Procurement with a team of two: Chris Kelley's AI playbook

How Chris got into procurement

Michael Denari: Welcome, Chris. It's so good to have you.

Chris Kelley: Thanks for having me.

Michael: To start off, I thought it would be interesting to get the origin story of how you got into procurement. Most people don't go to college with a dedicated focus on it; they tend to wander into it. I'm not sure if that was the case for you.

Chris: There's a funny viral meme on that recently of somebody saying, "I really want to slow things down, and I really want to stop things from getting over the finish line." Honestly, my story is pretty easy. I went to school for finance and started out in banking. I was actually selling subprime refis in 2005 and 2006. That clearly wasn't a good future.

So I ended up interviewing for a procurement position at CVS Caremark in Woonsocket, Rhode Island. I didn't know what procurement was at the time. I had to use Wikipedia to look it up. I understood it to be purchasing, and I've been frugal by nature my whole life, so it felt like a good fit. I was very fortunate to land with a team that really needed people, and I felt like I was able to drive a meaningful impact early on. I got exposure to really big deals at a young age.

Then I moved to the West Coast for personal reasons and found myself in a contract job at Kaiser Permanente, which weirdly opened a door for me to go to Salesforce. Since then I've done a couple of jobs, mostly for tech companies, and I've been focused most of my career on buying services.

This is the first job where I have much broader responsibilities. Now that I'm at an AI company, we're much more technical with the things we're buying. You bring the practices you know to the table, but it's also an exciting time to be in buying, because the people who have been buying cloud infrastructure for many years are seeing so much change in how these deals are structured.

It's an exciting time to take on a different role because there are more tools available and a faster path to get up to speed. From my perspective, it's a really cool time to sink your teeth into something new and try to build something that will be sustainable for the future.

Procurement's reputation versus what Chris is building

Michael: I find it funny, the meme that procurement is here to slow us down. What's your mental model for how the industry sees procurement versus what you're trying to build at Vercel and the other teams you've built?

Chris: Really good question. Prior to this role, I mostly worked in very large organizations where there's a need for process and policy. Internally, procurement sometimes gets quite a black eye. When your job is to manage the process and the policy, not every stakeholder wants to hear that when speed is the utmost importance to them.

I feel like I've spent most of my career trying to endear myself to stakeholders: meet them where they are, get an understanding of what they need, and then go figure out how to get it better for them. A lot of people view procurement as "I have to jump through these hurdles to get this thing done." I view it as, let's make sure you're doing the right steps and making the right decisions before we go obligate company money to anything.

You'll hear me say a lot that we really need to scope out what it is we're buying before we have a conversation with the supplier, or even allude to wanting to buy something. Let's make sure we know what we want to buy, and then go look at that. Unfortunately it tends to be the other way around sometimes. Somebody finds themselves in a conversation being sold on something, and they didn't spend enough time and energy understanding whether it's the right solution or the right contract model. They don't do enough due diligence.

So part of my proposition is to help think about that ahead of time, go figure out the right due diligence, bring it into the fold, and work with that stakeholder to make sure we ultimately achieve what he or she wants to buy.

Customer service as the value proposition

Michael: Kaiser, Salesforce, now Vercel; those are very different companies. When you think about the factors that let you deliver procurement the way you want it to be, what tends to matter most?

Chris: Good customer service. I was in this field for a long time without really thinking through what value I bring to the table. After some soul searching, the thing I've been best at in my career is what I define as customer service: understanding your stakeholder and meeting them where they are. Nobody is looking for me to tell them what the process or the policy is. They're looking for me to help them achieve some end goal, so I need to figure out what that is and meet them there.

Right now, Vercel is growing like crazy, and speed is the most important thing. As I look at our process, the improvements I'm focused on are mostly about where we can speed things up. That comes in a bunch of different ways. You can parallel process certain things. You can automate some things. It's a constant look back at where the inefficiencies are in our overall process and what we can do to make them better.

Not every inefficiency goes away, unfortunately, because some of them are there for a reason. You want the right level of friction in any type of buy. You don't want to make it so easy that mistakes can happen. So I find myself hearing what stakeholders need and what that speed needs to look like, and then interjecting where I think it makes sense to take a pause or get another perspective.

I've been in this long enough to have a perspective on what we give up to achieve that speed, and to make sure stakeholders are aware of the trade-offs they're signing up for. In procurement, that could be a number of things. It could be signing supplier paper when that's not the ideal thing to do. It could be not getting competitive quotes and then worrying that you're overpaying. In my opinion, those are the low-hanging fruit as we help stakeholders think about any other factors that should come into play before we make a spend decision, and then be in the foxhole with them on whatever challenge we have to solve.

Sometimes negotiation gets a little uncomfortable for people who don't live in that space all the time, so my internal joke is to look at procurement as the corporate bad guy. Internally, people get frustrated because they feel like we're slowing the process down or enforcing a policy. Externally, suppliers certainly don't want to pick up the phone and have it be the procurement person on the other end.

If you can show your value by meeting your internal objectives while meeting the stakeholder's expectations, or at least have the stakeholder see you there with them while it all gets figured out, I think that's really important. Especially since I'm still relatively new at this company, it's important for my stakeholders to see me and the one individual on my team showing up every day, being in that foxhole with them, troubleshooting how to approach something, and bringing an outside perspective: "I hear you, but we have a perspective on how we can get this done more efficiently, and hopefully cheaper."

You get a couple of wins, you show your value, and the relationship starts from a much stronger place once they've seen you go through one of those cycles and seen the outcome.

In a company that's growing super fast, it's kind of obvious that not everybody anticipated this company would grow this fast. So there's a lot of low-hanging fruit where we put a deal in place 24 months ago that we've outgrown. Everybody in the equation recognizes we've outgrown that deal and it needs to be redone. So I come into that equation trying to work with all parties and be honest about the situation. For us, it's a lot about making sure we're seen as the company we're growing into and not the company we're coming from.

Michael: I love that.

Chris: It's a size thing. I feel very fortunate that we're growing the way we are, because honestly it makes these negotiations a little bit easier. It's certainly harder to predict from a volume perspective, but I'd rather have this problem than the inverse.

Running a two-person team in the AI era

Michael: 100%. Being able to pre-commit to growth, there's a lot more appetite on the vendor side to grab that logo. I can only imagine the leverage Vercel has in a lot of these negotiations. You're a team of two today, is that right? I'm assuming there are disadvantages to being that lean, but in this new world with AI, maybe there are some pros too?

Chris: I'm very fortunate. This is what I wanted to do at this point in my career. I had spent a lot of time in large companies with large matrixed organizations, and in the world of AI there are a lot of hurdles that come with that, versus being forced to solve a lot of problems from a wide lens.

Doing that with only two people has forced us to get very efficient at everything we do and figure out where we can build automations. We're so fortunate to work for a company with the right tooling. We've used our own internal tool, v0, to create a couple of things we now use day to day. I've reached out to former colleagues to say, "You should try this," or sent them screenshots of something we built, because it's so cool and would have saved us so many cycles and meetings in previous lives.

I'm constantly trying to be an advocate for my peer group, and not just to make people more aware of our company. It's to say, "Did you know we just did X, Y, and Z?" and give them the perspective that this is completely possible. If a team of two non-technical people can do what we've done, any team could get a lot more value out of AI than I think most are getting.

The vibe-coded training game

Michael: I heard one of these use cases was a video game that you vibe coded. Do you want to tell the audience about it? It sounds like an awesome use case.

Chris: This one's really fun, and I take zero credit for it. This was the idea of the one person on my team, Christina Nguyen. It was her brainchild, she built it all by herself, and I'm very proud of it.

We use Zip, and one of our weekly hurdles is that we want people to take training before they get access to a purchasing tool. We want them to know how to use it, why they're using it, and what level of due diligence we expect them to bring to the table. What we found in doing weekly trainings is that everybody's multitasking. It felt like we weren't getting a ton of value, because people weren't paying attention and we were still answering all the starter questions we thought we had solved for once they got access.

So Christina took it upon herself to use v0, one of our tools, and she vibe coded a Super Mario-style video game. Right now I believe it's 14 levels. You run your little guy across, jump up and down, and there are coins in the air that are little Vercel triangles. When you grab one, it gives you a fact about our policy, our process, or something we want you to know. You run along, grab the coins, and get to the end of the level like in Super Mario. But to go to the next level, you have to take a one-question quiz based on those facts.

That has forced the stakeholder to actually pay attention to what they're reading. It's added a level of interest and intrigue, because most people haven't done corporate training via a video game like this, and the feedback has been amazing. We've had such good partnership with the peripheral teams we work with: IT, security, accounts payable, legal. They've all given us a set of questions, and Christina has created a level for each of them so we're coaching our mutual stakeholders on what they need to know for the process.

People can get through the game in 10 to 15 minutes, and then we get a notification that the training is done. We're still working out how to auto-provision the license at that point, so right now there's a human in the loop. But we're sure we'll get to the point where it auto-provisions a license, and we'll have taken ourselves out of a half-hour training every week, plus half-hour ad hoc trainings at least once every other week.

So there's a material time reduction, but it's also a very cool use case. People walk away saying, "You used our own product to build this? This is amazing." So far it's been all positive reviews, which is wonderful.

Michael: I can imagine Vercel coming out with a new procurement solutions package for fellow procurement folks. I also think it's interesting how that signals procurement as a progressive part of the organization internally. That's someone's first experience with procurement, and you're changing that perception from the get-go, starting with something as basic as training.

Chris: It's awesome. Honestly, it was designed out of an efficiency need. We didn't have the bandwidth to keep doing these trainings, and that turned into an investment. It took hours for sure, but hours to create something we'll get value out of for the rest of time. We'll change the questions over time and further enhance it, but at this point we have a team of two non-technical people vibe coding a video game with zero support or help from the peripheral teams, who are all amazing and lovely but have a million things on their own plates.

When we want to call into those teams for help, we want to reserve it for the times we really need it. What we're realizing, and Christina certainly is, is that she can do more and more with the tools we have, so we keep looking for more places we can do this.

Automating a weekly approval flow

Michael: Tell us two or three other top use cases. The video game is very interesting, but what's next?

Chris: We have another one we've worked on. It's a specialty approval flow that we go through a lot. Right now we're monitoring a report coming out of Zip, and we're monitoring our Gmail inbox for that report to hit. When it arrives, the report gets picked up and compared to a Google document. A new tab is created on that document, and all of the metadata that didn't exist on the last tab is populated in the format we want, so we have it all tracked in a shared spreadsheet.

Then it goes a step further and loads all of that metadata into a Slack canvas, staging the messaging in the format we want. We can make any final tweaks before we send those messages on to our executive team. This process was taking probably four hours of our time every week, because it's a lot of deals with a lot of metadata. We're very interested in which of our suppliers are also customers, so we had to find a way to automate as much of this as possible. There's still some manual ownership, but we estimate it's saving us four hours of manual work each week.

It's impressive because it spans multiple applications: Gmail, Zip, and Slack. We've been extremely happy with the results. More than anything, it's shown us once again that a couple of non-technical people have found a way to start automating processes.

So part of our day-to-day is thinking about where we have inefficiencies in everything we do. At this point, we're using AI for everything. For the smallest of tasks, we usually have a skill already prepared, because anything that's rinse and repeat, it's "Okay, we need to use this." We've probably gotten to the point where we're creating a little too much, because we get excited, try to build something out, and then have to decide whether we're really using it. We have to be careful not to create skills that run behind the scenes when we're not using the output.

But it's very exciting, because for most things we go through, at some point we're asking: could we automate some component of this? Could we get ourselves to a better starting point? Being in procurement for as many years as I have, I've seen multiple teams show up to the same meeting with different sets of data or different presentations that everybody spent all this time on. Now we're trying to centralize everybody onto one set of data, and then use an LLM or v0 on top of it to present it back, to dashboard it back, the way we think it needs to be done.

We're immediately seeing value with our stakeholders. We've created a whole dashboard that we sit with stakeholders and go through. The access to information is great, and it centralizes everybody in the same place, so we don't spend any time debating the numbers. Instead we get right into the action item and the next step.

Which models the team uses

Michael: Before we move on to build versus buy, I'm curious: are you a Claude shop, ChatGPT? You're obviously using v0 for a lot of use cases. What's the day-to-day tool?

Chris: Because we have an AI gateway, we're contracting with all the LLMs; that's part of our job right now. But as a company, we use Claude more than any other. It's a personal preference thing. You can use what you want, and it depends on which teams are using it for what. I'd say right now there are a lot more of us using Claude than anything else.

I oscillate between "wow, Anthropic's going to take over the world" and feeling like some of these other models get traction really quickly, and that world could change much faster than before. It's interesting, because we have a front-row seat to what developers are using. You get a little flavor of "this new shiny toy is coming out of XYZ company and people are loving it," and then you get interested: could I use that? How would I use that?

For our two-person procurement team, we definitely use Claude a lot. But the dashboards we've developed and the video game, we've done in v0, because, to use your phrase, we want to dogfood our own product. We've had great success with it, and we've enjoyed understanding our product more by actually using it.

And the two of us can't help ourselves. We joke that in our personal social circles we're the people who don't shut up about AI, because we can't wait to tell somebody, "Hey, we vibe coded this video game. I bet you could do something like this in whatever function you own at your company."

Negotiating consumption-based contracts

Michael: Love that. Tokens are coming into more and more vendor conversations, and I'm sure you're on the front lines. What are you observing across your vendor base, how are you approaching those negotiations, and what's your tolerance for these more variable cost models? This is front of mind for a lot of folks in our industry.

Chris: I've equated this to the cloud computing contract model spreading into so many other places, because of the question of what the right amount to commit to is, and what happens above or below that line. We're in this fortunate position of hypergrowth, so we're dealing with a lot of consumption-based suppliers, and we're trying to figure out the right math on what the commitment should be. That's a game with endless variables you could throw at it.

What I've been focused on in the most recent negotiations is this: because of our growth, the traditional commitment model doesn't work for us. We spend all this energy and time figuring out what the commitment should be. Then if we're fortunate enough to bring growth into the relationship and reach that commitment, and the supplier's answer is "we'll just redo the commitment at that point," there's a lack of understanding. We don't want to start negotiating a new commitment in the middle of a contract when we thought we had all of that out of the way.

A lot of companies are willing to talk about discounting at a commitment level, and we'll have that conversation. But every one of those conversations with me right now includes: "I need to know what happens above this line. If you're not giving me a tiered pricing structure where my unit economics go down, we have a problem. If we stay on the pace we're on, we're going to outgrow you, and we're going to get frustrated with this relationship because you haven't found a way to grow with us."

Again, a wonderful problem to have. But maybe the model has been borrowed from the cloud computing industry and isn't being used in the best way by other providers who are trying to change their model and figure out how to pay for the compute power tied to it.

What we're seeing is that suppliers don't totally know what their compute costs are, because they're experiencing growth and compute challenges similar to other companies. For us, we really need to lean into this concept: we don't just want a conversation about how to get to a specific point. We need to know that if we blow through it, you'll be the right partner to grow with us.

It's an interesting world to be in. I've been in multiple supplier negotiations now where I realized this isn't something everybody has been experiencing. It's new, and it's more of a factor for companies that are really leaning into AI and seeing this explosive growth. We're an AI-native company, so we're reselling a lot of this, and our volumes are much higher than a company that's only consuming these tokens internally. There are a lot of variables we have to figure out within that.

Managing token spend

Chris: Then every company is going through this reality check of, "Okay, yes, use AI and find efficiencies," but somebody has to go back and scrub through where we're burning these tokens and whether we're still getting value out of them. Plenty of people are creating things that run in the back end. If you leave that thing on and you're not getting value anymore, is anybody paying attention to what it's costing you?

Michael: And recently tokens have been getting much cheaper across these different models. You mentioned Vercel has its own gateway, so some of these use cases are probably underwater from an ROI perspective today but might be okay in three months. Speaking for Zip internally, it's a challenge to keep up with. It's going to be pretty amazing to see the number of use cases that aren't worth it today get unlocked in the months and years ahead.

Chris: Or move it to a lesser model, right? Everybody wants the latest and greatest model to create what they need, but for some of the rudimentary stuff, the older models can do it and they're much cheaper. So how do we put a focus on that?

I think eventually every consumption-based supplier will need to give you some metering ability, so all users have a perspective on how they're incurring costs. I've seen a few cases where we had spikes in spend, and the stakeholder didn't realize what they were doing was going to cause that cost or compute output on the other end. The easiest way to solve that is transparency.

There's been plenty of internal conversation, and I've seen trackers built internally that people use to monitor their token usage across these platforms and make sure nothing gets out of control. To me, it feels like every big company using AI is going to have to build that muscle: consistently come back to the table and evaluate what you've built, what the continued ROI is, and whether you should let it keep running.

Getting leadership buy-in

Michael: For your team today, and over the course of your career, what's your philosophy for engaging with your leadership team? A lot of leaders end up as a workflow step in a tool like Zip, but there are broader strategic initiatives you're driving within Vercel. What are your tips for getting leadership buy-in on your objectives?

Chris: It's a great question. I have a very fortunate scenario here in that I interviewed with that leadership team, so I knew what they had to say before I came into this role. Having direct access to our CFO and being able to say, "Here's my perception of the challenge you need this role to solve, and here's my approach," and getting that buy-in from day one gave me immediate confidence. I knew what I needed to go do, and I knew I'd be aligned with leadership.

We're still a small but mighty company, so I can interface with the CEO plus ones as needed to get their involvement and buy-in on whatever deals we need. We're a very communicative company, so everything's out in the open. We're all on Slack, and we try to keep most questions and conversations in channels so everybody can access what they need to know.

We also use Slackbot to help us determine what's in Slack, what people are saying about a given thing, and how to quickly get an assessment of a project so we know what we need to know walking into that engagement. It's very helpful to have the tools on top of the data, so we show up prepared to drive the best value for those stakeholders.

Building a procurement team from scratch

Michael: We have some questions from our Zip community. Shameless plug: jump onto zip.com and join our community of practitioners. The number one question we got was: if you were building a procurement team from scratch today, what would it look like, using the lessons you've learned at Vercel and elsewhere in your career?

Chris: I'm in the middle of that right now. We just put out a couple of positions; we had one offer accepted, and we have another open position right now.

What I'm looking for is in some ways different from what I would have looked for historically. Because the tooling is so good and the access to data is so good, I don't necessarily think you need to hire the person with 20 years of experience in that particular category. The world has changed so quickly that the individual you need is somebody who's nimble and adaptive with new tools and understands that whatever process or policy we have today will likely look very different six months from now.

Because speed is such an important factor, I'm looking for autonomy. As I talk to candidates, I want to know I don't need to micromanage you. There's no time or energy to do that. You'll have all the tools available, and you can come to me with whatever questions you have, but we need to trust that you can lean in and do this work on your own.

So it's less about finding the long-tenured person who's been in the know, and more about finding the person who's very open to finding new ways to solve old problems. Our talk track focuses on flexibility and creativity, and in the interview process we're trying to determine whether this person exudes the attributes we'll need them to have in their tool drawer.

When you drop them into different situations, they can quickly come in and say, "Let me draw on some of my experiences. But for all the things I didn't know and didn't have experience with, here's the analysis I just pulled out of our existing tools and ran through a model, and now I have a pretty succinct bulleted list of what I need to know coming into this."

It certainly feels like the expectation now, when you show up to stakeholder engagements, is that you'll have a decent sense of what they're looking to buy or do. That's not because you have 20 years of experience doing it, but because you took the time to search and use the tools available to figure out the right starting point.

On the personal side, I have friends who've said, "I'm a little hesitant on AI. I'm a little scared of AI." Or the pessimists who say AI is just another software application. Personally, I'm always looking for people who are really interested in where this technology is going and what more can be achieved with it. Some of the things I've described to you are table stakes at organizations that are much further along in their tooling journey.

I find it exciting right now to understand what people are doing in their own time and who's got some exposure to it. I find myself in conversations with friends who explain, "I had to connect this application to that application via MCP," and you realize these are also non-technical people who are learning what they can do with all the tools now available to them. I have a close friend who was out of work for a while and created a website and started selling T-shirts, and he used our v0 product to do it.

Michael: I bet.

Chris: I think it's so cool that somebody has the ability and access to use the tooling that's out there and teach themselves something that new.

The individuals we'll want to hire in the future will be those very curious, very inquisitive people who can show us in a job interview, "Here's the way I've used the tools I've had access to. Here are the things I've built." People with a building mentality about what's possible.

I've worked for large matrixed organizations my whole life, and a lot of companies grew by headcount. If they needed to do more, they went and got headcount. Here at Vercel, it's possible to get headcount, but you really have to prove it, and the expectation is that you've already kicked the can, or whatever the phrase is, on whether you did enough due diligence. Did I automate what I can? Are there any other inefficiencies I should be jumping on before I revert to "I need to hire another body to do X, Y, or Z"?

It's an exciting time. It also sets a really high expectation bar for the type of person we bring in. Traditionally, you've probably done your role. Here, we'll expect you to do that times three, and you'll have to get creative in how you find a way to do it.

Hiring for AI fluency

Michael: It sounds like you have a good filtering mechanism in the hiring process, indexing more on creativity than experience. I'd be curious to get your take, because a lot of folks are going through this: how do you filter for whether someone is going to be AI-first, beyond the creative side? Are you looking for tangible examples? Are you putting them through a live test of using AI?

Chris: We put them through one. I had to go through it myself: you build a presentation within our v0 product. That was the first time I used it. On the next call I got on, I could not stop smiling. I was super excited: "Holy cow, this was way cooler than I thought it was going to be."

We've seen that in multiple interviews with candidates. They go through this, and all of a sudden their demeanor changes: "This is so cool. I didn't know I could do this." Going back to the idea that we don't necessarily need to hire the person who's done this a million times, you can see on their face that they understand the possibility in that moment and realize this is a very different approach to problem-solving.

It's that creativity and flexibility that I think will change how you hire for these roles, and who you hire. Over the last 20 years, I've known people who were brilliant at what they do in purchasing. Their brilliance in understanding those SKUs and how to negotiate them was so valuable then that maybe you overlooked some of the personality traits or other challenges that came with that type of person.

Michael: Very interesting.

Chris: I view AI as something that's going to level the playing field for the individuals who see it as a tool to pick up and figure out, as opposed to "this is something my company is forcing me to use," or "I'm hesitant because I don't want it to take over my job." I think those are naive perspectives that are really going to hurt people over time. The technology's not going anywhere.

Michael: Right, the opposite.

Chris: The question is how to dig in and get value out of it.

One thing to start, one thing to stop

Michael: One final question for you, Chris. For our audience, what's one thing you would stop doing and one thing you would start doing in the world we're heading into?

Chris: That's a good question. I would start thinking more about what I need overall. It's hard to take a pause from your day-to-day work and think outside the box about what could make a task easier. We're very focused on our day-to-day, but if I had to start over, and we're kind of doing it now, I would require myself to do a monthly check-in on where I spent my time over the last month, and whether there's another way to solve that problem. All too often I'm not doing that enough.

What I would stop doing: personally, I say yes too often. I find myself wanting to appease and work with all of the people who are asking for help, and I could get better at ruthless prioritization. For me and my currently two-person team to keep getting better, we need to find a way to draw more lines of materiality. Maybe it's not a no. Maybe it's "I personally can't be the man in the box on that, but here's a solution you could use."

We've flirted with providing benchmarking details to our stakeholders. We have access to a benchmarking service, there's tons of information online, and we've built a couple of scrapers that scrape X and Reddit for certain things. Equipping our stakeholders with those kinds of resources would probably be an easier way to get us out of the day-to-day grind on smaller deals that aren't moving the needle long term.

Vibe coding from a chairlift

Michael: Makes a ton of sense. Chris, really appreciate you spending the time here; awesome insights. For our audience, what I'm hearing is that they need to go check out v0 and build their own vibe-coded app.

Chris: Yes, 100%. Use v0; it's going to make your life a lot easier. You can also just have Claude tell you how to use v0. Tell it the application you want, and Claude will give you the prompt to put into v0, and it'll build it for you. It's easier than you think. I have one other example. Can I share one for fun?

Michael: Absolutely.

Chris: During my interview process, I wanted to use v0 for the first time. I sat on a chairlift and said, "I'm going to build a flight tracker. I want to know the price for all direct flights between San Francisco and Boston." I skied down the run, got on the chairlift, and the application was built for me.

It wasn't perfect, so I added a one-sentence prompt to put a day-to-day filter on it. I skied down the run, sat on the chairlift, and now I have a working flight tracker that my wife and I use. It was all vibe coded with two or three sentences. It's amazing what the technology can do at this point. Just lean into it and try using it, and you'll immediately find some way it provides value for you.

Michael: Orchestrating your agents between ski runs. I love that. Good way to finish off. Thank you so much again for being on, Chris.

Chris: No worries, Michael. Thank you very much for the time.