought-out, insightful questions than from a hundred that are written poorly.

Nolan: But in the past, if you were given an hour to write ten questions, do you not believe you could write the same quality questions in thirty minutes with AI? Or are you saying it still takes you an hour?

David: If I’m not trying to elevate quality, I can absolutely produce more, probably fourfold. If it took an hour to write ten, now I’m writing forty, maybe more. But I’m not elevating the quality. More bad questions still gives me more bad data. I’d rather have the same number of questions at a higher quality so I get a better analysis of the learners I’m assessing.

Why Experienced People Are Slowest to Adopt AI

Nolan: I’d be interested to know if you face this. At Infopro, we develop content and solutions all day long. What we’ve found is that if we get someone who knows AI incredibly well but only has two or three years of experience, they haven’t lived the job long enough to have the context. They’ll produce an asset in two hours that used to take eight. But then the person with the actual context takes longer to review it because they have to go in and add what’s missing. Unfortunately, the people with the most context tend to be the least likely to adopt new tools.

David: That’s the scary part. One of the biggest misuses of AI I’m seeing is people thinking, well, you couldn’t do this before because it wasn’t in your skill set, but now with AI you can. Except you still don’t have the skill or experience to evaluate what the AI produced. Let’s not forget what AI stands for. It’s artificial intelligence. It is not intelligent. It’s made up by fallible people. It’s a predictive tool, always trying to please you, and it’s often bad at it. You constantly have to evaluate it.

Discriminant AI tools, the ones that have been around longer and are designed for a very specific task, tend to be better. For people just getting started with AI, or under a company mandate to use it, my advice is to check your tech stack first. Find out if you already have tools in-house that are pre-programmed for a specific task, so you don’t have to be an expert prompt writer right away. The tool we use for generating questions is embedded in a platform built for that specific task. It’s a good way to get comfortable with AI before jumping into wide-open, generative use.

The Human Element of Shifting from Writing to Evaluating

Nolan: You mentioned earlier that saving time is supposed to let you dig back in and create more value. What are you actually seeing? Is your org using that freed-up time to cut costs, or to do something you weren’t doing before?

David: It’s not about cost savings for us. It’s about being able to do things we couldn’t do before, investing more time in the quality of the work, going deeper instead of just scratching the surface. With the question authoring shift, we’ve turned our question sets into knowledge contests. We use AI to generate the questions, and we also use it to run the data analytics, so it’s not just about who participated or answered the most questions. We’re looking at performance growth, how fast someone improved, using AI to generate a multi-factor performance metric and process that data quickly.

Going back to the subject matter experts, when we shifted from asking them to write questions to asking them to evaluate questions, the feedback was that it felt like a better use of their time. Writing questions felt distracting and laborious, not a use of their expertise. Evaluating did. There’s a real human element there. I think that’s a good example of how using AI gets us more engaged with the things that make us uniquely human. I felt better as a subject matter expert evaluating the quality of something rather than just writing it.

Nolan: I never thought of it that way. I usually think about using AI to cut down someone’s workload just in terms of time saved. But you’re right, they probably feel better about the job because they’re applying their expertise instead of just producing output.

David: And look at what most people use AI for today. They use it to circumvent the process of writing, because writing is one of the most complex cognitive things we do. You’re taking a mental thought, translating it into words, and then into physical symbols for someone else to understand. That’s a huge cognitive load compared to reading and evaluating. Writing is labor. Evaluating is much less labor.

We had a roundtable conversation where most people talked about using AI to write emails or draft documents. One question was, is anybody using it to analyze spreadsheets? That’s where I’ve gravitated, because writing doesn’t bother me much, I just avoid it, but I deal with spreadsheets constantly. For our contest data, I’m cross-referencing three spreadsheets with over 20,000 records. Manually, that’s a full weekend. With an agent, it runs in under fifteen minutes. Then I can dig into what the data actually tells me: knowledge gaps, strong performers, weak performers. We already adjusted a training program based on data we got in June and July for a session coming in September.

Where AI Does Not Belong: Niche, High-Context Content

Nolan: So where are the areas where you say AI does not touch this for you? Where it just doesn’t belong in this world of sales training?

David: When you’re trying to use AI to generate, alter, or modify content that’s very niche or industry-specific, the more nuance and context there is, the lower the value of AI. By the time you get a prompt to the level where AI understands all that nuance, you might as well have just done the work yourself. You have to evaluate how unique your ask is, and how much you’d have to teach the AI to produce what you need. Think of AI as another member of your team. How much effort do you have to put in to train it, versus doing the work yourself or handing it to someone who already understands it?

Treating AI Like Another Team Member

Nolan: That connects to something I talk about a lot. If a task used to take ten hours and now takes one, it usually isn’t good. What you should be doing is spending the first twenty or thirty minutes training the person, or the AI tool, giving it context. I tell people in marketing, if you’re asking AI to create a landing page or an email, speak to it like you’ve hired an outside agency. If you just say “write a sales email,” it’s not going to get you anything good. It’ll infer a lot, but the more it has to infer, the less value there is.

David: Exactly. Speak to your AI tool the way you’d speak to an agency. For people just getting into this, use the microphone function if it’s available. Writing is labor, and you’re never going to be as detailed in something you type as in something you say out loud.

The second thing is repeatability. If a task is truly a one-off, there’s little point in using AI for it, unless it’s especially complex or you’re using it to brainstorm a starting point. The bigger value is in repeatable tasks. If I only need one landing page, maybe not worth it. If I have five, six, seven products and need landing pages for all of them, now there’s real value, because the tool learns and gets faster with each one.

Why Repeatable Tasks Get the Most AI Value

Nolan: You mentioned niche, high-context work as one place to keep AI out. Are there other areas where you just say, that’s not for me?

David: The one-off task is the main one. I see this a lot with people writing emails. Someone shows me a great email they wrote with AI, and I ask how long it took. Thirty minutes. I ask how long it would’ve taken to just write it. Five minutes. That’s going backwards.

I’m also still in a learning curve phase of figuring out what different AI tools are actually best suited for. What writes better, what handles spreadsheets better, what handles graphics better. I’m still experimenting. Some of that might be my own limitations with prompt writing. But I think eventually there will be certain types of requests I just decide are not an AI ask.

Tool Sprawl and the Rise of Homogenized AI Platforms

Nolan: The tool sprawl in L&D is interesting. Take presentations. Claude could create them, but historically they weren’t great. Then tools like Gamma came along and did much better presentations, but the content inside wasn’t as strong. So people write prompts in Claude and push the output into Gamma.

David: Right. It looks good, but the text isn’t good. There’s so much nuance in these tools. The ones generating entire e-learning modules, you can see them coming a mile away. No instructional design skill behind them.

Nolan: And then you get authoring tool companies bolting AI onto their existing product, and tools like Colossyan, which started with video and avatar-based learning, now adding translation. Every tool is becoming fairly homogeneous because they’re all trying to expand beyond what they were originally built for.

David: The tool creep is real.

Nolan: It used to be that you built a course in one tool, got images from Shutterstock, edited them in Photoshop, loaded them into Storyline, wrote the text in Word, did voiceover with one vendor, translated it with another. Now you can do it all in one tool, but the all-in-one tools today tend to do one thing well and everything else poorly.

David: Jack of all trades, master of none. I had an interesting hallway conversation at ATD about a headless LMS. LMS systems have bolted everything together into one behemoth. Someone on a manufacturing floor who only needs his work instructions is paying for a license that also includes a content generator and a library he’ll never use. What if you bought only the best-in-class tool for what each user actually needs, and used an AI agent as the entry point, with another agent handling the data analytics on the back end? Best-in-class tools in the middle, AI agents on either side. Is that a better replacement for the current LMS, on quality and cost, because you’re only paying for what people actually need?

The Shift from LMS to AI Agents

Nolan: That’s not a planted question, by the way, David and I didn’t discuss this beforehand. Infopro Learning is actually launching a product this July that leans into exactly that. So, the question becomes: are people really going to log into an LMS, find the training they need, take it, and be done? Or is an agent just going to serve up everything they need?

David: Two basic use cases. One agent says, here’s what you have to go do and how to get it. Another agent is for someone who wants to learn something and asks how. It’s like an Oreo cookie, best-in-class tools in the middle, agents on either side.

Nolan: The way I’ve been explaining it to people: we were promised personalization at scale with LXPs, everyone gets a personalized learning journey. But it wasn’t really that. It was personalized the way an extra-large shirt fits differently across brands. What I actually need is something more precise than that, because not every marketer needs the same content, people’s skill sets aren’t all the same. That’s why we have competency models.

David: Right.

Nolan: If you have an AI agent on the back end that’s essentially a digital twin, one that really knows me, my capabilities, my competencies, what I’ve been asked to do, and has actually seen me do the work, connected to the tools I use, it should be able to do exactly that. I’m fascinated to see how this plays out, because it’s a big leap for a company used to a traditional LMS front door, everything routes through it, to shifting toward a chat interface instead.

David: It’s huge. But it’s the same cycle. When the web first launched, everything went digital, with a lot of niche applications that eventually consolidated into giant platforms. Now everything’s flattening out again. Everyone has an AI coaching tool, an AI content creator, it’s in every system, but that dilutes the value, because they’re not necessarily good at the parts they bolted on. Eventually the market swings back toward best-in-class tools for the people who need them.

The Video Game Analogy: Why Less Is More

Nolan: You know what I equate it to? Video games. Did you ever play them?

David: I had an Atari 2600.

Nolan: You can only play one game at a time. Companies like Xbox and Sony have tried offering unlimited access to a thousand games, but that’s not how people actually play. They want their five premium games. If I want something, I’ll get the cartridge out because I want the value of it specifically. Don’t give me six thousand versions of something kind of like Halo.

David: I play five games and don’t want to navigate through the rest to find them. It’s the same thing.

Nolan: Give me one Halo, and I don’t care what the rest are, because that’s where I’m investing my time. I think that’s exactly what’s happened with training content.

The Irreplaceable Value of In-Person Training

Nolan: As we wrap up, I want to ask about the value of in-person practice and training. Is it rising, shrinking? Where does it fit as AI continues to move into coaching and other areas?

David: Simple human connection. We still need it. As great as this conversation on screen is, I got more out of the two minutes we talked in person on the floor at ATD than this. AI cannot replace true human connection. I talk about this in my training classes: communication is roughly 7 percent words, 38 percent tonality, and the rest body language. If I don’t see your whole body, I’m missing pieces of communication. In our live classroom training, PowerPoints have practically disappeared. People come in on day one and present back to us. It’s all conversation, feedback loops, human feedback loops, which drives quality.

Proper deployment of AI, like any other tool that’s come before it, reduces the mundane, rote task and increases the quality of what we produce, and the speed at which we produce it. But speed alone is no good if all you’re doing is producing more of the same junk. You have to use the speed to elevate quality. I think the next evolution with AI is elevating the things that are truly, uniquely human: building a team, building culture. I don’t believe AI is going to do that.

Closing Thoughts: Save Time, Drive Quality, Create Connection

Nolan: I completely agree. We’re seeing across the board that the value of training has become less about the content itself and more about linking arms with the people next to you. Even under the 70-20-10 model, seventy percent on the job, twenty percent through peers, ten percent formal, if we put people in a room together, we’re magnifying that twenty percent, and it becomes twice as effective.

David: Every new hire class bonds that way. Our VP of Sales, who’s been here almost twenty years, still has a text thread with the people he went through new hire training with. That cohort, completely different backgrounds, bonds over the shared experience and becomes a resource for each other. That’s what you can’t replace, that human connectivity.

Nolan: That’s a great way to end it. That’s actually how we reconnected, a chance encounter at ATD after not talking for a couple of years. If we’re in a position to create those experiences for our teams, let’s find the budget to do it.

David: So you use AI to save time, drive quality, and create more opportunity for human connection.

Nolan: David, thank you so much for the time today. Loved having you on. Let’s run into each other again soon.

David: Hopefully. This was good. Always a pleasure. Thank you.