Subscribe
Share
Share
Embed
We follow the journey of Faith Forster as she creates an AI native tech startup & product.
Welcome back to Building Out Loud with me, Randy, and Faith And we left last episode on a cliffhanger. We're back this week to take a look at what you've been working on. You showed us the original version of what you were working on. In fact, you also left us on another cliffhanger.
Randy Silver:You said you had to pick a name, but you didn't share the name either. So let's start off with what is the current name of the app?
Faith Forster:Yeah. Okay. The big reveal. I've named it Discoveree with two Es at the end instead of a Y. I kind of wanted the name to reflect one of two things.
Faith Forster:Either this is the stage, it's helping PMs do a certain job, or the discussion we had at the end of last episode around linking it through to commercial goals, customer value, and as a product team, what are we trying to achieve? So we were playing around with ideas on those two fronts. It just seemed like the obvious answer. So Discoveree is it.
Randy Silver:And when we left in the last episode, you'd showed us the initial in Lovable, but you ultimately decided not to go forward with Lovable and you wanted to start building in Replit. You're going to show us that in just a minute, but why Replit and what's the difference been?
Faith Forster:So Lovable is really good at building a really simple, what I would call a clickable prototype. More recently, Lovable has introduced a feature where it can create the database, but it wasn't working the way I wanted to. I just didn't find it reliable enough. I did look at, as you know, a number of tools. One that was quite promising was called Solid.
Faith Forster:They're early stage still. Their whole proposition is they build it using agents, but in the way a developer would build it. And I really liked that principle. But the actual platform is too early. It's not fully fiction enough for what I wanted to do.
Faith Forster:The challenge with B2B or enterprise apps, not only do you need to store information in a database, but you need to be able do things like user authentication and connecting into APIs, sending notifications. A lot of the platforms aren't quite there yet. Replit was able to cover those things. I could also connect into GitHub. So I had the code base and could transfer it to Cursor if I need a developer to work on it.
Faith Forster:It just felt a bit safer as a starting point that I was comfortable with, but it could be turned into a working product quite easily.
Randy Silver:Okay. Let's see where you've gotten to.
Faith Forster:There's definitely still work to be done here. It is still a working version. If you want to create a new product, this takes you through an onboarding flow where it asks questions around what markets do you operate in, who are your competitors, who are your adjacent products, What are your business goals and your product teams? And so it uses that to then set up the structure of the application. A lot of businesses have more than one product.
Faith Forster:You can set up multiple products within the organisation and assign which users have access to which products. So it is a little bit more robust. The way I've organised it is quite different. Rather than looking at what your competitors are doing, feature feedback or customer feedback as separate things, I came to the conclusion that all of those are just inputs to identify your opportunities. What we want to know is what are the opportunities.
Faith Forster:These are AI generated, but as a team, can also create your own opportunities. Within those opportunities, it's really about the evidence behind those.
Randy Silver:I'm sorry, you said those are AI generated. It's based on the feedback and it's doing a first triage and suggestions of what the opportunities might be. Is that the case?
Faith Forster:Yeah, exactly. So what it's done is gathered the feedback on the publicly available platforms. Also what competitive solutions in this space. It's used that as evidence to say this is an option that you are, something that you could do to improve your product. Likewise, as a team, if you create an opportunity, it will use agents to find supporting evidence.
Randy Silver:So I can see you've got an example here based on an organisation that you used to work with. You're familiar with the opportunity space and thinking at least from a while back, how good did it do? How reliable were those AI generated opportunities?
Faith Forster:When I first did this and created the agent, I was like, oh, actually those sound quite reasonable. Those are things that we had to work on. But then when you looked into it, if I click on the feedback, the link doesn't actually work. So I'm still not confident that this is working exactly how I want it to work. But I also haven't published it yet.
Faith Forster:So I'm wondering if that maybe has something to do with it. The sentiment scoring has improved though. Not only does it capture the feedback, actually gives it a sentiment score. I have refined that agent quite a bit now. And I've spoken to some colleagues who have worked on other platforms that have sentiment scoring and learned from their experiences.
Faith Forster:So So it actually goes through a multi step process to not only look at the feedback, but then identify different products that they might be referring to within that comment or different features, and then looks at the comment in relation to each of those things before assigning the score. So it's a much more robust scoring system. And I've set that agent up to do that centrally so it's consistent across all the different areas in the platform where we've got sentiment scoring.
Randy Silver:Fantastic. So what have you learned from doing it this way? What else has evolved?
Faith Forster:Yeah, there's been lots of things. Some of the review platforms have APIs that you can connect into to pull the feedback directly. So G2 is a good one for that. I have actually got it talking to the API now. We use Gemini to do the web scraping.
Faith Forster:Gemini is the best at that being a Google product. You can actually add a team member now that sends the invitation. I can then accept that invitation and join the organisation and the product. So there's still a lot more work to be done. We want to make this a lot more dynamic within the product teams that identify as optionees.
Faith Forster:You can then turn those into activities in your plan. I can do now, next or later, and the amount of time investment I think this will need. If I go into our product team plan, you can see there the ticket.
Randy Silver:You've built out quite a bit here. And part of it is because you can. One of the things I've learned in our conversations and I've learned from you over the years is your approach to make things the lowest risk, make sure that they actually are going to deliver value before you spend time and effort on it. How have you done that with this? How much of this is play and when do you get into validation?
Faith Forster:Yeah. It's a good question because I did jump straight into building something. The first version, the lovable prototype, I did share and I got feedback. A lot of that feedback was actually more ideas, more features I could do. Like, you know who our competitors are, you know where our adjacent products are.
Faith Forster:Yes. You've identified the opportunities, but what about some moonshot ideas? Something more innovative or different, which is a great idea, but probably not MVP. When I spoke to Ed Biden, before I started the Build with AI course, I told him about my idea and he was like, stop work. This is still a product.
Faith Forster:You need to validate whether or not people have this pain, whether they want a solution for it, whether they're willing to pay for it. And so I have had seven conversations now with different product leaders and had some really interesting feedback and input. One of the most important of those is the conversation I had with Fernanda, who's currently VP of Product at Clio, about the topic we were just talking about, sentiment scoring. And sometimes that's not very reliable. And how do you deal with that so people don't just give up and switch off the platform?
Faith Forster:One of the things I'm now thinking about is how do we make this more of a toolkit where you're giving the base plate to product teams, but they can then make changes themselves. So you can see here the list of all the agents that I've built. You can see what model we're using. You can see the prompt. You can actually edit the prompt if you want to.
Faith Forster:You can also see where it's run and what it's done.
Randy Silver:So this is where you are today. What's next? What's the next thing on the Kanban board?
Faith Forster:There's still a lot to do to finish refining this. I've been focusing on getting the opportunities and the agents working well. The idea is that you use all the context. So you've got the user story or the PID already written for you. You can share that into Jira.
Faith Forster:The bit that I need to think a lot more about is exactly what we're talking about, that goal setting and the OKRs and how you make sure that anything that product teams are working on is driving commercial value well as customer value. Different organisations do goals in different ways, and it needs flexibility. But it can also do the hard work based on the impact on goals that you think that initiative will have, as well as your time investment. Is this a good return on that team's time and capacity? And giving a structured way of being out and monitoring that every product organization I've worked in, that has been a really difficult thing to understand and have visibility of.
Randy Silver:It's interesting to hear you talk about I still need to add more flexibility here. I wanna do more customization there. I have a feeling that if I was asking you for advice on something, you'd tell me to focus, to test it with an ICP and get one thing going and then add that kind of additional functionality once you validated a single flow. How do you feel when I throw that back at you?
Faith Forster:It's such a good challenge. And it is something that I've gone back and forth on so many times. It's like, what's the MVP here? I started out wanting to gather all the feedback from all the sources. Actually from a technical perspective, it's far easier to start with just pulling publicly available information together.
Faith Forster:I haven't yet got the ability to integrate it into your internal tooling and pull your internal feedback as well. But do you think just pulling the data together is useful? I think that needs to be within the context of actually what are you trying to achieve for it to be really valuable. One of the strong themes I've heard when I've spoken to different product leaders is if you could help me make sure that my product teams are always working on the most valuable things, that's why we exist. And so that's where I want to get alignment set up.
Randy Silver:And it sounds like you need to prove the feasibility and the viability before you do the extensibility in this case.
Faith Forster:You're absolutely right. I should get more people onto this and using it. And actually testing whether or not the agents are pulling useful information together or it's a slob.
Randy Silver:Well, being that it's publicly available information, you can do it both with actual customers and you can simulate customers and do things in advance. So that's gonna be really interesting to see where it goes. But I think we are just about out of time for today, so we will come back very soon to see where you've gotten with the next step in the next couple weeks. Hopefully, we'll start following up with the actual users and feedback.