Proprietary AI Models

Why Your HR Team Might Need Its Own AI

Welcome to FullStack HR, and an extra welcome to the 93 people who have signed up since last time. If you haven’t yet subscribed, join the 8400+ smart, curious HR folks by subscribing here:

---

Happy Tuesday,

This is the first article of 2025, and I’m picking up on the three trends I think will be prominent in the HR and AI space for 2025. You can read about it here.

I’m excited to see what 2025 will bring. I have a feeling that it can’t be crazier than 2024 but I might be wrong here.

During Q1 I’ll speak at a couple of conferences; I’ll be in London on the 22nd of Jan to speak at an CRF Event, I’ll speak at the HR Mind Summit 2025 and of course, I’ll be speaking at HR Tech in Amsterdam in March.

Will I see you at any of these events? I hope so!

Anyhow, let’s dive in on this years first article.

---

Every day, your team types the same company-specific terms into Copilot or ChatGPT: your job levels, your compensation bands, your internal processes. And every day, you have to re-explain what they mean.

While everyone's rushing to adopt generic AI tools, some companies are quietly taking a different path: building AI models that already know their language, their context, and their way of working. And surprisingly, it's not just the tech giants doing it.

Enter the last trend I talked about in my webinar before Christmas—Proprietary AI models. Fancy term, right? But really, it just means companies are building their own custom AI tools instead of relying on the usual suspects like Copilot and ChatGPT.

So, what’s driving this?

Mainly because they want something that works for them, but let’s brake it down.

First, control. If you’re in healthcare, finance, or anywhere dealing with sensitive data, you can’t just hand that over to someone else’s AI. Building your own means, you get to keep everything in-house.

Second, customization. Generic tools are fine—until they’re not. Proprietary AI lets you build something that understands your business, your industry, and even your quirks.

Third, cost. Yes, it’s expensive upfront, but for companies using AI all the time, it can actually save money long-term. Plus, you’re not stuck dealing with a vendor’s rules, pricing, or sudden policy changes.

And let’s not forget the competitive edge. Training an AI on your own data means you’ll have insights and capabilities no one else can match.

Why now?

The timing isn't random. Three key shifts have made proprietary AI models suddenly feasible: First, the cost of computing power has dropped by roughly 70% since 2020, making AI infrastructure more accessible. Second, open-source language models like Llama 2 and Mistral have emerged, giving companies a foundation to build on without starting from scratch. Third, there's a growing pool of AI talent beyond Silicon Valley – countries like Canada and Israel are producing specialists at nearly twice the rate of 2021.

In other words: Just two years ago, building your own AI was like trying to construct a Boeing 747 from spare parts. Today, it's more like customizing a car – still complex, but doable with the right team and resources.

Companies like McKinsey, Salesforce, and Gong are already doing this. McKinsey’s Lilli helps consultants sift through massive amounts of internal documents. Salesforce’s xGen-Sales automates complex sales tasks. Gong’s models analyze billions of sales calls.

How can we prepare?

Building a proprietary AI model isn’t something you do overnight, but here’s how you can start thinking about it:

1. Spot the bottlenecks. Look at your processes. What takes the most time? What frustrates you or your team the most? Make a list of repetitive or high-impact tasks.

2. Collaborate with IT. You’ll need technical expertise to bring your vision to life. Start a conversation with your IT team about what’s possible.

3. Start small. You don’t need to reinvent the wheel. Focus on one or two high-value areas where AI could make an immediate difference.

4. Get your team on board. Talk about AI openly and educate your team. The more familiar everyone is with what AI can do, the easier it’ll be to implement solutions.

"But wait," you might think, "isn't building our own AI model just reinventing the wheel?" It's a fair question. After all, companies like OpenAI and Anthropic have invested billions in their models. And the skepticism isn't unfounded.

Building a proprietary AI model comes with significant challenges. The most obvious one is resources – not just money, but expertise. While building a model might save money long-term, finding and retaining AI specialists who understand both machine learning AND HR is exceptionally difficult. You're competing for talent with tech giants who can offer substantially higher salaries.

There's also the maintenance question. AI models aren't "set it and forget it" solutions. They need constant updating, monitoring, and refinement. As your organization's policies, procedures, and data change, your model needs to keep pace.

That said, the choice isn't binary. Many companies are taking a hybrid approach – using generic AI for basic tasks while building proprietary models for their most sensitive or specialized processes. Think of it like your phone, you use general apps for basic needs, but your company probably has custom-built apps for critical internal processes.

Is This Trend for Everyone?

Probably not—at least not yet.

Most organizations aren't ready to invest the time, money, and talent it takes to build their own AI. But even smaller HR teams can start taking steps in this direction.

Take for example how you could approach recruitment automation. Instead of jumping straight into building a full proprietary model, you might start by creating a custom knowledge base of your company's past successful hires, interview processes, and role requirements.

This becomes your foundation.

Then work with your IT team to fine-tune an existing open-source model on this data. While not a full proprietary solution, it's a practical first step that can reduce screening time by helping you generate more relevant interview questions and identify promising candidates who match your company's unique culture and requirements.

Going back to the “how can you prepare” - here's my challenge to you.

Look at your most time-consuming HR process. What specific knowledge does your team repeatedly use that a generic AI wouldn't know about?

Write it down.

This could be your starting point for a custom solution – whether that's a full proprietary model down the line, or a smaller, focused tool that makes your team more effective today.

Share your thoughts in the comments: What's one unique aspect of your HR processes that you wish AI could understand better?