top of page

The Check with Joseki Tech | Episode 2: HubBoost Fine Tuning and Prompt Engineering




Adam

Hi everyone, this is Adam and thanks for joining us here. Also from Joseci Tech. With me today I have Dan, our vp of engineering, and Jasmine, our director of marketing. Thanks everybody. Doing, doing good. So we brought everybody together today we're going to show a little bit more about Hub Boost, which is our generative AI integration between HubSpot and OpenAI to help sales guys get that first email out with a little more personalized content. And we're going to do a little bit of test and tuning on that. So with that, Dan, you want to bring up the app? And we'll start from a baseline email.


Dan

All right, so we're in HubSpot here and we're already driven into the contact section. Okay, mister Bill. Okay, so we've navigated to the contact and we'll select email to create an email. This is the standard pop up within HubSpot itself. And this is also the first place where you get to see the hub boost plugin that we wrote. This is the first place that it presents in services. So pressing on the Hub boost will pop another pop up, but it will also submit a request to OpenAI and it will use various pieces of information of the contact, including the name and the title, any about information that contact may have written about themselves on LinkedIn. And I'll use that information, and I'll use the magic of AI in order to try and generate a more personalized sales letter. So let's click hub boost.


Dan

See that in action? Okay. And this is the first rendition of AI attempting to write a personalized letter to Bill Gates. With that, I think, Jasmine, I'll turn to you for some feedback tune so.


Jasmin

That we can see essentially what the current downfalls are of this. So I'm just gonna take a quick peek at this, decide what it is that we like or do not like about this exact immediate offering from hub boost, and then tweak it using our feedback section. So I'm gonna take a second to read it. I hope this finds you well. So this is great and talks a lot about us at first. And I would say that from somebody who wants to get sold something, I want to know why it is that we're targeting me in the first place. So kind of having the first paragraph be more focused and geared toward the client rather than the business that is trying to sell, I think would be a better bet.


Jasmin

So maybe starting off with like a compliment from their work history or whatever's in their about section so that they know that portion of their career is why we are contacting them in the first place. So we just let them know, like, sure, all that we do. We want to start off with, this is what you do and this is what we do and this is how we can work together.


Dan

Sure. I guess for starters, what we have here, from a feature, perspective or capabilities, is the ability to give feedback right away. So this is meant to be part of an iterative loop, just like you use chat GPT, you know, you ask it to do a thing for you, it does a thing, and you have the ability to tweak it conversationally. So for starters, we can try that feedback here. Here we're going to compliment the contact on their professional achievements. So let's see what impact this has. So this, essentially that feedback gets added to the conversation with AI tries again.


Adam

How much do you think it costs to license to jeopardy music alrighty, so.


Jasmin

We got something up. I'd like to extend my sincere admiration for the unparalleled impact. That works way better, because for someone like Bill Gates, who is so notable already, we need to make sure that obviously everyone knows who he is, but we want to make sure that with that last sentence that we have added in there at Jaseki Techland park, we harness the power of technology to enhance the effectiveness of organizations committed to making a difference, much like yours. So now we have connected the initial person to our, like why we are contacting them and why we think we would make a difference. So that's a much better beginning. I would say that a compliment, that isn't necessarily the way that we need to go, but just a connection to the person. So maybe we can kind of remove the feedback.


Jasmin

Let's compliment their professional achievements and let's try to just say, start off the conversation with a connection between them and Jaseki tech.


Dan

Let's give that a try. Now, AI's ability to make that connection will be, you know, based on the contacts information, essentially whatever we have in the about available to it and whatever we've supplied to it on behalf of Jaseki Tech from like a knowledge base. So our assistant has a knowledge base and it is aware of Jiseki Tech as an entity. It's aware of who works there, what their skill sets are, the positions. It's also aware of the offerings of Joseki. I'm going to read through here and see what it did, but the result is kind of based on what information had to work with.


Jasmin

It does have all the information about Jaseci tech because we are the ones that created it, but also with connecting it with HubSpot and LinkedIn specifically, we also have a good amount of information about the people that we're trying to contact and the people that they have in their contacts as well. So I think a great next step to kind of enhance this message alone would be to bring in, for example, Melinda. So they know right now that we're talking to Bill. So we're talking to Bill Gates right now. But it is the Bill and Melinda Gates foundation. And so Melinda has a whole different sect of information that she works with and businesses that she runs.


Jasmin

So saying the work that you lead at Bill and the Melinda Gates Foundation, I know that the work that you've done with blank and the work that Melinda has done with blank gives us a little bit more of that background of we know what you're doing and we know why it is that we think that we are the best option for you.


Dan

Sure. Sure.


Jasmin

Right now.


Adam

So, Dan, a question. So when we add more feedback. So let's say we want to add, you know, highlight more about the work that we do. Do we need to add that to the feedback, or can we just start over with a new sentence and say and highlight what we did? I mean, does it build the conversation, or is it, do we start at the beginning each time, or.


Dan

It's actually additive? So it's adding it to the conversation as we go. You saw in the last one, I kind of blanked out the feedback and just wrote it new. So, yeah, we don't need to keep the feedback around. It's smart enough. It's. You're adding feedback. You keep adding feedback into the point at which you are outside.


Adam

Okay. So just to try it, can we.


Dan

Say, like.


Adam

Yeah, not that this is something we would use, but just, there's.


Dan

Lots of ways that you can ask it to be funnier. Right. You can say, use a funnier tone, be more humorous. There's a lot of ways to do it. This is affectionately known as prompt engineering. You ask the right question, you get a better answer.


Adam

Okay. Not exactly perfect, but it shows what we're talking about.


Jasmin

It took the idea of having fun without any specific kind of indication of what we think is fun and just added some funny terms.


Adam

Yes. A little something we added the other.


Jasmin

Day that I thought was pretty funny was I asked in the feedback session, talk to me about unicorns. And it gave some pretty interesting feedback in a way that integrated it into what were talking about while also still making it a little bit more like.


Adam

The. So we're all for rainbows and unicorns, but I don't know that's going to really close any deals for us. But it shows a little bit. Exactly. Exactly. No, it's a good time. I think it's good to have examples where we're doing something dramatic just so we can see how it's affecting change. But one thing that we hit on here that I think is really the core of what we always want to be doing is we want to be sending sales messages that are more personalized so when they read it, they can say, okay, this is a person that understands a little about what I'm doing. We can start to form a connection between our sales guy and our lead to say, we're people that can help you solve your problem and start to build that conversation.


Adam

So I don't want to have to go in here every time and put the feedback in. I want to say, okay, as the sales manager, I want all my sales guys to be following these rules. So how do I tweak our system so that it's not each person doing the feedback, but it's just sort of a better starting position?


Dan

Sure. Let me cycle back because you guys started with, really, the very first one you started with was good. You did compliment the contact for their professional achievements. So rather than having everybody have to type that in, you want to more or less institutionalize that, right? You want to tell AI to always kind of do that. That's just a matter, a simple matter of going to the assistant that we built for this and giving it a little bit better instruction. So I'm going to switch tabs here. I have OpenAI up in the background and I'm going to proceed to our assistants. We have a couple assistants written because we're testing different things, but the important one here is hub boost. And this is the initial assistant we used just straight up openaPi, and we gave it an initial set of instructions.


Dan

So, you know, it's a salesman trying to compose a sales email. There's some instructions here, things like, we wanted to shorten it to two paragraphs and we wanted to have it less than 200 words, and it messes up the name once in a while. So we gave it explicit instructions to try and, you know, put up some guardrails and keep it on the road here. So we took that exact feedback you guys gave that we put into the feedback section and we can add it right into the assistant instructions here. So it's as simple as that. Yeah, yeah. Anyways, you'll be able to get out.


Adam

Of the screen so if we just close that, does it start a whole new conversation? Okay, so let's start from scratch again. Just hit the old hub boost button.


Dan

If it doesn't start a new conversation from there. It will if I restart the email. So we'll see if it's silly email or not here. Right?


Adam

Serious emulation. Yeah. This is looking more like what we did before. Starts with a compliment. Excellent. So I think this is really cool steps and what I'd like us to be doing. There's going to be some more features we're going to be adding to the tool and some more tuning we're going to need to do. So when we get together next time, let's look at the starting place a little bit more. So something we talked about previously was as opposed to the full body of the starting position being over there on the AI side, we have some of it that it just pulls in the template. So essentially the starting instructions are improve this email so that we'll be able to have Jasmine or I can go in to the templates on hub boost and drop the copy in right here.


Adam

And that when we click hub boost, it'll start from that position. So I think that'll be a good feature that'll give us a lot more ways to dodge and weave. I think that's going to be one avenue of our next discussion. And the other one I think, is really going to be around prompt engineering. So getting together so we understand as users of the system how to better use it so we can get better results. I think one of the things we're going to see that will be interesting in these next couple of weeks is how do we get these emails to better and better with little tweaks? I don't think our goal is to have the perfect email.


If you got a guy on your sales team that's been doing this for 30 years, he's going to better than AI. And I think that's true of AI across the board. But the trick is we want him to better than your average sales guy. We want to codify the rules that you want all your sales guys following so that we're tracking that and we want it to be quicker so that our low players start to become our high players and everyone starts sending out more emails every day. I think that'll be a real good point.


Jasmin

We want to just increase the effectiveness of our baseline, essentially.


Adam

Absolutely. Okay. So thanks for joining us today. Thanks for your time and attention. And when we get together next, we're going to try to see what the next iteration of the tool is able to do. Thanks. Awesome.

Recent Posts

Commenti


bottom of page