Philipp Stelzer: Hello, and welcome to another episode of the RevOps Lab Podcast. My name is Philipp, and I'm here together with Janis. Hello, Janis.

Janis Zech: Hey. Hey. How's it going?

Philipp Stelzer: Okay. And our guest today is actually someone who's been here now for the — or he's here for the second time, Eric Portugal Welsh. Hey, Eric. How are you doing?

Eric Portugal Welsh: Good. Good. How's it going, guys?

Janis Zech: Yeah. Great to have you back. Great to have you back. The best beard on the podcast so far.

Eric Portugal Welsh: Alright. That's what you — it's my playoff beard. My wife is pregnant. It's a little bit of a superstition I have till Kate is born.

Janis Zech: Congrats. Congrats. Good news. Alright. Cool.

Philipp Stelzer: Yeah. Eric, for those who don't know you yet, who are you and what do you do?

Eric Portugal Welsh: Yeah. Well, I'm Eric Portugal Welsh. Neither Portuguese nor Welsh. That's just how the cards lie. I am currently the head of RevOps at a database company called PlanetScale, formerly the head of RevOps at the global payroll and workforce management company, Deputy, which I believe I was at when we last got in.

Philipp Stelzer: Yeah. Alright. Perfect. Okay. And our topic here today is how AI can help a RevOps team of one, which I think is a very hot topic. I'm not sure if the word vibe coding will come up. I have a feeling it will. So, Eric, I'm curious, like, why is this an important topic for you? Are you a RevOps team of one right now?

Eric Portugal Welsh: Yeah. Yeah. So I came from being a team of about six, or leading a team of about six. So I had a lot more firepower from a human perspective. Came over here to being a team of one and working at a company that really values high output and high performance. PlanetScale is really designed to kind of keep headcount numbers low and give us the really fun and performance-based AI platforms and tools that can help us do the job of like a handful each. So this really — I was looking at it from the outside in previous to PlanetScale like, oh yeah, AI is nice and all, but I don't really need it that much — to, oh yeah, I need it every day all day in order to do the job of like six people.

Janis Zech: Yeah. So would you say like you made yourself six hundred percent more productive with AI?

Eric Portugal Welsh: I'd say I'm getting there. I'm definitely maybe two hundred to three hundred percent more productive. But I mean, this is gonna be fun to talk about. Like, I'm only scratching the surface and I'm starting to get deeper because I have kind of daily inspiration from some of the most brilliant engineers I've ever seen and gotten to work with here. So it's been — I feel like the last three months have been a very intensive bootcamp for me.

Janis Zech: Yeah. Okay. Maybe to get started, like ease into the topic. Can you describe just, like, where you are right now in terms of your setup and sort of, like, what you use AI for?

Eric Portugal Welsh: Yeah. I can give you a lay of the land on, like, the tech stack and what I started with and what we have. PlanetScale is a Notion company. We use Notion as our knowledge base and intranet. We also have access to — I can't remember what the name of the actual tool is — Nightshift, the AI agents, native AI agents that they have. I have access to Cursor, Claude, CoWork, Claude Code, and then Gemini. That's kind of my AI platform suite right now. The way we have Claude and Anthropic set up internally is it's connected to most of our internal systems. And so I tried to take advantage of that from the get go when I started at PlanetScale. I think one of the interesting things is when you're a manager of a large team and you're constantly onboarding new people, you have to build out these elaborate onboarding plans, find documents to reference that are stale, maybe still relevant, help get training materials together for new systems. Using all of our integrated systems here, I built out my own onboarding plan, and I did so because I'm the only RevOps hire, my boss is the COO, he's not gonna be able to go deep on the systems that we have. So I built out my own plan. I just built an agent in Claude and in Notion. I did one in both just to check to see how close they would come back to build out that first thirty day onboarding plan for me. It referenced all of the right materials and documents. And I didn't really have to go to people for the first week to ask questions. Notion and Claude just had the answers for me based on like everything they had access to. And it had all the historical context. I think one of the cool things also is like it has Slack context too. So like any public Slack channel conversation, like it could reference for me. So if anything had been talked about in the world of RevOps or anything related to it, I had context in my first couple days.

Janis Zech: So you found out that all the different definitions changed three times very quickly. Saw all the process maps, all the systems that got onboarded and retired.

Eric Portugal Welsh: Was great and a little depressing at the same time, I would imagine. It got me — you know, when I talk about like operating at that like two hundred, three hundred percent capacity, like it got me there quick. You know, what would have taken a couple of weeks, got there in a few days.

Janis Zech: Yeah. I mean, and realistically, if you talk to anyone, right, most likely they won't remember or, you know, they don't really have the reference point. It's very inefficient. So if you basically can tap into the entire knowledge of the company, obviously, that's extremely unique and useful. And I think we all know the situation even just in Slack, right, where you — oh, where was this? And then was it in Slack? Was it in email? Which channel was it in? It's just, like, painful. Right? Like, and I think everybody knows that feeling, especially in RevOps where you get pinged all the time from many different areas at scale.

Eric Portugal Welsh: I've got solutions for that one too. When we're ready to talk about it.

Philipp Stelzer: And which system, like, is most effective for you at the moment? Like, it's just like speaking for myself, like, I currently have a ChatGPT subscription, a Gemini Ultra subscription, a Claude subscription. Right? And we use Notion as well, so use Notion AI and stuff. But I'm always falling back to Claude at this point, and I'm not sure whether it's just like some kind of bias because everyone has been talking about Claude so much, or whether it's really just because the output is so much better, but I basically just stopped using ChatGPT entirely. Like, it serves zero purpose for me nowadays because I just like, the style of Claude just works better for me. And if there's ever a moment where I need to create, like, an image, then I just use Gemini because that's anyway included in our Workspace subscription. So, yeah, like, how is that for you?

Eric Portugal Welsh: It's similar. I mean, I would definitely say like if I have intense web research that I need to do, I'll use Gemini's deep research. I think that works really well. As we know, like Google owns the internet. So if you're gonna do web research and you don't need to go like too deep on people, Gemini's great. I also use it for image generation if I need to, but I don't really need to do that very much. It's usually just for a comic meme or something. Claude tends to be the one I use the most. I use the desktop app and I use CoWork almost more than just the chat. Actually, I use it a lot more than the chat. I have a lot of plugins to Notion, to Slack, to Clay even. So it just works better for me. I found this out because I ran a handful of tests — and we're gonna talk about vibe coding in a little bit — but I did run a handful of tests against different reporting models that I wanted to ship out to our team here. With like building out pretty intense spreadsheet models as a starting point. And Gemini was awful. Like really, really bad. I have examples. I'm not going to show the examples, but I have some examples of — you know, Gemini versus Claude Opus, which is what I was using — Opus, I think four point five at the time, now four point six — built me some very robust pipeline prediction models. The output I got from Gemini was a rose on a spreadsheet. It was comically bad.

Janis Zech: Was like — I actually talked to Kai this morning and he showed me our new financial model, which he built with Claude. And I was like, oh my god, this is like intense. Right? Like, so it's yeah. It's wild.

Eric Portugal Welsh: It's pretty crazy. I did that. And then I started to dabble with like Cursor vibe coding and building an actual app for the pipeline model. And then pipeline trackers and trackers for our biz dev team and our AE team and all the standard RevOps funnel metrics that you can imagine. And I did that with the help of our finance lead who is an accountant and an engineer by trade. So she got really into Cursor vibe coding. And if you'll believe it, she hasn't used or built anything in a spreadsheet — this is somebody on finance — hasn't built anything in a spreadsheet in I think like three or four months. She doesn't even go into Excel anymore. Everything she does is vibe coded and built into an app now.

Janis Zech: Okay. That's crazy. I mean, I don't shed a single tear for Excel. But it is surprising that Gemini is so bad at building spreadsheets given Google Docs and Google Sheets. I mean, it's huge. But I do have a similar experience. And, like, what is, like, crazy to me for Claude is, like, Claude is so proactive in just building you, like, a custom app. Like, sometimes I even have to tell it, like, stop. Stop. Like, not another custom app for this. Like, just actually give me a spreadsheet. Like, she actually wants a spreadsheet for this use case because it's easier. And like, an app brings its own, like, problems with it also. Right? Like, then you need to maintain it. Other people don't understand it and so on. But maybe that's, like, a good segue to start talking about, like, the vibe coding piece. Right? So are you currently vibe coding for yourself, or are you vibe coding for the benefit of others?

Eric Portugal Welsh: Yeah. Yeah. Yeah. I mean, a bit of both. I mean, personally, after starting a few work projects via vibe coding, I started building apps to make my life at home easier. I can talk about that in a minute, but built a really fun one. But yeah, it started like I think any RevOps person would start a vibe coding project. Like I inherited a new instance of Salesforce and needed to understand like where things were and what was going on. We had no real pipeline metrics and nothing set up in Salesforce to traditionally track pipeline velocity, conversion rates, didn't have fields like timestamping for your stages — kind of just like the basic stuff that you would expect out of a legacy instance to have. And so I started — where you would normally start is like a history report, right? A history report has a lot of duplicate entries. It was really difficult to like, if you just put it into a spreadsheet, really difficult to manipulate and figure out what your conversion rates are and your pipeline velocity is. So that was my starting point. I just got one single report, put it in Cursor and then began writing like everything I wanted out of it. And it was like intimidating to try to build an app to start. But once I got started, once I built out a plan doc in Cursor, then started asking questions and then started prompting the agent to actually build, I was able to get to essentially what you would traditionally have built into like a full pipeline dashboard in like a ThoughtSpot, a Domo or whatever BI tool you're using. And that only took maybe an hour and a half to do something that my previous company would have taken weeks to build. This just took a couple hours. It was great and got me really good insights into where our funnel is and, like, how things have progressed over the past few years.

Janis Zech: And do you feel like the data is reliable so you would also share that with the rest of the team? Or is this rather to quickly onboard and understand, get some initial views — like how robust is that? Right? Like, how much can you trust it?

Eric Portugal Welsh: Probably a combination of both. Like, one, it gave me good insights into like where things might be a bit messy. And I did share it with the team. I shared it with our COO, Kevin, maybe to gut check to see where things might be off because he has the most context. And Megan, our Head of Finance. I think things like that are good to push and get public early so that if there are issues, we can all see them in real time. Part of that was to show where things are actually a mess. We're involved — we're doing a full rebuild of our lead to customer funnel right now. And that includes our deal funnel.

Philipp Stelzer: One thing with vibe coding that I — and I think this is why Claude is also so preferred as a solution — is it's asking you questions before it starts. Right? Like, in some cases. And what I found is to really speed up, like, my own vibe coding — and I've said this word now, like, more often than I think the entire year so far — but, like, whenever I do vibe coding for myself, I basically first create, like, a basic description of what I want. I put it into Claude or some other LLM, and I basically tell it, create, like, an acceptance criteria document for what I'm actually trying to achieve. It then creates the acceptance criteria document, and then I take that document, and then I put it back into Claude, right, and say, like, and now build this. And just like the returns are so much better. So I think that's something I just wanna share with our audience because I think it does work to just tell it to build something, but it's definitely worth to invest those five minutes in the beginning or ten or whatever because, like, the output will just be, like, so much more reliable and, like, yeah, bug free than if you just give it a basic description and say build.

Eric Portugal Welsh: Yeah. It's a bit like the typical traditional process of building product, right, where you would do that anyways just in one person and in a very time constrained experiment, right, which is actually really cool to see.

Philipp Stelzer: Yeah. It's a no brainer, but you gotta think about it. Right? Like, basically.

Janis Zech: Yeah. I'm curious, like, obviously, right, like, you know, understanding pipeline health risks, right, metrics, always crucial. But, like, you know, are there other things you've done, you know, to basically clone yourself and just deliver more impact on the RevOps side? Curious what other use cases you're seeing.

Eric Portugal Welsh: Yeah. There's a ton. And these are — for a RevOps person, they might be exciting, but they're kind of like the boring ones that just help remove the tedium of the day to day. We have a lot of tedium all the time. I'll premise this by saying like, I found it with the stack that we have to be really important to write down as much of the work that I'm doing as possible in the platforms that I'm using. So I document just about everything in Notion and Slack because those are the two places that are really important for what I'm about to talk about. The other thing is internally we have adopted the idea of recording every call and every meeting that we're in — even if it's in person, we record it — and edit with the Notion meeting recorder app. I found that to be really helpful when ideating on different ideas because I can then take that transcript and then build like the outline of what I'm gonna do or even like start the build using the AI, like the Notion AI or Claude. Those two things are important because there's a couple of things that I've found to be really helpful being a team of one — communicating like the priorities and the work and doing the kind of standard stuff of making sure everybody knows what's above and below the line. I have a couple different agents running right now on a weekly and daily basis, depending on priorities and stuff. One is I have a weekly summary of everything that got done for the week that goes — that pushes to Slack. And this is all built in Claude CoWork, or Claude projects, sorry. That just pushes like, here's everything that we talked about and here's everything that got done and here's everything that's on the agenda for next week. It gives people the opportunity to kind of like plus one or minus one stuff that my people have aligned on. That I think is kind of table stakes. There's nothing really exciting about it, but it takes one thing off my plate and I don't get constant questions from colleagues about what are you doing? Where's so and so and such? The other thing is I have Notion agents running in several — I got this idea from the rest of the company — where we have a handful of agents running in different channels to allow people to ask questions and get actual responses. And so I have the same going in certain channels. If people ask questions directed at RevOps or directed at me, it'll scan Notion and reply with a response with an answer. That's kind of a time service desk automation, but on steroids and not just for intake, but also support, you know, dashboard questions, metric questions, definition questions, you name it. Like all the typical stuff that comes up.

Janis Zech: Yeah. Okay. Mhmm.

Eric Portugal Welsh: Yeah. It's also — for us, like our flow, it's really important. We have most of our Salesforce opportunity data and call intelligence data pushing into Notion too. We have a weekly pipeline meeting on Monday mornings. Typically what you would do in kind of your standard procedures like Friday afternoon, you'd run, you'd go into your pipeline reports or your forecasting reports and you'd look at all the data, write down notes, ask questions. I built a Notion agent that just does that for us and pushes it out. I think I have it set to Saturday or Monday mornings at like seven AM. This call is at eight thirty, gives you enough time — gives you the executive overview of like everything that we got going this week, all future calls, all previous calls. It does it based on the way our COO likes to communicate. That's another fun one. I also built a Claude project that allows me to give him updates in the tone and like the detail that he wants specifically. So it analyzes all of his Slack conversations, picked out his tone, picked out all the questions he tends to ask and then preempts all of it. So all I have to do is input my update for him or question for him and it spits it out in language that he loves. I saw a noticeable change in how he responds after I started doing that. But it uses a handful of those communication agents that I've built in this sales pipeline update. And so far it's been a pretty good success. It gives you a really quick snippet to show where things are for the week.

Philipp Stelzer: Just a quick question for the audience — how big is the sales team? How big is the company overall? Because obviously I think it's — I'm pretty sure, like, you're all listening here, and you're like, well, okay. You know, I see a bunch of different challenges with permissions. Right? Like, you're kind of a very engineering-forward company, probably, like, access to everybody. Everybody has open access to everything. Right? You record every conversation. Everything is like going into the context window of the LLM. So I'm curious just to share for the audience.

Eric Portugal Welsh: It's a good call. We're a company of about seventy five people. And I would say — I haven't done the full math on this — about eighty five percent of the company is engineers. So it is very much like an open data concept here.

Janis Zech: Just one remark here. I have to actually constantly smile because this is what you're basically building through different systems — that's essentially what Philipp and I are building with Weflow, very much from the data pipelines to automate activity, conversation data capture, field data capture with an app, to then the query layer, the unification layer of CRM activity, conversation, field data, to essentially then being able to either use chat based systems or workflows and agents to query the data and then push the data or reports into Slack or email or different other places, Notion and so on. It's actually so interesting because there's — I think the reality that you're describing, and I think this is how startups are run today. This is how we run. This is how you guys run. But if you look at scaled setups, five hundred people, different hierarchy, most companies don't run that way. And so there's then complexity to get into that kind of place. But I think it's fascinating to hear you talk about this because it's basically — this is going to happen to everybody. And this is awesome. Right? This is, like, really a superpower because in my mind, right, like, what you're basically able to do is you can basically orchestrate the entire workflows end to end. And so you sit there and you really do resource allocation on steroids based on huge context data that is available to us. And so even you describing that you document all the internal meetings, that you push more data into Notion to have more context data so that you can basically have better outcomes and better output — I think that's just a reality of what's coming and we can embrace it, and then this gives RevOps superpowers in my mind. Because I don't think that the VPs of sales or CROs of the world will sit there and build something like you're describing, even if it's a prebuilt system like Weflow, right, where it's easy to configure. It's still fairly complicated to — I mean, it's not complicated, but it's like it's a lot easier maybe to do what you're doing because you have to build that yourself. But at the same time, it's also something that you need to then think about the different use cases. So it's fascinating to hear you explain this, actually.

Eric Portugal Welsh: Yeah. One, just to give the VP of sales credit for stuff that I'm seeing them building — our head of commercial has done a fantastic job. He has leaned in really heavily on agentic workflows and doing a lot of this stuff. He built a really, really great Notion agent that can take all the context data from Notion and builds really good sales sequences and sales cases. And our outbound motion is really new and so we're iterating on a lot of different ideas and trying to figure out which play, which kind of motion we move forward with quickly. And being able to spit out really good content in the PlanetScale voice, like in our brand, has been extremely helpful. I know that being at a really small company like this with access to just about every tool I can imagine makes it a lot easier. But those kinds of things are just — it's really good to have a partner who can just do that. And then all I have to do is take that content and spit it into the systems that we're using.

Janis Zech: One hundred percent. I mean, this is — I mean, this is why working at, like, an early stage company can be maybe sometimes a bit more rewarding in the sense that you have, like, certain freedoms that you don't have elsewhere. But then I think it's also understandable. Right? Like, so one thing that we do at Weflow is, like, we automatically respect your hierarchy in Salesforce, so you can only access the data that you should be allowed to access, which then obviously is, like, a huge pain point. Right? So if you just — and a lot of criticism, I think it's just, like, you know, if you work at a company with, like, five thousand people, you just can't throw everything into Notion. It's just, like, very hard to do. Right? Like, the security and compliance team will freak out and for good reason. And then you need, like, these safeguards. You need to kinda, like, anonymize before you do things, and then there's GDPR and CCPA and yeah. All the crazy stuff that slows you down, but it's also there for a good reason sometimes. Always depends. So yeah. And, yeah, I think, like, as soon as you, like, hit, like, a certain threshold, it becomes really hard to move that fast. But I do wanna say that all our customers from big companies, the ones that use our AI the most, are VPs of sales, CROs. Yeah? So they have lengthy conversations, you know, trying to understand, hey. Like, who's doing this? How are we preparing for this big deal? Are we in a good position to actually, you know, renew that account? It's those guys. Like, it's not the sales reps. I mean, they also use it, but, like, a lot less. It's the VP of sales. And I get it because, like, that's their core problem. Right? Like, huge amounts of unstructured data, and no one can pull a report for them easily to give them, like, a simple answer to what maybe sounds like a simple question, but it's often just very hard to answer.

Eric Portugal Welsh: Yeah. For sure. And it's funny because, like, two years ago, I would have said VP of sales, maybe, like, less tech savvy overall if you compare it, like, you know, with other people in a company. And, again, it's not a criticism to VPs of sales. It's just I think also, like, they're super busy, and they have too many things to do all the time. But the chat interface and the agents that just push reports directly into their inbox — that is just like it feels natural. Right? Like, it's just like normal, natural usage of technology in a way that it just wasn't available before. Like it's not a Tableau report. Right?

Janis Zech: Oh, I was gonna say, I think it's honestly — it's yes. And it's less about getting agents to push reports. It's more about giving them the context behind the report and giving them the ability to ask the report questions and get the context that they need.

Eric Portugal Welsh: That I found to be like the way things are going, the way things are moving for this cohort of startups right now that can adopt these tools really fast and get the most out of them. The concept around building Salesforce reports for anything other than auditing and updating fields is gonna go by the wayside, I think. The way people are gonna interact with their CRM is going to be in the chat interface. Those who like adopt a Salesforce or HubSpot are going to do so for just like data input and entry, but they're going to expect to interact with it via chat agents. They're just gonna want to ask questions and get context. Nobody is going to go to Salesforce and look at a record page.

Janis Zech: I mean, this has been true for a long time, right? Like, this has been true for a long time. The only positive side now is that actually you really don't need to and you can unlock insights that you just couldn't unlock because there's too much data and especially unstructured data. The way I think of how this is going to play out is you essentially have RevOps orchestrating those agents, and essentially across your operating cadence and different questions and reports, thinking about what are the different executive reports you want to schedule, what are the actionability you want to drive, how can I then also automate entire workflows for the reps so that essentially on a Monday morning they all go get prepared into the deal review or forecast meeting, the pipeline reviews, that they know what's at risk, they know where to follow up, they have the pre-written email that has all the context data? And so I think the orchestration layer will continue to be RevOps. The interaction layer will be chat, whether that's in Slack, that's in a tool, that's in email, it doesn't really matter, but you want to basically have kind of a set system and then the ability for the end users, whether it's reps, managers, VPs, CROs, to have flexibility around, like, what else they get out. Right? And I think that kind of is actually — I really like that kind of combination. And then, yeah, many workflows will be just fully automated that just should also be automated, ideally. And I think with, you know, obviously meeting prep, meeting follow-up, CRM admin, all these — these are the earliest things. There's a lot more to come that you can automate, right? Like if you think about like presentations and stuff like that. So, yeah, it's a very interesting one and it's cool to see you building that at your current role, basically from scratch in an engineering environment where this is also something that often is lifted by the culture. It's really very insightful. That said, I think we're out of time. Final question: is there a book, research report — you can also do Claude's best report on how to build prompts or so — but what would you recommend? It can also obviously be a non-business book for our listeners.

Eric Portugal Welsh: Yeah, I actually built a prompt to help me build prompts in Claude, but that was a fun one. The book that I'm reading right now, and it was recommended to me by my dad and that I really like, is The Last Stand of the Tin Can Sailors. Really good book about the Navy in the Pacific in World War II.

Janis Zech: Awesome. Awesome. I think we never had that one. Really appreciate you coming on and sharing all these lessons learned.

Eric Portugal Welsh: Yeah. Thanks for listening. Thanks for joining.