Why AI Fails Your Sales Team - And What to Fix First
Most sales teams have AI tools. Almost none get results. Texas A&M's Arnold Castro and sales builder Valentyn Yaromenko break down the 5 real reasons - and what to fix first.
Big Sister AI
May 21, 2026
Open LinkedIn right now. Someone just replaced their SDR team with AI agents. Another founder built a CRM replacement over the weekend. A new tool promises to 10x your pipeline - launched yesterday.
Meanwhile: is your team actually selling more? Is anyone spending less time on admin? Is any of it sticking?
That's the question we brought to Sales OStin in May - into a conversation with Arnold Castro, Assistant Dean of AI at Texas A&M's Mays Business School, and Valentyn Yaromenko, who has built sales organizations for hundreds of companies. Not theory. The actual mistakes, the logic, and what works.
What this article coversReading time: ~11 minutes · Jump to any section
Valentyn Yaromenko, Arnold Castro, and Lucy Yaromenko at Sales OStin · May 20, 2026
The tool comes first. The problem comes second.
Here's the uncomfortable truth that Arnold leads with: your ChatGPT is identical to your competitor's ChatGPT. The model isn't the differentiator. It never was.
"The tool is not the answer. We all have access to the same tools. It's all about the context - who you are, what problem you're trying to solve."
Arnold Castro - Assistant Dean of AI, Texas A&M Mays Business School
The board says "we need to be AI-ready." Someone subscribes to ChatGPT - Gemini, Claude, whatever is on the list that week. The team checks the box. And nothing changes.
Arnold Castro - Assistant Dean of AI, Texas A&M Mays Business School
"A lot of organizations say: our board told us we need to be AI-ready. They adopt ChatGPT and they're like, okay, we're done. It's like - no, that didn't solve anything. You need to figure out what problem you're trying to solve. Because a lot of the time you don't even need a model. It could be simple automation. Or a tool that already exists that gets you 99% of the way there."
Valentyn adds the pattern he sees from the sales side - and it starts even earlier than the AI question:
Valentyn Yaromenko - CEO, White Sales & Big Sister AI
"Some companies say they want to use AI, but they don't have CRM - they just use Google Spreadsheet or something else. When you start using AI without zero structure, you have zero results. It's like a chat - just only chat."
There's also the build-fast trap. A team spends a weekend replacing HubSpot with a custom AI tool, celebrates - and then quietly realizes nobody can maintain it when something breaks.
Valentyn Yaromenko - CEO, White Sales & Big Sister AI
"The problem isn't how you build it. It's how you can support it and keep it running in real time. They don't have the technical team to support this tool, grow it, fix it. It's much easier to build - but it's still hard to keep the tool in life."
The first question isn't "which tool?" - it's "what problem?" Map the process. Then ask where AI fits.
The data is siloed. Everyone knows it. Nobody fixes it.
Everyone obsesses over prompts. Nobody wants to talk about data quality. Arnold's framing cuts through it:
"Everybody obsesses over prompt engineering, but honestly - it's the last mile. The data is the road."
Arnold Castro - Assistant Dean of AI, Texas A&M Mays Business School
Most organizations have data in five different places - CRM, spreadsheets, email threads, Slack, someone's local drive. Multiple versions of the same record. Nobody agrees on which one is real.
Arnold Castro - Assistant Dean of AI, Texas A&M Mays Business School
"One of the first things I always asked my former clients: let's create one source of the truth. We're not asking you to get rid of your ERP or your CRM - but you need to be focused on creating that one source of truth so you can actually do BI and then AI."
Two years ago, bad data meant a wrong answer in a prompt. Today, with agentic systems, the stakes are different.
Arnold Castro - Assistant Dean of AI, Texas A&M Mays Business School
"Garbage in, garbage out - we've all heard that. But in the world of agentic AI, garbage in will scale to become massive garbage out. That simple little bitty mistake that could have been caught early on is going to keep growing and growing as the automation starts happening."
So where do you actually start? Not with the data - with the problem.
Arnold Castro - Assistant Dean of AI, Texas A&M Mays Business School
"You shouldn't start worrying about the data at the beginning. Zero in on the business problem you're trying to solve. Once you figure that out, you do some process mapping. You break down that process, find every moving piece from beginning to end. You'll find out where each table of data resides. You'll see in what shape that data is as you start looking into it. That's step number one."
Don't start with data cleanup. Start with process mapping. The data problems become visible - and you'll know which ones actually matter.
Arnold Castro and Valentyn Yaromenko at Sales OStin · May 20, 2026
Agents get vague instructions. Then they fail.
A C-level comes back from a conference fired up - "we need 10 agents by next week." Someone builds something quickly. It fails. Now nobody trusts agents. Valentyn has seen this enough times to have a clear explanation:
Valentyn Yaromenko - CEO, White Sales & Big Sister AI
"C-level goes to the team and says: we need to build agents. I heard at a conference that this company has 100 agents, this company has 1,000 agents, we have zero - let's build 10 agents next week. And for some reason, mostly all these experiments fail."
Think about how you'd delegate to a new hire on their first day versus a senior colleague who's been on the team for three years. Most people give agents the senior colleague version.
"Building agents is like delegating a fully transparent process. You say: here's what you need to do, here's what success looks like, here are the steps, here's the deadline, here are the rules. You cannot just say: you're the smartest AI, help me find 10 customers tomorrow."
Valentyn Yaromenko - CEO, White Sales & Big Sister AI
His rule: test manually first. Prove the logic works in a simple chat before touching any automation tool.
Valentyn Yaromenko - CEO, White Sales & Big Sister AI
"We make a prompt and test it manually. Try it in Claude or ChatGPT - understand: does it help or not, does it work? When it works, make the skill, test it for the whole team, then build the flow agent in some tool. We don't start from building a flow in n8n, because we don't understand how it works yet."
Arnold adds the layer that should run through all of this - regardless of how well the instructions are written:
Arnold Castro - Assistant Dean of AI, Texas A&M Mays Business School
"You need to be questioning everything. Does it look correct? One of the things we've learned is that everything AI produces seems very convincingly correct. You're giving instructions, giving all this context - but you still need that human in the loop somewhere to vet it. To make sure: yes, this is true, this is legit."
Delegate to agents like you'd onboard a new hire: define success, write the steps, set the rules. Test manually before you build. Human review stays in the loop.
The tools get upgraded. The people don't.
Every sales team has a superstar. Most of the time, nobody knows exactly what makes them the superstar - and that's exactly where Arnold gets called in.
Arnold Castro - Assistant Dean of AI, Texas A&M Mays Business School
"The boss pulls me in to say: how do you replicate this person? Every sales team has their superstar, and everybody else is lagging behind. What you end up finding most of the time it's the human side that's the secret sauce. That's what makes them the differentiator. If they can focus on the human side by taking the mundane stuff away - it makes it even better."
Valentyn makes the same point from a different angle - and points to a mistake that's just as common as studying the superstar:
Valentyn Yaromenko - CEO, White Sales & Big Sister AI
"If we try to measure only the best person, we forget to measure the ones who are struggling - to understand what they did wrong. Nobody says: Ronaldo does this every day, so you need to do this too. You need structure first - a knowledge base, the base skills. And then you can differentiate."
Most coders stopped typing code. They talk to Claude, Copilot, Cursor. The same shift is coming for every role in sales.
Arnold Castro - Assistant Dean of AI, Texas A&M Mays Business School
"It's the soft skills. Problem solving. Critical thinking. Those things that are going to really make you successful in the world of AI. Today, most coders don't type code anymore - they code by talking. It's going to be more about talking than number crunching or hitting the keyboard. How you frame your question becomes extremely important."
He tells a story that stays with you. His eight-year-old son has a nightly ritual: he talks to AI about space, about natural disasters. Over time, the kid learned how to keep digging - tell me more, explain this like I'm 8, why does this happen? Now Arnold is the dumb one at the dinner table, listening to his son explain quantum mechanics.
"I don't have a PhD. But I always say I have a PhD in curiosity. That's what's kept me at the edge of tech my whole career."
Arnold Castro - Assistant Dean of AI, Texas A&M Mays Business School
Valentyn translates this directly into sales:
Valentyn Yaromenko - CEO, White Sales & Big Sister AI
"I have a rule I share with my friends: the best prompt is the question. When you ask AI how to do something - it's better than starting from a command. And it's not just about getting information. Sales is about trust. Questions are the number one option for building trust with a customer too."
The skill gap isn't about knowing which tools exist. It's about curiosity, critical thinking, and knowing how to ask - whether you're talking to a customer or to an AI.
Phone calls happen every day. The AI stack never sees them.
Ask any sales team what data sources they've connected to their AI stack. You'll hear: CRM, email, Slack, maybe LinkedIn. Almost nobody says phone calls - and that's the most underused data source in sales.
Arnold Castro - Assistant Dean of AI, Texas A&M Mays Business School
"The VoIP side - for telephone - becomes one of the most extremely important things. A lot of sales folks are using their cell phones, making calls on the road, as opposed to using an app that transcribes all that information. That's data that could be leveraged to not only help the rep, but for the company to do stuff with."
Arnold illustrates what this looks like when it's actually working - through a story about his doctor that lands closer to home than it first sounds.
His doctor now walks into every appointment with a small recording device. He talks through the exam naturally - asking questions, saying his observations out loud, checking the usual things. By the time it's over, he looks at his screen: all the documentation is filled out. Prescriptions suggested. Everything done. The doctor focused entirely on being present with the patient. The AI handled the rest.
Arnold Castro - Assistant Dean of AI, Texas A&M Mays Business School
"Nurses spend 70% of their time doing paperwork and 30% being human to the patient - when we wish they were more human to us more of the time. The same is happening with any role related to communication on phones. How can we automate all that back office?"
Imagine a rep on a call - and as the customer talks, their past deals, open issues, product questions surface automatically on screen. No searching, no "let me get back to you." Just the conversation.
Arnold Castro - Assistant Dean of AI, Texas A&M Mays Business School
"It's like having all these assistants around you, listening to the same call and popping out information in front of you to help you do your job better. It becomes very important for a human to actually be human at that point - instead of searching for different things, saying 'I'll get back to you.'"
Valentyn summarizes the sequencing that Big Sister AI is built around:
Valentyn Yaromenko - CEO, White Sales & Big Sister AI
"First you need to record data, then connect the data with AI or a knowledge base - and work with this data. Connect them - let it be MCP, API, whatever you're trying to do - into one unified place where your agents can help you."
Phone calls are the richest and most ignored data source in sales. Connect VoIP first. Let the rep be human. Let the AI handle the context.
Where to start
The teams getting ROI from AI in sales right now aren't the ones with the most sophisticated stack. They're the ones who got the sequence right - problem first, process second, data third, tools fourth.
The short version:
The tool comes first. The problem comes second. Start with the business problem. Map the process. Then ask where AI fits - and who will maintain it.
The data is siloed. Everyone knows it. Nobody fixes it. Create one source of truth. Bad data in an agentic system doesn't fail quietly - it scales.
Agents get vague instructions. Then they fail. Define success, write the steps, set the rules. Test manually before you build. Keep a human in the loop.
The tools get upgraded. The people don't. Curiosity, critical thinking, knowing how to ask - these are the skills that compound. The best prompt is a question.
Phone calls happen every day. The AI stack never sees them. Connect VoIP. The rep stays human. The AI handles the context.
If you're thinking about how to build this foundation before layering in AI - the previous Sales OStin conversation on moving from founder-led sales to a team that runs without you covers the process layer in detail. Same logic, different problem.
Arnold Castro · Assistant Dean of AI, Texas A&M Mays Business School
Watch the Full Conversation
This article was written by Lucy Yaromenko, Co-founder & COO of Big Sister AI - an AI platform that analyzes every customer interaction to score your team's performance against your sales playbook, so founders can step back from sales without flying blind. See how it works →
This article is based on a live conversation at Sales OStin - a monthly event for founders and sales practitioners in Austin, TX. Want to join future events? Subscribe to the Sales OStin calendar: luma.com/salesostin →
