How This Product Manager Turned a Failing AI Project into Profit in Just Six Months

OK. So a new product or business model from scratch.

OK. Um, sure So, let me tell you about a time when um I just joined the uh product team at Xerox PARC. Um, as a, um, product strategist on the AI Ventures new AI new Ventures software team. And um, Um, I was working on a product called Draft Spark, which was a, um, an RFP assistant, and AI meeting assistant, um, which let people leverage their, um, past proposals and create new proposals. And, um, the team had been working on this for 2 years, we had 0 customers, and my job was to help get the first customers for this product and make it profitable. We had about, just real quick,

that's right. That's right. I just joined the team, and, uh, Xerox, which owns, owns the company, had, uh, given an ultimatum that you have a year to, to make this turnaround. So my job was to, uh, within 6 months, turn this around and get some customers. So, um, I'd like to talk about that story if, uh, that's OK. Great, um, and by the way, at the end of this, we were able to get our first profitable customers, um, eight profitable, eight customers, um, and, uh, in

6 months, yeah, and, um, 1500 in revenue, uh, of signed contract revenue, uh, with more interested customers. So I'd like to talk, talk about that. Yeah, it's OK. Let's go for it, um, great, so. The, the first thing I did was, um, I spoke with the team to really understand what we had done in the past, what knowledge we had, and what we can leverage, um, you know, what we knew about the customer, uh, what things we had tried, what our roadmap was. I tried to really uh get a baseline of what that was. Um, so I spoke with everyone on the team. Uh, the next thing I did was I started from ground zero to really, uh, you know, do detailed customer discovery. So I hopped on calls with sales, customer service, uh, about 30 calls, uh, mapped the entire customer journey, um, And um I basically, uh, so that was one thing, I mapped out the entire customer journey. Uh, and then I uncovered like what were the gaps between where the product is today and what uh is needed by the customer. Uh, and we were at that time, we were exploring both SMB customers and enterprise customers. So I tried to map with the Capos. Um, following which, I, um, I made a proposal, um, with the, uh, the VP, my manager, um, to explore the, uh, to go after the SMB segment versus the, um, sorry, another one month before that. Um. It also turns out that we had, we were capturing usage data, like in the form of logs, but no one was looking at it, just because, not because the team was not capable, no one had the bandwidth. Um, it was a small team, so no one really had the bandwidth. So what I did was I went to the engineering team, I got a dump of the logs, and I created a tool out of that, which was. Um, something that we could reuse, um, but that gave me insights, and so I turned that into actionable insights in terms of who's logging into the product, what kind of actions are they performing, um, you know, where are they dropping off the Q&A

Yeah, sorry, I didn't explain that very well. Um, so we did have prospects who are trying it out. Got it, but we're trying out some of the some of the, um. Uh, features, but no one had signed up as no paying customers.

Yeah, uh, so, no one was monitoring like what their usage was like, so I was, I had a chance to really dig into that and create a tool from that. So it was like a bunch of pivot tables, um, which told me who's using it, what is the, uh, are they, are they an enterprise customer, SMB customer, um, you know, what kind of features are they using, where are they dropping, where are they dropping off, how often are they using it, and also what's the, uh, like what, what do they click on, and, you know, there was. Q&A pairs for the RFPs, um, that would enable people to really quickly create RFPs. Um, so what was the quality of that? So all of that was being logged, but I've created a tool that derived insight from that. After that, I was able to make a case using the qualitative results I got from the customer, uh, from the customer discovery and from this. Analysis that I did, I took that to the manage my manager and made the case for going after the SMB segment versus the enterprise segment because the SMB segment, we were further along when it comes, uh, as far as the product is concerned, to meet all the needs of the SMB segment versus the enterprise segment. Now, there was some pushback from my manager because we had invested a lot of energy, uh, and effort into the, into exploring both segments. There was a sunk cost, so there was some pushback, but I was able to eventually convince her and we can go into that if you like. Um, but the, um, uh, what we ended up doing in, in the end is we focused initially in the SMB segment after we got everybody on board. Um, and then, um, I prioritize the features that were, uh, gonna help us, uh, solve the needs of the enterprise, of the customers, um, of the SMB segment first. And um we prioritized those features we built those out and in less than six months we had 8 signed uh 8 paid customers and uh about 1500 in contract revenue um but happy to dig deeper into any of these elements

Yeah, great question. So, it was a bit of both, more of, we made a lot of changes to the product. I can go into that. So basically, enterprise like uh Uh, so if you look at SMBs versus enterprise, SMBs, uh, they did not care about things like, um, you know, high-end security, uh, they didn't care that much about, um, you know, um, on-premise solutions. Um, they didn't care about integrations with Excel, they didn't care about integration with SharePoint, right? Uh, as enterprise customers care about all those things. Um, also, SMB customers didn't really care about detailed workflows, right? So when you create an RRP, uh, they didn't really care about it going through many different approval processes, uh, but they did care about, we want to leverage the entire team's collective knowledge. So I want to assign this question to this person and this question to this person, so that we can, they can best, uh, and make sure that the, what, what the AI is producing is actually relevant and valid, and, uh, we create the best possible proposal as a team. So we made all those changes, uh, on the technology side, we had to shift from, um, to a hybrid AI model. So, instead of Uh, you know, the, the customer just uploads a bunch of proposals and then, and that they have responded to, and then so it has question and answer pairs, and then uploads a target proposal and then just gets an output. Instead of that, we shifted to, hey, upload a bunch of proposals and upload a target proposal and then you train the AI so it's like human-assisted AI. Uh, so it required a shift in the technology, uh, so that it was more human assist and that required changes to the UI as well. Um, and of course setting expectations with the customer that, you know, for the first, you know, 7 to 10 days if you use it, uh, you'd be training the AI and then You'll be able to get a good initial draft and that came to the marketing side. So instead of um messaging it as, um, you know, you, you can win more bids and win, win more deals, instead we shifted our our focus to um get a great initial draft that you can then use and submit. So those were the changes we made.

Yeah, so

Absolutely, great question. So, um, so what we did was, so. I had a, you know, uh, I had done this analysis from all these calls I, I've had from sales with sales and customer service. And I've mapped out all their needs to the currency of the product. So that was one thing that really gave me good data. The second thing we did was, um, so I went to the VP, um, obviously, she was initially opposed to this because of that sunk cost, and we had come very close with some Enterprise customers, which is why she didn't want to like completely lose focus of Erin Price customers. Um, but, uh, the way I position was, hey, I've spoken with the rest of the team, marketing, sales, everyone's on board. And um we think that this would um be a good fit to start with. We're not gonna like say, you know, like completely ignore the enterprise segment, but you know if we can focus initially on SMBs, get that initial traction, then that would be a good idea. Uh, but the way I validated this coming back to your point is we, we actually Um, did a, uh, did a 3 week trial with both SMBs and Enterprise, and, um, we told both,

Sorry, uh, yeah, SMB is an enterprise, and, uh, we said, hey, so you know, we're giving you a full, full access to the product. Previously it was limited, giving you full access to the product for 3 weeks. Um, and at the end of the three weeks, you know, would you be willing to sign a 1-year contract? And it turns out that 2 of those SMB customers ended up signing, and none of the enterprise customers did. And what they wanted was they, they said, yeah, we love the product, but we want X, Y, and Z, which was a clear sign that, you know, the SMB segment, we are closer for that segments. And over the subsequent, you know, uh, 1.5 or so or 2 months, we were able to sign another 6. OK.

Yeah, so a couple of things went wrong in this, you know, obviously I presented this in a very short, uh, it was, you know, there was quite a few things that went wrong or, uh, a few things, a few stumbles I had. One was, um, It was, it was not exactly easy to get everybody on board on the SMB segment versus the enterprise segment because we hadn't, there was some cost as I mentioned, and everyone had invested, you know, there were leads we were talking to, um, you know, we had come very close with some customers. Can I name the names of some customers or, um, should I do that, or no, I

fine. OK, so, uh, yeah, we had come really close with some customers and Um, yeah, so that was, that was one thing.

Yeah, yeah. So, one big one for both SMBs and enterprise was quality. Um, the quality of the AI, um, and the quality of the Q&A pairs was not exactly up to what their expectations were. Um, so I had to work with engineering on, um, you know, a new AI model basically which involved in that hybrid AI as I mentioned, uh, which, and we shifted focus from, you know, You know, you just add a bunch of proposals and you add your target proposal and you get an, you get your filled out proposal versus the customer has to work and train uh the, the AI over a 1 to 2 week period and then it learns from its past proposal, uh, it's responded to and helps it create a target proposal. So, Uh, that took a good amount of time, like I had to really understand that it was my first time working in an AI, uh, in the AI domain, um, with an, uh, AI product, so I had to really spend time with the engineering team to really understand it, um, and, uh, research it on my own time as well. Um, so that was one hurdle. The other hurdle was, uh, um, actually, uh, I did have some friction with sales because. Um, you know, sales had, uh, made a bunch of requests initially, this was early on, sales made a bunch of requests. Like product requests. Track requests, yeah, for new features and um I had said no to them because uh there wasn't enough data saying, you know, to, to back up that that, that more people wanted this. It was, they were like one-off requests, um, that created a bit of friction and uh eventually what I did was I, you know, I acknowledged their contributions. I said, hey, you know, I'm just, uh, I'm here to, uh, basically, uh, I'm here on the team to basically help us arrive at a, at a good result, to help get our first customers, um. But I also started to like say no more diplomatically to really explain that, you know, we want to make sure that there's, uh, you know, one-off requests are one thing, but if there's enough customers that want it, that, then it makes sense because we have limited resources. How can we really invest our limited resources to serve the customers that we want to serve the best. So those are the, the biggest areas where, you know, there were um stumbling blocks. OK.