There was an interesting discussion on a customer forum recently. People were asking each other how they handle PII scrubbing tools for LLMs. What is everyone using? How do you take all of your company information, feed it into ChatGPT or Anthropic or the cloud, and anonymize it along the way so you stay compliant with data privacy?
PII is personal identifying information. Phone numbers, email addresses, anything sensitive that identifies a person. Data privacy is a huge issue in today's world, and in the AI world it's an even bigger one. So the question is a good one to be asking.
But I think the scrubbing-tool approach is usually the wrong one for the long term.
This is a pattern we see again and again in the industry right now. How do we approach a problem? There are lots of ways to solve any problem, but what's the best way? It's easy to say, well, there's an X-Y-Z tool, we want X-Y-Z, so how do we make it fit in that peg or that hole? A lot of the time, that's just not the right long-term approach.
Think about what a scrubbing tool actually does. If companies are using your data for AI training, they still know who you are. They know your account. You're feeding the data in. So even if you anonymize it, they can still aggregate it and collect information about you. And it's not that anyone is being devious. It's simply that AI training is becoming very human-like.
If I asked an AI a question about a company, its data, what its gross sales were, or any number of things, that's not out of the question depending on the plan and the restrictions in place. It could still draw conclusions. It would still be intelligence about your company. Scrubbing the obvious identifiers doesn't change the underlying dynamic.
The real path forward for most companies is to partner with folks who protect your information. At Kodaris, we're on Amazon Web Services and use their tooling so that data stays private to each customer. All of our AI tooling is very specific to you.
This connects to something I've been saying over the last couple of years. Your training data and the specific data around your company are going to become intellectual property, and many times a competitive advantage. That advantage comes from the context, the data, and the brain trust you put into your AI specifically. Without that, you're just like every other company that AI already knows about and can tell other people what to do.
So circle back to the personal information question. Sure, you can create scrubbing tools and do all of that and put it in place. But is that really the best approach? Or is the best approach to partner with someone who is going to silo and protect your information while still allowing you to use the capabilities of AI?
What we find is that AI makes us more efficient, more productive, and faster. But making an inefficient or wrong process faster doesn't necessarily mean it's the better long-term solution. Especially in a world of fast-changing AI tooling, we have to start thinking as customers. What is the right process? What is the best long-term solution?
It's like upgrading your ERP from on-prem to the cloud and gaining so many more capabilities in the process. Do you try to bolt everything onto on-prem and maintain it, or do you go to a cloud solution that's more advanced and moving faster? It's the same with AI. AI is not a magic wand. These are tools to drive innovation and efficiency.
But you're still going to see competitors driving the right long-term solutions with AI and pulling far ahead of anyone driving the wrong solutions or the wrong processes faster. As these tools get implemented, I can even see a role for consulting and guidance that helps businesses rethink their processes rather than just bolt AI onto the way they've always done things.
So that's something to think about as you contemplate the future. As your team looks down the road three to five years, the evaluation isn't just which tool fits. It's whether you rethink the process and partner to invest in the future rather than bolt AI onto the old way.