Subscribe
Share
Share
Embed
Serious People Podcast: An Operator's Manual to AI is a podcast for business leaders and operators running complex, real-world businesses. The ones selling supplements, managing caregivers, or running service crews. Not software.
Host Noah Levin brings nearly two decades of experience at Amazon, Whole Foods, and healthcare tech to weekly conversations about what AI outcomes work inside a business, and what doesn't.
Equal parts practical and irreverent, but useful above all else.
April Lane (00:00):
We opened the Pandora's box. We gave everyone access to Claude.
Noah Levin (00:03):
How many employees are we talking about?
April Lane (00:04):
Our corporate staff is around 300.
Noah Levin (00:06):
300 people are getting Claude installed on their computer.
April Lane (00:09):
Unfiltered access to Claude.
Noah Levin (00:11):
Was there any limit on usage, tokens?
April Lane (00:13):
No.
Noah Levin (00:14):
So that's a pretty bold bet.
April Lane (00:15):
It's a very bold bet. And that's when we had to start thinking about token limits because all of a sudden our AI bill got very large.
Noah Levin (00:21):
Welcome to Serious People. I'm Noah Levin. Today, we're facing down your CTO's biggest fear, giving everyone access to Claude and letting everyone, even the business team, write code. My guest is April Lane. She's the chief merchandising officer at Thrive Market. Her team runs a massive online grocery business with thousands of items and as you might expect, a lot of spreadsheets being emailed back and forth. Earlier this year, Thrive made a bet that by giving everyone access to AI tools and unlimited tokens, employees would invent solutions to their own problems and it worked. They let a thousand flowers bloom and they didn't regret it. Here's my conversation with April. You have agreed to talk us through all the things that Thrive is doing with AI and specifically all the things that non-technical people at Thrive are doing with AI.
April Lane (01:08):
Exactly.
Noah Levin (01:09):
Are you technical at this point? How do you classify yourself?
April Lane (01:13):
No, I've always wanted to be technical. And in fact, actually, one of the most exciting things to me about AI right now is that I finally don't feel like I need to go back to get a computer science degree. I studied sociology and poli-sci at Northwestern, and immediately once I got into the business world, first couple of roles, that seemed fine. The minute I landed at Amazon and realized that Amazon wasn't a retailer, it was in fact a tech company, I felt like I should have studied computer science because I'd get into the room with a bunch of engineers and I could never figure out if the estimates they were giving me were accurate, if what I was asking for was too big or too small. And fast-forward, now I can build a prototype myself. I can go build a tool that automates really mundane tasks that I would've used to have to have gotten on a roadmap or waited for engineering bandwidth.
(02:08):
And so now I can build it, my team can build it. It's great.
Noah Levin (02:11):
Okay. So before we get into what you built, we should level set for everyone on what Thrive is. And then we were talking about is Thrive a tech company? So tell us what Thrive does and whether you consider it a tech company.
April Lane (02:22):
Thrive Market is an online grocer on a mission to make healthy living easy and affordable for everyone. And so with that, we're an online grocer. We're a membership model. You pay $60 to join Thrive, and with that you get access to our curated assortment of healthy essentials. It's everything from snacks, pantry staples, frozen, it's really across the board, and vitamins and supplements, health and wellness products. We're trying to meet the needs of the health and wellness consumer.
Noah Levin (02:53):
What's it been like to adopt AI as an organization? And Nick, who's the CEO and founder, what's Nick been saying about AI that's kind of led us to what we're going to look at today?
April Lane (03:05):
As an exec team led by Nick, we got together in Q1 and we said, look, the world is changing and it's changing faster than we ever thought. Our engineers were using AI already and they had access to Claude and all these tools. And we started to realize that this is actually something that's not just going to be kept within the confines of engineering. This is something we needed to give access to the entire company. And so we very quickly said, here are all the tools. We opened the Pandora's box, we gave everyone access to Claude.
Noah Levin (03:36):
I'm going to interrupt you. You gave everyone access to Claude?
April Lane (03:39):
We gave everyone access to Claude.
Noah Levin (03:40):
How many employees are we talking about?
April Lane (03:42):
Our corporate staff is around 300.
Noah Levin (03:44):
300 people are getting Claude installed on their computer. They're getting -
April Lane (03:48):
Unfiltered access to Claude.
Noah Levin (03:50):
Was there any limit on usage, tokens?
April Lane (03:52):
No.
Noah Levin (03:53):
So that's a pretty bold bet.
April Lane (03:55):
It's a very bold bet. And our idea was we would give everyone unfiltered access, we would see what happened. And the way we talked about it internally is said we're going to let the wildflowers grow. We're going to let every team, every person experiment. We're going to encourage them to, we're going to give them time to experiment and then we're going to see what happens. We're going to see what gets built. We're going to see who our super users are. We can put caps on usage and tiers on usage later. We'll start by just giving everyone unfettered access. And it was fantastic. Those first couple months were truly incredible. Of course, we had the skeptics, we had folks in our organization who were like, "Ooh, what is this? What are the sustainability impacts of this? Is this something that we should really do as a mission-driven company?" But overwhelmingly, people started using it and they started realizing that they could take the just boatloads of manual work they were doing and make it happen a lot faster.
(04:54):
And as we were experimenting, the tools kept getting better and better. And so I think the combination of the acceleration of AI, the acceleration of the model development and just the integrations when Claude's integration with Excel came out, oh my goodness, that was huge for our teams.
Noah Levin (05:10):
And that's because your team, the merchant team does a lot in Excel. You run your business on Excel, right?
April Lane (05:15):
Excel and Google Sheets, yes.
Noah Levin (05:16):
Let's pick an example. Think back to the early days of rolling out Claude where it was like Wild West and you had no idea what was going to happen. Do you have any memory of first example or two of someone doing something where you're like, "Holy shit, everything's going to be different now?"
April Lane (05:31):
There's a couple of examples. I think the first one is the one I was planning to show you today, which is our ingredient vetting portal. So as I mentioned earlier, Thrive's an online retailer with incredibly high ingredient standards. We ban over a thousand ingredients in our catalog, which means that every product that our team is thinking about selling goes through the gauntlet. We take that brand, we look at the ingredients behind it. We want to make sure everything that's food is non-GMO. We want to make sure it's free of artificial colors and flavors. We want to make sure it doesn't have preservatives that we don't like. Our standard list is incredibly long. It's so long that none of us can memorize it. There's too many ingredients that we ban. And so historically you'd take a new idea, a new product that you'd think you wanted to sell, you'd put it into our spreadsheet and you'd say, okay, I want to sell this item.
(06:21):
Then that spreadsheet would go over to our food safety team. They would look at it. They would one by one research every ingredient. Now they would cross reference if we would basically VLOOKUP a spreadsheet to our list of banned and researched ingredients and then one by one they'd say, okay, well this ingredient is restricted. It's palm oil. We need to make sure it's RSPO certified or it's high fructose corn syrup. Okay, that's banned. And so then they'd go through the entire list of ingredients. They would send that manually back to the category manager. That would take weeks. Category manager would then take that back to the brand and say, Hey, we can carry this or we can't or we can only if you can prove X, Y, and Z. And so then that loop would take another few weeks, if not longer. And so end to end, it would take us sometimes up to six months to launch a new item because we were going through this really thorough vetting process.
(07:19):
There's other reasons why it was long too. We were uploading everything by hand. We didn't have a PIM. And so what I think is really interesting about the introduction of AI is it came at this perfect time for us as a business. It came right in the midst of a serious push for automation and tools because we're now at a scale where automation and tools make sense. Prior to a couple years ago, our business was growing quickly and not quite ready to invest in the kind of tools to basically make people's lives easier on the team. Instead, it was really just focused on every ounce of engineering was focused on how do we make sure our product is as good as it can be. Now we're to the size and scale where of course we're still investing in the product. We need to
Noah Levin (08:09):
Make
April Lane (08:09):
Sure the product is perfect for our members, but we are also starting to invest in tools. And so we had just launched a new pricing tool. We had just launched a PIM, both from third parties, great SaaS companies. Now with AI, we're building those tools internally instead of having to go buy them. I think for a business our size and scale and for context, we're at about 1.7 million members. And so that's a pretty decent number of customers who are shopping Thrive every day. So when you think about how do we measure the impact of, do we want to spend this amount of engineering time on a new feature that's going to make those 1.7 million members' lives better, or do we want to spend that same amount of engineering time on a feature that's going to make a team of 10 people's lives better?
(09:02):
And so it's a really hard scale to tip. So we've really been cautious about how much we invest in internal tooling, which is why when I started, our merch team was run off of about 125 Google trackers. And I made an admission that, okay, as we continue to scale, if we want to be a multi-billion dollar retailer, we're going to have to get out of these Google trackers. We're going to have to build systems that are durable that can withstand the type of pressure we're going to put on them. And so we took that, we did some time studies. We looked at where we were spending our time, what products were the most manual, what teams were spending the most manual workload, and then we started to triage and prioritize. Eventually we brought on board a PIM so we'd have our catalog, have a single source of truth.
Noah Levin (09:53):
What does PIM stand for?
April Lane (09:54):
Product information management system. And so a PIM in the retail world is basically the single source of truth for everything you sell. There were still parts of the process that frankly, there weren't third parties we could go buy a solution from. Ingredient vetting was one of them where there wasn't a place where we could go buy exactly what we needed. And so the fact that we can now build it in a bespoke way and not have to use hundreds of hours of engineer bandwidth to do it is incredible.
Noah Levin (10:23):
So you invested in your PIMs in... So it's PIM?
April Lane (10:27):
PIM, yeah.
Noah Levin (10:28):
So you invested in your PIM in 2024?
April Lane (10:31):
2025.
Noah Levin (10:32):
2025. Before you could vibe code things internally yourself?
April Lane (10:36):
Correct.
Noah Levin (10:37):
Do you wish that you had vibe coded this yourself or do you think about... Because I think there's this notion that SaaS is dead, you can vibe code anything for yourself. Why is PIM something that you wouldn't put in that category and where you're happy to have a vendor relationship?
April Lane (10:54):
Two reasons. One, one of the reasons why we picked Syndigo specifically as a partner is because there's so many other retailers using them. So our brands are incented to keep the data in that software very up to date because it impacts not just Thrive, but all the other retailers that leverage it. So there's something about the scalability of that that actually makes it easy for our brand partners. They don't have to input their data into 20 bespoke tools. They put it into one and then it gets syndicated out, which is actually I think a lot more efficient than everyone vibe coding their own PIM. The second is it really needs to be production to grade software. Our catalog information needs to vend a hundred percent of the time and it needs to be accessible with the right security externally and internally. It's to a level where it's important enough in our system that we need a team of people somewhere, in this case at our third party, the SaaS provider who are keeping that software up to date at a hundred percent level.
Noah Levin (12:06):
I love that answer so much. Okay. Two lessons I heard from that. One is if you're working with a system that is your source of truth and where lots of downstream operating processes depend on it, you have to take that system seriously.
April Lane (12:20):
It needs to be a tier one system.
Noah Levin (12:21):
It needs to be a tier one system. That's Amazon language for, it doesn't go down, has to work. And so it doesn't mean that you couldn't code it yourself at Thrive, you could.
April Lane (12:28):
Totally.
Noah Levin (12:28):
But you would have to put real engineers on it, spend actual time on it, and you'd be making a much more consequential build versus buy decision. You have to maintain that and so on. And then if you're a SaaS vendor, part of what you just said about Syndigo I think is so important is they're not just selling you software, they're selling you a network effect.
April Lane (12:45):
100%.
Noah Levin (12:46):
And so they're doing something useful with their data and their position in the market that's above and beyond something that you can code yourself.
April Lane (12:53):
Great. Exactly.
Noah Levin (12:54):
Did we do it?
April Lane (12:55):
We did it.
Noah Levin (12:55):
Okay. SaaS is not dead.
April Lane (12:57):
SaaS is definitely not dead.
Noah Levin (12:58):
Not investment advice. That said, there are all these tools that are core to the workflows that your team does that you empowered them to go build for themselves. Set the stage for the dojo and how you incepted this process into the broader team.
April Lane (13:13):
So when we gave everyone access to Claude, we wanted to see what the whole company would build. We also wanted to bring everyone together to get folks inspired because we did feel pretty strongly that you'd get pockets of the organization building, but not everyone. The thing about AI that's so powerful from my perspective is you take people with the most domain knowledge and if you empower them to actually build directly, you end up building better solutions more quickly if they're hands-on keyboard involved. And so we wanted everyone kind of hands on keyboard involved with building these different types of tools. So in order to inspire them, we get the whole company together twice a year. We do Thrive Week every June and we do Winter Week every December. And so for Thrive Week this year, we said we are going to do a science fair of all the different things people are building around the company.
(14:08):
And so in order to inspire that, get people up to the right level of quality, we started this process called a dojo. The dojos really came about by our VP of platform engineering, Michael Winslow. He came up with the idea and he said, look, why don't we pair these non-technical builders with our AI engineers? We have a handful of really adept AI engineers and they were to pair together to really help empower, help instruct, help mentor our non-technical builders. Across the company, every team nominated two or three people and they had to nominate people with a particular project in mind. Those projects ranged from how do you build what we call mission control, which is a dashboard for our fulfillment centers where the leadership of fulfillment center sees every task they need to accomplish that day. They see everything they need to know about their team, their productivity.
(15:07):
That was a really incredible build out of Scott's organization. In my organization, we started with one of our hardest problems, which is ingredient buying. It's how do we make sure that everything we sell meets our standards? How do we vet products quickly? We've got a bright young woman on our food safety team, Marya Caples, who is our senior manager of food safety. She'd been really passionately starting to build, so we nominated her for the dojo. She'd been already working on this prior to the dojo, by the way. She had V1 built. We nominated her for a dojo. She built a really incredible tool that now our entire team relies on day in, day out to vet all of our products, but she built it within that environment. She got to work with one of our best engineers. She got to truly pressure test it, add a ton of features, learn really how to take it from being built.
(16:03):
She started building it in Claude chat actually, not even Cowork. She'd built it in chat initially and she, through many iterations, has taken it and turned it into a much more robust tool built with Cowork. In doing that, everyone got to build their software. We brought the whole company together in this science fair-like environment. We built these really fun posters to highlight all the projects.
Noah Levin (16:32):
Every app that was built got its own little movie poster,
April Lane (16:36):
Right? Got its own movie poster. So we had 20 something movie posters. We hung them up all around the office and we spent a half day where the builders each had a screen, they demonstrated their product to the rest of the company. After we did that, our AI usage again skyrocketed. So in the beginning we saw a lot of people experimenting, but it was probably a third of a team working on it. After our dojo, all of a sudden that adoption rate skyrocketed and everyone's using it. And that's when we had to start thinking about token limits because all of a sudden our AI bill got very large. So then we started to segment our users and start to add tiers, but that was a really interesting way to kickstart it and it felt very grassroots, it felt very fun. It's a great way to highlight people who are inventing, and we did it in, I think, a very thriving way.
(17:28):
We made it fun. We made it have a sense of community. We wanted to highlight folks across the organization.
Noah Levin (17:36):
Something that you said very quickly that I think is worth calling out is what was Michael's title?
April Lane (17:41):
VP of platform engineering.
Noah Levin (17:43):
I think there's something special worth calling out about him being the driving force behind a non-technical facing AI initiative because a lot of IT companies, and you didn't say IT, but there are a lot of IT teams out there that are more in buying and configuring software land, and a lot of engineering teams at companies that have non-technical teams working with them, they don't have that posture by default. The posture is guardrails and is lock it down and manage costs. It's trained before you let people use. And this sort of let's all get in a room and make it into a fun science fair model is like, first of all, it's very thriving, but it's a great way to nurture and experiment and see what works before you double down and invest.
April Lane (18:30):
Yeah. We also wanted to do two things with it. One, we wanted to really shine a spotlight on the people who were taking risks, going back to are we a tech company or a retailer? We're very much a tech company in our DNA, and so we wanted to make sure that everyone felt like they could be builders. We wanted to highlight that. The other thing we wanted to do, back to your comment around training, is we actually did want to make sure that everyone got trained and we thought not only is it important that they see what's possible, so the dojo was all about the art of the possible. We wanted everyone to be able to get inspired by things that other parts of the teams were building and get access to that. But we also brought in experts. So we brought in every very top tier AI consulting firm, come and do training for the entire organization, and that was really powerful.
(19:19):
We had done a session with them earlier in the year as an exec team where we got our hands dirty, building tools for our own teams with their help, which was fantastic, but we wanted to make that accessible to everyone. Again, everyone within the company, very thrivy mentality. So we brought them in.
Noah Levin (19:37):
What do you think Every taught you through their trainings that you couldn't have learned on your own or taught yourselves?
April Lane (19:45):
I don't think there's anything bespoke that we couldn't have taught ourselves. I think the spirit of it was actually, they made it feel very easy. I think when you are a non-technical leader or a non-technical team member, you are looking at a tool like Claude and you're not going to know that where to even start. You can get on YouTube and watch a bunch of videos, but you're not going to know what is an MCP server, what's a skill. They broke down those concepts in really simple ways, in really user-friendly ways. And so by the end of the session, we were all building skills. We're all building skills to help Claude write our voice.
Noah Levin (20:28):
And then you have a common language you can use to talk about what to build next.
April Lane (20:31):
Yeah. Exactly. And there's something about having a professional trainer come in that you don't need it for everything, but there are some topics where it's just better if you've got someone who understands not just the subject matter expertise, but also how to communicate it, and I thought they were great.
Noah Levin (20:50):
One more thing before we look at the laptop and actually get to see some of these tools. I think you said something that it mirrors what I try and tell people all the time, which is that you really want the domain expertise and the AI expertise to live in the same person.
April Lane (21:03):
Yes.
Noah Levin (21:04):
And it's hard to appreciate from the outside just how nuanced some business processes are, but man, to be a user and build the thing for yourself, not only do you take away some of the communication gaps, but you just build this iteration cycle that's a thousand times faster because you're just watching it happen in front of you and changing your mind again and again. There's really nothing like building your own tools. When you're able to take someone out of a team, let's say there were 10 people on a team doing a process, and one of them happens to be the AI native one, that person can be so powerful in developing tools for that narrow team, whatever they specifically do, that you really couldn't replicate it by bringing in someone from the outside.
April Lane (21:46):
Totally. Actually, it takes me back to part of the genesis for the dojo was actually to be able to embed those AI native people within our teams. So once someone had been through a dojo, they spent two days with one of our incredible AI engineers, then they were so much more confident. So then they came out and helped their peers. And so you would see an entire team get better at AI because one person went through this experience. And so we very much were intentional about the people we chose for those early dojos because we wanted to make sure they were people that would have a network effect within the organization. And to your point, there's nothing like the end user being the builder because they're so close to the pain, so close.
Noah Levin (22:37):
There's this other positive externality of doing what you just said, which is that these people are AI native, whether or not you use that or acknowledge that at work, and this is the direction the world is moving.
April Lane (22:48):
Totally.
Noah Levin (22:48):
By giving them a place to put roots in and grow their branches, they're able to build this capability in this part of their career at Thrive. And if you didn't do that, they would leave and they would go find a place where they could do it.
April Lane (23:02):
Totally. I think just to double click on that a little bit, not only is it a career development exercise, we're investing in our people, giving them access to these tools, giving them the time to experiment and learn and bringing in resources like Every to help teach. So there's a great development opportunity there. But what's even more important or more impactful is through this process, we actually took a lot of folks who were AI skeptics and turned them into AI natives. And that I thought was the most interesting. We'll talk about the ingredient vetting portal. That was a tool built by someone who was a very early adopter on our team. Fast-forward, our RFP tool was built by someone who was very initially skeptical, was raising a lot of concerns, but she saw a problem in her business. She saw something that took her and her teammates a ton of time, and she taught herself how to use the tools to build a better RFP tool for the entire owned brand team.
(24:01):
And her teammates were so grateful that they nominated her for Thriver of the Week, and just a couple weeks ago she won Thriver of the Week because of the work she did on this tool. You see this great shift happening where in the beginning of the year, I told the team, "Hey, it's a gift that we're letting you guys all spend the time learning this. You need to take advantage of this, jump in headfirst." Not everyone listened initially, but by the end, by now, I can sit here 10 months later and say, "Everybody on my team is using AI every single day." And not for everything, but they're using it. We were talking before we started about how I use it every day. I've got a morning briefing tool. It's really simple. It connects into Slack, it connects into my email, but it launches first thing in the morning at 8:00 AM.
(24:56):
I open it up and it gives me an outline of the day as if I had a personal assistant and if briefing me on the day gives me a heads-up on all the things I need to respond to, additional context. It's great. It takes five minutes to start my day with that. I read it, and then I go quickly do the top things it tells me I need to get done. It's fantastic. We've had other tools built like the ability to automate our weekly business review metrics. So now on Mondays, instead of the team cobbling together slides from this dashboard or that dashboard, Claude does it for us and the deck is automatically created. Now then the team can spend the time looking at the numbers, analyzing it, writing in their commentary instead of creating the dashboards or creating the slides. It's huge.
Noah Levin (25:47):
Let's look at the ingredient vetting portal so that we can make this real for folks.
April Lane (25:50):
That's right.
Noah Levin (25:50):
Okay. So now enter the ingredient vetting portal thanks to Marya.
April Lane (25:55):
Marya and Michelle.
Noah Levin (25:56):
And Michelle.
April Lane (25:57):
Michelle is on our nutrition and regulatory team. She's the expert on the vetting portion of it. Marya was our AI native member of the food safety team who jumped in to help build.
Noah Levin (26:08):
Okay. So how does it work now?
April Lane (26:09):
So it works really simply. We've got an interface that everyone on our team can access. It looks like this. There's a new submission form. You click what standard you're looking for. So first food and beverage, then you take brand type. Is this a branded product or is it owned brand, meaning it's a Thrive Market product. In this case, we'll use branded. And then just to play with this, I'll take a product that I know will fail the ingredient portal. So we'll take Oreos. So let's say Nabisco comes to us and says, "Hey, we want to sell Oreos on Thrive." This has not happened, but as an example, you would take the ingredient list, you'd paste it in, and then you would say, "Analyze ingredients."
Noah Levin (26:48):
And this is being hosted on Google Apps Script? Yep. Is that right?
April Lane (26:53):
Yep. Google Apps Script is where we host it, and it's internally only facing at this phase. We would love to make it externally facing. And so one of the things that Michael and our platform engineering team are looking into right now is what's the right way to host these things externally so that we could expose a version of this to brands themselves. So rather than having a conversation with our category manager saying, "Hey, is this going to be Thrive approved or not?" They could figure that out in advance, which we think would be great. So as you can see now, the analysis is complete. And so if you look at the ingredients, you'll see they're really coded in three different ways. Items are either acceptable, restricted, or unacceptable. So you'd look at this product. In this case, you'd see Oreos. They have artificial flavors and high fructose corn syrup.
(27:42):
Those are two ingredients that we do not allow under any circumstance, so this item would be banned. With that, if you wanted to, you could submit it to food safety for further review. In this case, as a category manager, I wouldn't because it includes unacceptable ingredients. However, if it just included restricted ingredients, then we might need more information from the vendor. We might need to understand, okay, is the soybean oil in this from a non-GMO source? Is the palm oil in it RSPO certified or not? And so we would go back to the brand to get that additional information. In this case, with high fructose corn syrup and artificial flavors, it's a no-go. But this tool has really sped up that initial analysis. So now our nutrition and regulatory team doesn't need to look at every single thing that a category manager is thinking about.
(28:34):
We only look at it once we've decided to submit it because we think it's going to be accepted.
Noah Levin (28:43):
What's happening behind the scenes here? Where's it getting the data or what ingredients are approved and what may not approved?
April Lane (28:49):
So we fed it our ingredient standards, which in order to put it into this format, we did a lot of work to make sure that our ingredient standards were incredibly comprehensive and that it would catch things that were obviously acceptable. Wheat flour isn't something that you'd need to put on an acceptable ingredient list, but in order for this tool to work properly, we did. So we had to take all of our acceptable ingredients, all the restricted ingredients, and all of the banned ingredients and really build that into a really comprehensive ingredient standard. The other thing is we had to create permissioning. So there's different parts of the team that have to access to different parts of the tool. So there are submitters and there's approvers. So our category managers, our own brand managers, they're the folks who are determining our selection, they are submitters. The food safety team, which is kind of the enforcer of our ingredient standards, they're the reviewers.
(29:45):
And so you had to create access tools so that, heaven forbid, I don't think anyone would do this, but you wouldn't want to see them to be able to submit something and then approve it themselves. You want to keep that division and appropriate permission structure. There's a couple different views for that. You can actually look at it. If you are looking at it at my level or at our team level, I can look at metrics to say how many ingredient submissions are in the pipeline, how many have been approved to date, what's the median time to approval? Are we getting to our queues fast enough? How many are closed within seven days, which is our goal? Who owns different things in the process? Is it sitting in a CMS queue? Is it sitting in the vendor queue? Is it sitting in the food safety queue?
(30:31):
So now I have an auditing mechanism that I didn't have before when this was all happening over email and in spreadsheets. And it's a chance for me to really check, okay, are we making our new SLAs? Do I need to resource the team differently? Do we need to adjust something about how this process is working? It's really interesting. It's been game changing for our team.
Noah Levin (30:56):
I think that what I love about this tool is that even though ingredient vetting might not feel like it is the strategic pillar upon which all of Thrive is built -
April Lane (31:06):
Oh, it totally is.
Noah Levin (31:07):
It's our
April Lane (31:08):
Secret sauce.
Noah Levin (31:09):
First of all, it's your secret sauce. And so as an external person, you might not know looking inside of Thrive that this is actually what's happening. But also you identified a constraint, a good old operations class. We found the bottleneck in SKU onboarding, and then you created a solution to that bottleneck. Now not only that, but you also have this new visibility into that little part of the pipeline so you can keep finding the bottlenecks and keep alleviating them. And so now presumably the bottleneck is somewhere else. So we have to go find the next one, but that's how you attack a problem like this iteratively is you just look for the bottleneck, relieve it, move on.
April Lane (31:48):
The ingredient vetting portal's actually been through multiple revisions. So since it launched in June, we've continued to add feature enhancements. Now we've got ingredient standards templates that are downloadable from the tool that we can provide to our vendors. We've updated it so that we segment out each category really in a fine-grained way so that we're making sure that a vitamin and supplement product is being analyzed versus the vitamins and supplements ingredient standards, not general food and bev. So we continue to refine it, continue to add features, make it easier for the team to use.
Noah Levin (32:27):
I was asking you earlier about how the tools that your team is building get catalog information and other source of truth information. You mentioned that you built something called Genome. Can you talk about what Genome is? And I think don't tell me more than you know. Tell me what you know about Genome.
April Lane (32:46):
I did not personally build Genome, but again, the brilliance behind what Michael's done, we have Genome, which is an MCP server. It's basically an integration within Claude that pulls in information from Thrive and it tries to make our version of Claude more Thrive specific. So it accesses, understands different team names, it understands things about our catalog, things about our business. If you ask it to give you some data around Autoship, it knows what that is, which Autoship is our recurring cuff scheduled delivery feature. And so it's made it so that when you're just interacting and chatting with Claude, it's far more useful. On top of that, we've got these great integrations. I use the Claude integration into Snowflake very frequently, Claude integration into Excel very frequently, PowerPoint. And so between those integrations, between skills that folks build on the team, between the MCP server of Genome, we've been customizing our instance of Claude to really make it just easier and faster for the team to build and more consistent.
Noah Levin (33:58):
There is a right way to install Claude at a company or ChatGPT, and it takes something between a really business oriented lead and IT team and someone who's more technical building custom software to really make it not just a generic AI chatbot. It's something that's useful and endemic to the company.
April Lane (34:22):
I think when you are empowering non-tech builders, you want to make it as easy as possible. Make it simple, easy, and great for them to be able to create consistent product. I think that's the most important thing and Genome really helps that.
Noah Levin (34:38):
What is not gone well?
April Lane (34:41):
That's a great question. I would say there's a few areas that we are still working on. One, we've got some really big ideas for how to build really large scale tools. So for example, my team desperately wants a vendor portal, a place where brands can log in and get access to all things Thrive and have it link into our payment system, have it link into our brand insights dashboard, have it linked to our PIM, all the things. We started building that and there's a couple roadblocks. We've got a really great internal prototype version, but we've got to figure out how to make that secure. And to our earlier SaaS conversation, making that secure, externally accessible with always on runtime, that's not a small task. That's not something that a non-technical builder is going to be able to do. So we get bottlenecked waiting for engineering resources to come in and help or the right third party solution that's going to enable us to do the login provisioning for all these other external users to get into it securely.
(35:50):
So I think we've got really bold ambition, which is great. And sometimes I think we might be a little bit ahead of where the tooling is or where we're at least ready for it internally.
Noah Levin (36:02):
I think of it as concentric circles a little bit. And sometimes you'll see someone bring up a website with this amazing tool that they've coded for themselves and then they'll want to popularize it on the team and you look up at the address bar and it says localhost. And that's not to denigrate it. It's doing a useful thing for you in your web browser on your computer, but there's this uncanny valley thing of it looks like it should be able to sell over 100 bucks a month online, but actually there's this huge gulf between what you just built and what it takes to productionize that.
April Lane (36:39):
Totally. Well, and actually that's how we stumbled on Google Scripts. So we were building all these little apps that were local hosts and we were waiting for the tech team to come and tell us how to make these things accessible internally. And Marya actually stumbled upon it and launched things on Google Scripts before Michael figured it out. It was incredible. And so when we were going through her dojo, she ended up teaching our engineers a little something about how to make these things more useful, which she figured out by asking Claude. She said, "Well, how do I make this accessible to my teammates?" And went down this whole journey.
Noah Levin (37:20):
All right. Don't sleep on Google Apps Script. Is there anything that you wish you had known a year ago before you set out on this journey that would've made it easier?
April Lane (37:30):
I think a year ago I was still using ChatGPT to just ask personal questions, and I don't think I imagined how critical Claude and all these tools would become to our team. And so I would say the only thing I would do differently is give the team even more time to play, frankly, because the work they've built, the tools, the better data analysis, dashboarding, we've just been on this learning curve and that learning curve is only inhibited by time. So I think if anything, I would give the team more time to tinker and play earlier.
Noah Levin (38:14):
Is there anything that the Serious People audience can do for you if they want to be helpful to you?
April Lane (38:21):
Well, if anyone has a great way, a great tool we can use to take some of these ideas and make them external facing, we are all ears.
Noah Levin (38:28):
Okay, we'll talk.
April Lane (38:29):
Okay.
Noah Levin (38:31):
April and thank you for coming on the Serious People Podcast. This is really fun.
April Lane (38:35):
Hey, really good to see you.
Noah Levin (38:36):
You too. Thanks for listening to Serious People. If you want more conversations like this in your inbox every week, subscribe to Serious People on Substack at seriouspeopleai.substack.com. You'll get the episodes and practical notes in your inbox along with something to read and something to try. And if you're trying to make AI useful inside your own business, Serious People can help you find the right work, build the system, and put it into production with your team. Find us at seriouspeople.ai.