At this point, you’ve almost certainly used or heard about ChatGPT. But if you’ve only been exposed to the default version, you’ve only seen a fraction of what GPT technology can do. A generic GPT produces results, but a specialized GPT can be curated for all specific use cases.
Think of it like the difference between the roles at a startup and a company that’s started to scale. When a company is new, everyone wears every hat, doing what they can to stay afloat. However, the company hires people for more specialized jobs when it grows. Teams become more well-defined, and the company benefits from a more intense focus in each business area.
Custom GPTs, when trained correctly, can offer similar benefits within your tech stack.
GPT stands for “Generative Pre-trained Transformer,” a type of large language model developed by OpenAI starting in 2017. At its core, a GPT is a mathematical equation that can predict the most likely next word when given a set of words. The GPT must first be trained on a data set to do that.
Generic, all-purpose GPTs are trained on a wide data set — that’s why you’ll sometimes see a GPT contradict itself. When the training data set is wide enough, you will encounter conflicting data.
A custom GPT uses a narrower data set to specialize in a particular area. If you want to create a GPT that helps your sales reps determine the best course of action when negotiating a deal, there’s no point in training it on a data set that includes baseball statistics or sitcom scripts. Instead, you’ll want to feed it data about industry-specific business negotiation tactics, your company’s sales philosophy, and established best practices.
If using GPT is like going to a library with every book available to you, using a custom GPT is like finding and learning from the exact book or combination of books you need to complete a task.
This may seem like a question for your IT department, but custom GPTs exemplify how modern business tech shouldn’t solely be their provenance. Every department in the company depends on tech, and each department has needs.
Building a custom GPT isn’t like your traditional tech implementation that involves IT installing a program or platform out of the box and then adding customizations or configurations. It’s closer to hiring and training a new employee. And think about it: Would you want another department to handle onboarding and ongoing role-based education for your department’s new hires?
Getting started on building a custom GPT is as easy as signing up and paying for the current GPT model and playing around with it. You can get a feel for what the GPT can do more broadly and start thinking about how those things could be applied to specific departments and roles at your company.
When you’re ready to get serious, you can create a new GPT instance and train it on what you want it to do. That shouldn’t just include official documentation but also tribal knowledge. This is your chance to formalize processes and best practices that only exist in peoples’ heads – a precarious position, as that knowledge tends to leave the building along with the head it’s in.
Once you’ve spent time training it, you can test its functionality within the intended department or role. Whereas traditional software relies on QA testing by its authors and involves waiting for bugs to get squashed by patches and updates, a custom GPT can immediately incorporate user feedback to improve on the fly.
As you encounter unexpected scenarios and edge cases, you’ll iterate upon your custom GPT tool, making it more useful every time you use it. It’s like how specific, direct feedback helps employees get better at their jobs — or, to extend the library metaphor from earlier, it’s like you’ve read all the books and used your research to write several drafts of a paper, each one an improvement on the last, because the more time you spend thinking through your thesis, the better your grasp of it will be.
Of course, it’s not exactly like those things. It’s a good way to think about the process. Still, when you start to develop a custom GPT, you’ll quickly realize that you and your team will need to create a new muscle because it’s a very different approach to developing a business tool.
That’s why we encourage companies to start with the low-hanging fruit: What clear, simple use case can we build toward right now?
Right now, custom GPTs are best used internally. You may have seen them deployed as customer service chatbots, going viral for the wrong reasons. In short, the tech isn’t quite ready to do that at scale—not yet, anyway.
Right now, organizations looking to build and implement custom GPTs should focus on use cases that empower their employees to do their best work. GPTs are good at helping people work through processes, analyze data, and synthesize it into actionable next steps. With that in mind, you could train a custom GPT to:
Those are just a few examples, but any internal process that’s tedious or arduous could be an ideal candidate for using a custom GPT. The sooner you start brainstorming and experimenting, the sooner you’ll discover the best use case for your company.
At ProfitOptics, we don’t just experiment with GPTs for fun (although it is fun). When we develop a custom GPT, we start with a goal and work toward achieving it. We’re also fastidious about documenting the process so that we don’t let any lessons or opportunities slip through the cracks. We absorb everything and put it all to good use.
If your organization wants to start developing custom GPTs, we’re here to help. Contact our team to learn how we can ensure your custom GPTs turn into what you need.
