Insight of the Week: GPT Powered Self-Service is Here
I've been trying to build a properly intelligent chatbot since I was 9 years old. The tech has always held me back. ChatGPT changed that. With the underlying GPT tech we now have something that can translate, reason, and communicate.?
But outside of question answering, it's really hard to build reliable, transactional self-service with it, because it hallucinates, goes off-topic, and doesn't follow your processes. But I think we've cracked it. Check out this dialog with what we're calling taskGPT:
?
Cool eh? Let’s break this down.
I started by asking it a question about the returns policy for online orders. It got relevant information to answer the question from a vector store that we crammed full of knowledge articles. That's the same technique we used to build my AI clone, Terry But it also asked if I wanted help to return an item.
?
I said yes, and gave my (fictional) name, and mentioned the item I wanted to return. It figured out I needed to be authenticated, used the name I already gave it, and asked for my zip for verification.
?
Then it looked up my orders, and found one with a red speaker in it. It gave me a summary of the order and asked if that's the item I wanted to return. I said yes. (Actually, I said yep, but LLMs are good at translation ??)
?
It knew from the backend API that this item could be returned, but only if it was unopened, so it asked why I wanted to return it. And because it wasn't broken or damaged, it checked I hadn't opened it. Because that's the returns policy for electrical items.
?
Then it asked if I wanted to drop off the item at a nearby store, or get it picked up. I said I'd drop it off, but wanted to know where the nearest store was.
?
It checked my CRM record to find my address, looked up the nearest store and gave me relevant info about it. It even tried to upsell me on the key cutting and pharmacy there! Then it processed the return and let me know I'd get an email confirmation.
?
领英推荐
taskGPT just did what your agents would do. And it's also trained to handle order status, and cancellations. It can schedule or re-schedule a delivery. And of course, it can escalate to an agent if it can't do what the customer wants.
?
Answer questions. Complete tasks. Escalate to agents. That's the holy grail of digital self-service right there.
?
And it could be on your website. In your chat widget. Powering your SMS service. Soon, it'll be able to answer calls, too.
?
Me and my team have toiled for 20 years working around the limitations of legacy tech, and Conversational AI. We've built some awesome experiences in that time, but it took a lot of time and effort. It was a battle. One that only the largest enterprises could afford to win.
?
taskGPT changes that. It's not as nicely scripted as the best Conversational AI. It's a bit wordy, and sometimes lacks empathy and brand awareness. Occasionally it slips up, but probably no more than your agents do. With our approach, we think it's going to be much faster and cheaper to build GPT-powered transactional self-service than Conversational AI powered self-service. And the dialog is more nuanced. Many of its capabilities are emergent. You don't have to write down all the rules.
?
Outside of the reusable framework we built to make taskGPT possible, there’s only a few pages of custom ‘code’ to make the dialog above work.? Could you build this with Classic Conversational AI? Probably. Just. But it would be hundreds, if not thousands of pages of design and code. And there would be nothing nuanced, or emergent about it.
?
With taskGPT you can give GPT powered AI just the right context, at just the right time, to do something that I think is quite extraordinary.
?
But I'm biased. I've been working on this since I was 9! Let me know what you think. If you're interested in our proof-of-concept program for taskGPT, shoot me a message or a connection request on LinkedIn and I'll loop you in.
Kerry Robinson?is an Oxford physicist with a Master's in Artificial Intelligence. Kerry is a technologist, scientist, and lover of data with over 20 years of experience in conversational AI. He combines business, customer experience, and technical expertise to deliver IVR, voice, and chatbot strategy and keep Waterfield Tech buzzing.