The Science of Scaling · 12 Mar 2025
How to Hire Sales Reps for AI Products w/ Lauren Nemeth (Fmr COO, Pinecone)
These are notes on the conversation, checked against its transcript. The episode itself has the full discussion.
In brief
Lauren Nemeth, former COO of Pinecone, a vector database company, talks with host Mark Roberge about leaving Twilio's large sales organization to lead sales at a roughly 40-person, $25M Pinecone team. The episode covers how to hire sales reps for AI products, how Pinecone narrowed its ICP and reworked pricing and packaging, and how its sales process was built around customer use cases. Its central argument is that a top sales performer is worth more than AI knowledge because AI is very teachable. Mark adds frameworks for outside-in sales process design, three-lens pricing, and ROI-based ICP selection.
For founders
- Define your ICP by where customers get the most success and ROI, not where lead volume, lowest CAC or conversion rates are highest.
- Build pricing around three lenses: the ROI the customer can measure, the cost of the substitute solution, and your own unit economics such as LTV to CAC and payback period.
- Design the sales process from the buyer's view outside in, rather than starting from a deck of features, benefits and problems solved.
- Before setting prices, go talk to developers about what they would pay more or less for.
- In PLG, minimize up-front friction so end users can feel the product's value, then monetize at the enterprise level through controls like IT lockdown or CFO volume pricing.
For revenue leaders
- Lauren Nemeth would take a top sales rep with no AI experience over an AI-savvy middle performer, because AI is very teachable.
- Test coachability in interviews by giving a candidate one piece of positive feedback and one area to improve, running a role play, and having them redo it.
- Prioritize sales hires who have sold to the technical buyer persona they will face, since selling to developers differs from selling to HR or CMO buyers.
- Expect AI sales playbooks to need rewriting within three to six months, so the methodology should not be treated as fixed.
- Focus the sales team where meaningful AI workloads are concentrated, which for Pinecone meant ISVs and software companies.
What was said 25, most useful first
Would rather hire the top sales rep and teach them AI than hire an AI-savvy middle performer. Listen
In a hypothetical comparing a top sales rep with no AI knowledge against someone who knows AI well but ranks around 500 out of a thousand reps, Lauren Nemeth said she would take the top performer. She said AI is very teachable, just like any technology, and that she came to Pinecone with no AI experience herself.
“I'll take the top sales rep every single day.”
Listen to the episode Sales team, hiring & comp Link to this Report a problem
Give a sales candidate one positive and one improvement point, then have them redo the role play to test coachability. Listen
Mark Roberge describes testing for coachability during the interview. He runs a role play, gives one piece of positive feedback and one area for improvement, and then watches whether the candidate listens, takes notes and applies the feedback. He has them redo it, either in the same interview or the next one.
“I always say I'm giving you one piece of positive feedback and one piece of need for improvement.”
Listen to the episode Sales team, hiring & comp Link to this Report a problem
Many startups design their sales process inside out, starting from a deck of what they built instead of the buyer's view. Listen
Mark Roberge says that when a founder moves to selling and thinks it needs a sales process, they usually build a deck covering what was built, the features and benefits, and the problems solved. He calls this an inside-out approach and argues it goes against sales research. He recommends an outside-in process that focuses on how the buyer sees the world before they know the product exists.
“It's what I call an inside out approach.”
Listen to the episode Sales process & deals Link to this Report a problem
Define your ICP by where customers get the most success and ROI, not where lead volume, lowest CAC or conversion rates are highest. Listen
Mark Roberge argues Lauren Nemeth's ICP logic was about where people are seeing the most success and generating the biggest ROI, which he calls the value creation endgame. He says this points to product-market fit. He suggests plotting customer size against customer industry on lifetime value and customer value to decide where to scale.
“She said where people are seeing the most success.”
Listen to the episode Strategy & market Link to this Report a problem
In PLG, reduce up-front friction so end users get real value, then monetize at the enterprise level. Listen
Mark Roberge describes Lauren Nemeth's view of PLG pricing. Friction up front stops users from feeling the power of the tool, and monetizing the end user is common but less ideal. The better model is to let end users go hog wild and monetize the business, through IT lockdown or CFO-controlled volume pricing.
“we let the end user really experience the tool and we monetize the enterprise.”
Listen to the episode Pricing & packaging Link to this Report a problem
Price against three lenses: the ROI the customer can measure, the cost of the substitute, and your own unit economics. Listen
Mark Roberge describes a triangle of pricing lenses. The first is customer ROI, where a CFO buying big products might want to see a 12-month ROI and the price must be multiples lower than the value. The second is the price of a substitute, including a direct competitor or a combination of tools. The third is the company's own economics, such as LTV to CAC or payback period, and he hopes the three lenses land on the same price.
“The first framework that you see on this slide is a triangle that represent the three dimensions by which you want to think about your pricing.”
Listen to the episode Pricing & packaging Link to this Report a problem
Founding teams often ask sales leaders for experience with their technology and industry first, even though a strong sales performer may matter more. Listen
Mark Roberge says that when early-stage startups and their boards look for a sales leader, the first thing they ask for is experience with the company's tech and industry. He contrasts this with experienced leaders such as Lauren Nemeth, who would take a top sales performer and teach them the tech and industry.
“The first thing they say is experience with our tech and experience in our industry.”
Listen to the episode Sales team, hiring & comp Link to this Report a problem
Coachability is an attribute that is hard to interview for but matters a lot in sales hires. Listen
Asked what she learned from reflecting on the reps she hired, Lauren Nemeth named coachability as the attribute that stood out. She said it is hard to test for in an interview. Mark Roberge noted that coachable people learn fast, which matters at startups where the product is often being sold for the first time.
“it's hard because it's hard to interview for this, but coachability.”
Listen to the episode Sales team, hiring & comp Link to this Report a problem
Look for sales hires who are technical, curious and actually love the technology when selling AI products. Listen
For Pinecone, Lauren Nemeth said she wants exceptional sales people who are hungry, motivated, curious, technical, agile, strong relationship managers and solutions-oriented. She said not all salespeople are technical, and that the reps need to genuinely love the technology.
“you have to be technical by nature, which not all sales people are technical, and you've got to actually... love the technology.”
Listen to the episode Sales team, hiring & comp Link to this Report a problem
Prior experience selling to a technical developer persona is critical when hiring reps for a developer-focused database. Listen
Lauren Nemeth said many Pinecone reps came from other database, AI/ML, dev tool and infrastructure companies. She said they understand the developer buyer, which is very different from selling to heads of HR or CMOs. She called experience selling to a given customer persona pretty critical.
“having experience selling into that by customer persona is pretty critical.”
Listen to the episode Sales team, hiring & comp Link to this Report a problem
Lauren Nemeth learned AI by talking to customers use case by use case rather than playing with the product. Listen
In her first months at Pinecone she spent time with customers and asked what they were building, whether the work created new revenue, improved search or delivered efficiency or productivity gains. She also spoke with partners about the competition and the ecosystem. She said this learning never ends because AI keeps changing.
“So just going in, use case by use case, customer by customer to seeing what are people building”
Listen to the episode Strategy & market Link to this Report a problem
Pinecone adds roughly 4,000 to 5,000 net new developer sign-ups every week, mostly long-tail developers still working out what to build. Listen
Lauren Nemeth gave this figure for Pinecone's developer top of funnel. She said most of these sign-ups are long-tail developers trying to work out what they want to build in AI. At the high end, the customers are sophisticated enterprises and ISVs, such as Notion, whose assistant product runs on Pinecone.
“we have about 4 to 5,000 net new developers sign up for Pinecone every single week.”
Listen to the episode Pipeline & demand generation Link to this Report a problem
Pinecone narrowed its focus to ISVs and software companies, where it sees the biggest AI workloads. Listen
Lauren Nemeth said Pinecone stopped trying to sell to everyone, since sales teams running in 50 directions is flawed. Based on evaluations of existing customers and conversations with partners such as Anthropic, she saw interest consolidate around ISVs and software companies. She said a vector database that prides itself on scale, high throughput and data freshness needed to focus on those major workloads.
“You'll see sort of a consolidation of interest really around ISVs or software companies today”
Listen to the episode Strategy & market Link to this Report a problem
Rag chat-over-text applications were the largest workload at Pinecone but did not create ROI for the ISVs using them. Listen
Lauren Nemeth said rag is probably the biggest workload coming into Pinecone and is the talk of the town. However, many rag use cases were simple chat over text, and those did not create ROI for ISV customers. She said the clearer ROI comes when AI takes over human tasks through agents, gen AI applications or lower hallucination rates.
“those weren't actually creating ROI for those ISVs.”
Listen to the episode Strategy & market Link to this Report a problem
Before setting prices, go talk to developers about what they would pay more or less for. Listen
Asked how to handle founder, product and engineering resistance to charging, Lauren Nemeth said she started by talking to a lot of developers about what they would pay more or less for. Enterprise-grade needs such as security, compliance and audit also came up in those conversations.
“First, it's go sit and talk to a ton of developers today. You know, what would you pay more for or less for?”
Listen to the episode Pricing & packaging Link to this Report a problem
Enterprise buyers pay for security, compliance, auditability and professional services, which Pinecone used to shape its paid offerings. Listen
Lauren Nemeth said enterprise-grade companies want security and compliance, the ability to audit, and lots of professional services. Pinecone used intake from customers about what they valued and were willing to pay for to shape its packaging. She also said other database companies with transparent pricing were useful to study.
“They want the ability to audit. They want lots of professional services”
Listen to the episode Pricing & packaging Link to this Report a problem
Pinecone built gold, silver and bronze packaging tiers based on what different customer types would be willing to pay. Listen
Lauren Nemeth described packaging in tiers, similar to gold, silver and bronze, matched to what different types of customers might pay. The free tier is there so users can get a modest workload running and understand how Pinecone works before paying.
“we built kind of like, think of like gold silver bronze type packaging in terms of what different types of customers might be willing to pay.”
Listen to the episode Pricing & packaging Link to this Report a problem
Pinecone rolled out many features without repackaging them and gave a lot away in its free tier, so pricing and packaging needed a big overhaul. Listen
Lauren Nemeth said Pinecone launched many products across security, compliance, higher throughput and evals, but never repackaged them. Much of that value was given away free. She said pricing and packaging has since been a pretty big overhaul.
“There's so many features that we kind of rolled out that we never repackaged. We gave a lot of way for free in our free tier.”
Listen to the episode Pricing & packaging Link to this Report a problem
Many enterprises exploring AI had no large, meaningful workloads or significant investment, which made motivated buyers hard to find. Listen
Lauren Nemeth said Pinecone spent a lot of effort talking to enterprises about their AI plans. Those conversations were intellectually interesting but rarely led to meaningful business that could be closed. She said finding a buyer with a real AI workload at scale, developer resources and knowledge was hard up front.
“they don't really have large meaningful workloads and they don't really have a ton of investment”
Listen to the episode Sales process & deals Link to this Report a problem
Enterprise buyers resist adding another data subprocessor, so Pinecone works through cloud and partner integrations to remove that hurdle. Listen
Lauren Nemeth said nobody wants another data subprocessor. Pinecone tries to remove that objection through its partnerships and by integrating into the clouds. She also said data ingestion (chunking, parsing and moving data) is hard, so Pinecone invested in documentation, best practices and products like Pinecone Assistant.
“nobody wants another data subprocessor today, right?”
Listen to the episode Sales process & deals Link to this Report a problem
AI buyers asked for lower costs, so Pinecone needed more opinionated case studies showing productivity and revenue gains. Listen
Lauren Nemeth said many large customers needed to bring down costs to meet the internal metrics for their AI products. That made clear Pinecone needed to be more opinionated. It needed concrete case studies and art-of-the-possible examples of where people were improving productivity, efficiency and making money.
“we need to be more opinionated. much more in terms of case studies and art of the possible”
Listen to the episode Positioning & marketing Link to this Report a problem
AI sales playbooks may be obsolete within three to six months, so the methodology cannot be treated as fixed. Listen
Lauren Nemeth said the sales playbooks Pinecone runs are very unlikely to be the same three to six months from now. Mark Roberge noted that building a methodology, playbook and battle cards can take three or four months, and that it can all change quickly. Pinecone tracks competitors, S-1 filings and customer model usage to keep up.
“the sales playbooks that we're running, very unlikely those will be the sales playbooks that we're running even three to six months from now.”
Listen to the episode Sales process & deals Link to this Report a problem
Judge the business in order of people, then performance, then process, looking three to six months ahead. Listen
Lauren Nemeth said she learned this framework from a former boss. People comes first: whether the right people are in each seat, whether they are motivated and understand the strategy. Performance means the health of the business and dissecting the funnel step by step. Process means building for where the company will be three to six months out.
“So people. performance and process. People is always the first thing that I look at, right?”
Listen to the episode Leadership & culture Link to this Report a problem
Early AI application products have mostly been iterative, with copilots making existing tasks around 20% more effective, while disruptive products would remove most of a task. Listen
Mark Roberge says he thinks the first phase of AI at the application layer has been iterative. He contrasts that with disruptive use cases that remove around 80% of a task and rethink it, which he says can make the incumbent technology irrelevant and give an attacker an edge.
“You know, I think like this first phase of AI at the application layer has been very iterative.”
A hypothesis Mark Roberge hears is that AI lets software skip structuring data, which is a massive architectural shift and could challenge incumbents. Listen
Mark Roberge says he believes a common narrative is that earlier software vendors had to structure data because of architecture limits. With AI, it may be better to let unstructured data go in so the AI can choose how to structure it for each business and use case. He asked whether this is an innovator's dilemma for current category winners.
“that you're better off letting the unstructured data go in the AI so the AI can choose how to structure it for this particular business and context in use case, which is a massive architectural shift.”