Top Five Growth Lessons Scaling Stitchfix to IPO

so much for joining me today, Mike. Thanks for having me, Harry. I’ve been a long time listener and learned a lot from your show over the years. So thank you. Do you know what? I

Yeah. So I’ll go back and I’ll start with my undergrad. So I studied engineering undergrad. I finished up University of Michigan in 2007. Most folks at that time out of engineering school were either going to work in the auto industry in Detroit or moved to Chicago or New York to work in business. I’m using air quotes. Google was just opening up their Ann Arbor office at the time. And my objective when I was thinking about what I wanted to go do was really to go avoid routine, so avoid getting bored, and then go optimize for learning, which is a theme that’s held true throughout most of my career.

Right out of school, I joined Bain because of this pattern. I think from a distance, felt the breadth of problems and clients would offer that. I actually had a great time, and then if anything, it trained my brain to be a structured thinker. But I always wished I could have better aligned how I spent my time with what I actually cared about or had some passion for. At the time, one of my colleagues at Bain went to go work for Danny Meyer, then Grant Achatz and Nick Cocones, who were running this big restaurant group, Beelinea Group.

Yeah. Think at that time was the best restaurant in The US. And I was a food and restaurant geek. I think Chicago is one of the best food scenes in the country, and I was always deep studying the food bloggers and tracking new restaurant openings. I And was always inspired by that. So when I moved to the Bay Area in 2011, I started looking for interesting opportunities in the food space. You know, we’re looking at food startups. Turns out, these are really hard businesses, but I ended up working at this chef marketplace called Kitchit that I fell in love with, essentially a marketplace connecting chefs and notable restaurants with diners.

And it was really my first experience being the generalist athlete in a room of five to 10 people just trying to make stuff work early on. And I didn’t really know what growth was at the time, but that’s kind of what I was doing. One of the things we did there is we launched this group payments product, and it was a real needle mover on conversion. Shortly after, when I learned about this YC company, Tilt, what they were doing around building this crowdfunding or group payments API, I got really excited and I cold emailed James Becherra, who I think you know, and ultimately joined shortly after the company finished YC and raised its first round of funding from Andreessen Orwitz.

Most of the company at that time was writing code, I was kind of helped to figure out why the company was growing 20% month on month, and then help engineer things to help it grow faster. The stepping stone at that point, so our COO there, who hired me, Brian Burtwhistle, he was, I think, the first BD person at Amazon, ended up being a close friend and mentor. Three and a half years later, when I was leaving Tilt, which didn’t work out the way we wanted it to, he introduced me to his wife, Julie Bornstein, was the COO at Stitch Fix.

The story there was they had grown on the heels of organic and strong retention for the first few years. Famously And had a hard time raising venture money, so decided to just say, hey, we’re going operate under the constraints of profitability. And there’s a lot more color on that, but really, essentially, they never had to invest in building an in house user acquisition muscle until relatively late in the journey. So I saw this unique opportunity, and I was also coming off a time where I felt burnt by this high growth, high burn startup failure.

And so Julie took the bet that I was sufficiently analytical and competent to operate at that next level of scale, and that’s how I ended up there.

The first one’s probably about the importance of getting your objective function right. And so when I initially joined, there was a retention team, and there was a VP of retention who’s fantastic. She was my counterpart. And then I was hired to go run acquisition. And our early emphasis on retention meant that we had a strong understanding of our users and really an amazing foundation to add acquisition to. But with a structure like this, teams ended up thinking less holistically, really were focused on tools that they have within their disposal to move one particular metric.

If I viewed my metric as new users in CAC, I could have destroyed her retention metric downstream by just letting in shitty quality users and optimizing for an affiliate channel or something like that. And so it’s important for teams to have a holistic view of what growth actually means for that business. And for us, redefining our KPI to some down funnel metric was really impactful. The meta point there is getting your objective function right. I think as a related point to that, many companies today start doing things to grow without really having an understanding of their growth model or a holistic understanding of it.

Growth could be really detrimental if you’re not focused on moving the right set of metrics, and that’s a really important point that’s natched in me since then. The second point, which is a little bit more tactical, and it’s really around the importance of understanding incrementality. This is specific to paid marketing, but Stitch Fix had a lot of data science DNA from Netflix, and that was instilled in the company. One of the things Netflix uniquely did was they were pretty fancy at running at scale performance marketing.

All their measurement was based on these holdout tests. They didn’t use any precise attribution models. I think many growth marketers aim for this almost a false sense of precision to give credit to a conversion on any one channel or any fractional set of channels. Oftentimes, that’s just false precision. Even though we were highly analytical at Fix, we realized that marketing is rarely, like, really direct response, and the best way to really quantify efficacy is to measure lift by doing some version of holdout testing. So that one’s more tactical, but I think so many people could actually expect almost false precision out of paid marketing, and that was a really important learning

You’re right. It’s very inconsistent. I prefer a broad definition, which is someone who’s responsible for two things. One is accelerating the company’s pace of learning. This is typically done through experiments, but that’s a very important piece. Like, essentially, the growth team and the head of growth should be operating at a faster drumbeat than the company and bring those learnings back into kind of core product and what’s actually happening. And the second description might be engineering systems that help a company build more control over its North Star metrics.

And so that’s when I look at ahead of growth, that’s what I look for. If we think about increasing a pace of learning, what’s an example of that? Alright. Here’s a good example. If you look at paid marketing, and if you think about what a marketing team might be looking to do is set up a set of kind of different landing pages to actually capture different forms of intent, and better convert users. One way of actually capturing learnings on conversion would be to just up a page, send organic traffic through, and just see what happens.

The role of a paid marketing team here, or a growth team, could actually be, hey, we’re actually going to go up funnel, spend money in the spirit of learning, and actually drive much more traffic through a set of ads that are actually testing different messaging that will eventually roll into our landing pages. Again, that’s a very tactical one, but that’s like one example. I think maybe the better answer is, the way most organizations are structured have some form of dependencies. You have an engineering team, a product team, a marketing team, etcetera.

I think part of the magic of cross functional growth pods, which is generally the structure I prefer, is you can actually be fully autonomous in going and running an experiment to validate or invalidate a hypothesis without actually needing to step in the roadmap of other teams in a company. These are problems that happen more like at scale.

Yeah. So it’s interesting. There’s one meta point we talk about at Reforge. But one point that I very much agree with is if the practice of growth is successful, it should eventually abstract itself away and just be how product and marketing are run. So this is a discipline that I think can be instilled into other functions. But sometimes having a standalone growth team might not be the best long term answer. But yes, there is a role in many companies’ journeys where it’s really important, and it might be the most pragmatic solution.

So there’s typically structures. There’s like single function, which is actually how we had it at Stitch Fix, which is some VP of acquisition with different channel directors reporting to that. This is more common for e commerce companies or ones where it’s a pretty straightforward or linear conversion funnel. The second model, which I just referenced, is a cross functional pod where you have a PM, an engineer, analytics design, user research, working on shared business problems. Typically, PM might be, like, the most accountable person on that pod.

But in general, those are, everyone’s taking different disciplines and working towards a shared objective. So there’s no

What I was referring to was a growth team there. That would be the way the growth team is structured, and you would have a head of growth and various different PMs, and then those other functions, engineers, or whatever, either data lining in or actually sitting within the growth team. And so where it sits, was your first question, many successful companies have growth reporting into the CEO. And this is important given that you oftentimes need a blessing to just go run really aggressive tactics, And dependencies, like I said, could really hinder a growth team.

One way to solve that is just have it roll into the CEO. Between marketing and product, because the CEO at a certain scale might not be practical for that to be the case, I typically say this worked better under product. I think many sustainable growth programs require with engineering and product discipline. Marketing really only makes sense for e commerce companies where the product is like simple and linear. You’re likely going to have a performance marketing team sit within marketing, and that’s fine. Sometimes that’s used interchangeably with growth marketing.

But if we’re talking about growth programs that are going to touch the product, I think it’s much more ideal to have it sit under product.

Yeah. One principle that for sure is true, never hire a growth person pre product market fit. I do think it’s okay to have generalist athletes at this point, but they should really be talking with users and aggressively trying to find product market fit. And sometimes people who might have growth ish backgrounds could be helpful on that, but they should not be running a growth team at that stage. I think prematurely scaling top of funnel before you have a cohort of users that’s really retained, and also understanding who those users are, that could be detrimental to a company.

And so for sure, agree with you on that point. The second, analytics is a really necessary foundation for this. One question that’s related to what you’re asking is, who’s your first growth hire? What should they look like? Think for me, my first growth hire is typically an analytics person. In Silicon Valley, we talk about the concept of tech debt. Think that’s widely understood. I think analytics debt is just as common, if not more common. I’ve had a lot of success hiring people from biz ops teams out of a company like LinkedIn or elsewhere who are really SQL proficient, maybe engineering or stats backgrounds or math backgrounds, but they still understand the business context.

And so to be clear, this is not a data scientist. These are people that actually are analysts, but also are technical enough to be laying a lot of the analytics engineering foundation. Analytics, debt.

Yeah. A good indicator of analytics debt is if I come in and ask a very basic question, say, show me last month’s cohort of users and how they’re retaining a d 30, for instance. Not being able to answer that question in a simple way or a series of questions that one should be able to answer, or there being a really high cost to go answer those questions. Someone needs to go do a custom query of the database, and it takes a few days. That’s a good indicator that you have analytics debt.

So if I’m a founder Yep. I’ve got Google Analytics set up. Yeah. Is that enough for you? So that’s the problem. A lot of people go to visualization tools, and Google Analytics is a little bit different. But say Looker or Mixpanel, whatever, they go to visualization tools and say, hey, we have dashboards. Like, that’s we have our problem solved. The reality is garbage in, garbage out. Poor instrumentation of a product is actually a very common problem here. So what I mean by instrumentation is if you’re not actually capturing key user events in the product, it’s maybe simplest to understand this in e commerce.

Add to cart, check out, revisit site, etcetera. But for social products, it’s much more challenging. If you’re not actually logging the right events there, then you have no idea what’s happening inside your product and can’t start working on growth until you have that lens. You work with many different companies now.

to be so good at it? That’s a good question. Fair is a great business, and I think if I look at fair and think about what were the indicators early on when we were talking that showed me they’re really healthy in analytics, it was if you look at the dashboards and what they’re reporting out on, just very high signal noise. Rather than going and tracking everything and reporting on everything, it was here’s actually the three things that matter to our business, and this is what we’re gonna have rigor on reporting on every month.

I think they clearly had this set early on, and I think that propagated through their dashboards and everything else.

It’s a tricky problem, and I think it will evolve over time for a company. But I think it’s generally important across companies is actually having a growth model. Oftentimes this starts by getting into a spreadsheet, or actually pre spreadsheet, write down how your product actually grows and how it should grow. So for Pinterest, and there’s a concept called the growth loops that maybe we could talk about later, but new users coming in get brought in by existing content, go and post new content, which feeds the SEO engine and kind of drives more users there.

Maybe for a business like that at some point in time, it’s probably number of pins or active pinners at any given time. The point is having a just a conceptual understanding of what is the mechanic of your business that actually is driving the main metric that you care about. For most businesses, it ultimately ladders up to revenue. For some, it might be users early on. Just writing it down before you actually get into a spreadsheet is important. I think I yeah.

Yeah. And it’s That’s your growth rate, Harry. And there’s going to be this actually is a good one to dig into, because at a certain point in time, maybe early on, it’s getting guests that have over a certain threshold of audience is probably what your North Star metric for the first couple Later, there might actually be a guest retention metric to actually get their next audience on as their own reach is expanding. And then eventually converting listeners to guests might actually be like another kind of conversion metric that you care about later.

And so probably over your lifetime, there’s gonna be different North Star metrics. I think this is actually a good example to illustrate.

What should those stages look like? A lot of growth hires will ultimately fail, in part because a company doesn’t have a clear idea of what they’re hiring for. And so the fail case is, hey, we have a great product. Now let’s go hire someone to help us grow it. This might be true, but founders need to get a step more specific on what success or failure looks like for the role. What are the outcomes or systems they want to be in place? So defining success is step one.

I think step two is calibrating. Determine some set of companies with relevant growth patterns to yours. For you, it might be in other kind of creator, kind of like content oriented businesses. And go have calibration conversations with early growth leaders there. At Tilt, for instance, we knew that our growth playbook was gonna be community by community, so we went to talk to early leaders at different products that grew city by city or community by community, so Grubhub, Uber, Postmates, etcetera. The goal is really to build an intuitive sense of what great looks like, and sometimes those people end up being poachable.

That’s an important step as well. The third, writing a job spec and then the exercise. The spec oftentimes, like the best leads are not cold applying on your site, but writing the spec actually force clarity on what one is looking for. So if we have this growth spec now,

not that many people out there that understand this stuff that well. You’re either taking a bet on someone that hasn’t done growth before, and that’s all of us were there at some point, or you’re probably not running a very high volume process. I think of this as I would much rather be targeted, and I think in getting the spec out there, I skipped this, but I don’t think of it as, hey, we put on the website and then we see who comes in. It’s you’re sending it to investors and close friends in your network.

I think one of the biggest mistakes people make is doing references too late and treating it as a formality. Doing them early on I think is a great way to filter. Once you actually do get people in process, there’s a role for early filtering questions before broader exercise later. One of the questions I really like to ask people, pick your favorite product and tell me how it grows. That’s a really simple question, but it gets at whether someone understands the system and is able to think holistically versus specifying just one case study or one anecdote.

And it also covers both left and right brain thinking. Most systems should be grounded in both quantitative and user psychology. That’s like one good question to ask for that. You could actually even ask that before someone actually comes in the door just as a filter to see if someone is thinking the way you want.

What I’m about to say in some ways is a cliche interview question, but I think it’s important for growth, which is tell me about a time you failed and what happened from that. I think for growth, of the important things that you’re looking for is somebody who’s not afraid to fail. Failure is part of the experimentation process. The worst thing one could do with a growth role is go and try to cover up failures or just not acknowledge that it happened because what an opportunity to learn.

You were looking for someone’s risk appetite, but then kind of appetite to go and just put themselves in the stream of learnings that oftentimes come from failure and then go iterate on that. There’s a few that get a step more tactical, is show me your daily dashboard is one that I like. This maybe gets a little bit deeper, and this can be used in an exercise. It could also be used in a live conversation where you just ask someone to write it out on paper. But you’re looking for someone who’s analytical in nature with high signal in noise and is able to tie data back to a business context.

And if they’re too complex or showing too many metrics and they’re unwieldy with it, that could be red flag. Those are another couple ones I like. There’s a much longer list. I think specific anecdotes through references and actually through experiences they had in the seat. Ultimately, those are most important in getting as specific as possible as just general guidance.

One that I like is bring me a prioritized experiment roadmap for x feature. One of the important points of running a growth team is you need to have a system on triaging different experiment ideas and really having an ROI mindset. Some people are gonna come in and bring a laundry list of everything they wanna go test down to, like, button colors without having a system or some higher order context. I think that’s a fail case actually for kind of a new head of growth. You just come in and say, hey, I wanna have an impact.

I’m just gonna go and test everything and be the AB tester. And if you’re doing that without high order context of what’s actually gonna move the needle on a business, that’s problematic. And prioritization of an experiment roadmap is one way of getting at that.

Sure. I’ll go back to Facebook because a great growth team really. I know that one of the things they actually put the growth team on was when they spun out Messenger from the core app. It was the growth team that actually ran the product on that early on that drive a bunch of adoption to it. One question there could be, and actually this is separate from going and writing a whole growth model, although it’s probably helpful context to have one in place, but if you actually say, hey, your job is to come in and grow DAUs on Messenger from x to y, what’s your three month roadmap and how to go and do that?

I think writing the growth model is one. And we talked about this a little bit earlier, but write out a growth model for x product. It should be relevant to the business that you’re running. So if you’re running a marketplace company, have it be another marketplace business, just so people understand the set of levers and the interplay for that particular type of business.

them time for these exercises, because that takes some cognitive blush. Yeah. We expect them to do it in This is debatable. I always do. Call it a few days, and you could say, hey, don’t take more than a few hours on this. Yeah. I often do. If you’re testing for ability to produce this stuff on the fly, it’s another set of skills you’re looking for. Think there’s value to that, and maybe like the the dashboard question might be relevant on that. But I think for something like this, for these last two, I think you should give them time.

And for a model like this, really, you’re looking to make sure they can think holistically, that they could think in terms of loops and visualize growth into a spreadsheet. And maybe what might be a flag here is if they’re just bringing a pure kind of financial model or something that’s overly daunting or unwieldy, that’s probably problematic. You really want people in this seat who are simplifiers.

vary by few.

This is more of a general kind of comment, but I like being more generous with equity early on. Rabboy gave this awesome talk eight years ago at startup school on being a great operator. One point that always resonated was around barrels versus ammunition. I don’t know if you’ve heard this before. He basically states that all companies need high quality people in both ammunition roles and then barrel roles. But no matter how much ammunition you have, if you only have five barrels, you can only do five things simultaneously.

As you’re hiring people, just keep an eye out for who those barrels are. When you find one, do everything in your power to keep them engaged, throw more responsibility at them, more equity, whatever, etcetera, because that’s ultimately determining how much you can get done as a business and how fast you can move. In my experience, and obviously I have some bias here, but great growth people, especially those who join early, can be those barrels that a company really needs. I’ll go back to fair, like Olivia was the first PM there, and early on initially led growth, and then eventually like most roles in the company over some point in time.

A lot of companies have this individual. They’re oftentimes not as publicly talked about or whatever, but they were the first barrel in that company. And when you find that person, do whatever it takes to keep them around because they’ll probably hire other barrels too. Can I ask salary range?

Yeah. My I’m trying to knock out some trouble here too. But yeah, sure. Series B company, someone who’s probably mid career, because you actually don’t wanna hire execs too early on, but it’s also not someone who’s a few years out of school, probably you could go between 150 and $2.50 in salary, and then maybe a sliding scale between cash and kind of equity. And I think equity could go up to, I don’t know, for a Series B company, at the high end, these people could be getting north of a point, I think.

If you’re already series b and you’ve taken some dilution already, it might be shy of that. But you should be willing to par with half a point to a point, I think, for people like this. I would be hesitant if someone were asking for a whole lot more than the high end of that range for a company that’s not really proven yet. You said

What’s an example of of one? The definition for loops is like they’re closed systems where the inputs through some process generate more of an output that go then reinvested back into the input. And so there’s different types of loops. I referenced the content one from Pinterest early on. That’s actually a good example. A user onboarding and using the product, the output of that generates more content that goes and feeds back into the search algorithms that is the input into inquiring new users. Basically, usage begets more usage, new users begets more users.

That’s really the concept that you’re looking for. And so loops, they’re powerful because they lead to compounding growth systems versus typically funnels are more linear growth trajectories that will oftentimes decay with scale. And so the way to engineer loops oftentimes involves kind of product efforts versus just marketing. When we talk about growth, that’s a concept that’s fundamental to

I have that clearly. The traditional way that marketers think is funnels. So one framework, I think this might be Dave McClure’s, but it’s the pirate framework, RRR, RRR. So top of funnel really starts with awareness, then acquisition, activation, retention, revenue, and some might add referral to the end. The problem with this is this leads to siloed teams, and typically brand marketing might do the awareness stuff, then you have performance marketing doing acquisition, and then that might get handed off to some conversion PM that’s working on activation.

And so you’re really missing opportunities to have interplay between these steps, and you go and reinvest outputs from one step into the other. For some businesses, like ecommerce stores, it’s just the most practical way to look at a business. So it’s not funnels are bad all the time, but I think if you look at your growth this way, you will typically see a degradation of performance with scale. That’s the most common framework for funnels.

about the practice of growth, it’s about building distribution strategy into your product strategy, and then finding harmony between products, channels, and monetization. And so this is very different than saying, hey, we have a product team that’s going and building stuff, and then we have a growth team, whether it be marketing or sales or whatever, going and growing stuff. And I think the whole practice of growth is about creating more harmony and actually introducing distribution strategy into your product. And part of getting at this is org structure.

Right? So as I mentioned before, structuring org around a funnel creates silos, and teams will optimize at the expense of one another. And so I gave the Stitch Fix example of, hey, could send a bunch of shitty quality leads through, and then retention is hit. So when your company thinks in terms of loops, teams are forced to think about the interplay between different components. It’s less about road mapping tactics to reach some local optima on a specific metric, and more about seeking these compounding results. And most teams decide that cross functional teams are the best solution, and this is a core part of operationalizing because I think these cross disciplinary teams also come up with better ideas.

Other things that might fall under operationalizing is going and redefining new interim metrics and saying, hey, this is actually something that lives between acquisition and retention, but this maps more closely to the north star of our business, and this is what we’re gonna try to move.

How do we do that? The best ideas could come from anywhere within the org. So reducing friction to bring forth ideas is something that I think the head of growth should think of themselves as responsible for. And the second is the only real failure is not implementing learnings from an experiment. So you need to have a culture where it’s actually okay to take risks and fail in the spirit of learning. The process I’ve introduced in both the companies I led Growth for, there’s four main components of it.

So I think first is experiment one pagers. So anyone could submit an experiment idea, anyone, even if you’re not on the Growth team. But we built this very simple one pager framework that included an objective, hypothesis you’re testing, the design resources required to execute the experiment, timeline, so how long you need the experiment to run, and then what success quantifiably looks like, so you’re not actually going and having revisionist history and like, hey, it actually did work even though the data showed this. And then finally, the next steps that one takes if the experiment fails or succeeds.

So when you have that one pager, those ultimately fall into a triage list, which is most of time just a Google Sheet. So in one spreadsheet, anyone in the company could go and look at a prioritized list of experiments and see what’s running at any given time, what were the results of ones that ran in the past, and also the head of growth and the growth team should be triaging these. One framework that’s used to triage that’s common is the ICE framework. So impact, confidence, level of effort, you score every idea on these three attributes, and it’s a little bit rough.

But that kind of prevents the test everything mindset and the stuff that is the right balance between high impact, but maybe not high cost and a reasonable level of confidence that it might work, those will rise to the top.

These are actually the last two steps of the process I was gonna get out. So we would do weekly experiment reviews as a growth team. Every week, we had a meeting on completed experiments or anything that completed over the past week. The PM or whoever was responsible for it would go in and present on that. Lessons learned, implications, and suggestions for other functions too that were sitting around the table there. And then we would use that form to go and kick off new experiments. And so anything that was going to go live or that was in process, we would go and discuss those too.

So that’s within the growth team. I think company wide is equally, if not more important. And this is practical at certain company scales. I could tell, I don’t know, we were a 100 people. We did weekly all hands. At every all hands, the growth team would get up there and talk about these ideas. Sometimes engaging the company was important, so we would do sometimes like guess which variant won game. It was fun. It actually helped spark curiosity from different corners of the org on, hey, this intuition or kind of belief that I had on a certain part of the product was actually wrong, and the growth team went and invalidated that.

Communicating that out, I think live is best, and weekly growth team meetings, and then some presence at regular all hands for probably the most impactful learnings specifically I think is important. What makes good learning versus bad learning? It’s hard to answer with precision. Think in general, a good learning is specific enough to impact the way you move forward on the next iteration of a product or strategy. And so a bad learning might be one that’s not quantified or not specific enough to change your core product strategy.

A bad learning also might be one that’s just held within a single function and not propagated out to impact other teams. Paid marketing is easy. A learning that you get from spending millions of dollars on Facebook ads on what creative or copy resonates best, that should definitely inform your messaging on the product. If those learnings don’t make its way out, it’s a missed opportunity.

Yeah. This is precisely why, by the way, having a good analytics function in house is really important. Because, like, they’ll keep you honest on experiment design. Sometimes it’s gonna be hard to isolate all the variables, but if you’re running a test, you need to hold other variables constant. And sometimes the software, as great as your others might be, won’t capture this stuff.

What does that mean? One measure of product market fit health is what happens when you turn off all nonorganic acquisition activities. Ideally, you should see some ongoing user base that’s retained and ideally generating new users even if at a much slower clip. If you don’t have that, you don’t have product market fit. Going too heavily into paid marketing too early can lead to a number of problems.

Paid marketing is relatively simple and straightforward to go and acquire new users, and so it gets addicting when you’re more likely to design a single disciplined marketing function and miss the opportunity to do more challenging kind of cross functional initiatives, when you see the magic of being able to put money into a Facebook machine and deliver users back. The thing that many people miss early on is that paid degrades with scale. Yeah. Everyone says you’ll get better at it, so you’ll fight laws of gravity. That might happen in a short term, but on a long term time horizon, performance always degrades.

That’s just how it works. You could really get false signal on early CACs as a business when at low volume and end up with a business maybe a year or two down the line whose economics just straight up don’t work. And that could be very problematic. Given the speed of the feedback loops here, it’s addicting. But I think if you haven’t done the harder stuff up front, you could really just head down the path you don’t want.

It’s gonna vary a lot by business. If you’re a business of an e commerce store, you’re probably gonna be 30 to 50% like paid marketing. If you see business that’s north of 50 for any type of business, that’s probably like flag. This question itself is flawed, because I think this goes back to the point of incrementality. Attributing a single source of acquisition to a given user is usually a fallacy. And I’m not dodging the question here, but I just think it’s even this line of thinking actually part of the problem.

of channel performance to? The caveat on the answer on accelerant not a crutch thing is most businesses don’t find their one channel that works. And so if it needs to be paid, that might be okay. Just make sure you have enough room in your AOVs or customer value to support that. And the payback calculation is very important. And if you have enough buffer, it could be yeah.

wrong path as most VCs do? Completely agree with you. Seeming that you’re talking about early stage companies. Where I disagree is once you’re later stage I’ll take a Stitch Fix anecdote. This was around twenty sixteen, twenty seventeen. Cambridge Analytica do you remember the whole Cambridge Analytica thing? Basically like Facebook ads. There’s a lot of conspiracy on what exactly made Facebook ads non performing for everyone at that time, but there was a period of time when the shit just wasn’t working for people and performance took a big hit basically across the board.

We had a constraint that I saw with my team was we would not get more than 50% concentrated in any one channel, even if it were short term optimal to do. Because part of it, we were getting closer to IPO, but like you don’t want to be too exposed to any one channel. At scale, there’s a lot of diversification risk. We actually weathered through that just fine. Even though Facebook took a hit. We were nimble enough to be reallocating during that time. We had a broad channel mix where we were running a pretty diverse portfolio at that time.

But early on, I totally agree it could be problematic. And actually, there’s all sorts of other reasons why you don’t want to be too broad early on. It just makes measurement harder too. You’re looking at more interactions between channels that are harder to capture. At scale, you wanna be diversified. Early on, find your one. Ideally, that one channel is, like, not necessarily paid marketing. But I’m with you on that, Harry.

I think one of the lessons I had in doing this stuff at scale was I started buying Facebook ads in 2012. And at that time, there was a lot of alpha from just being a savvy media buyer, and people I used to learn from and talk with were these basically degen growth types that just continue to try to find arbitrage opportunities. And I’m talking specific to Facebook, but this kind of applies across most channels. That type of alpha gets arbitraged away over time, and ultimately what makes one great, or the only sustainable advantage on channels, is having great measurement, being sophisticated measurement, and then having great volume of creative, and great creative that performs.

And so ultimately, for the practice of performance marketing gets back down to do you have a unique kind of user hypothesis, and are you generating creative that resonates with them? Add enough scale. Some of the biggest, most impactful changes we made at Stitch Fix were lowering the bar to actually creative content we just put out there. So rather than having very high produced TV kind of quality video assets, let people shoot stuff on an iPhone, just get a lot more out there and feed it into the algo.

I think the biggest lesson I will go back to is incrementality testing as a real north star. So many marketers just rely on these overly simplistic last click attribution models, which give basically a 100% of your credit to one channel, which oftentimes looks like search just because it’s the most high intent channel you have in your mix, or like some black box model that you’re paying millions of dollars a year for to go and assign fractional credit and try to get you precision on optimizing your channel mix down to a T.

The reality is impact is rarely that direct, and that doesn’t mean that you shouldn’t be measuring stuff. And I think a lot of brand marketing gets, it’s kind of like, hey, this is marketing that’s not measured. I would argue actually brand marketing should be held to performance guidelines just on a longer time horizon. But incrementality testing involves running holdouts across channels that actually incur a cost. I’ll take a TV example, right? So this is actually an anecdote. TV was an important channel for us over time, and we ran it out of our performance marketing group, it wasn’t a brand marketing channel.

And typically when you’re running TV, the measurement method you will use is called spike analysis. So run a spot, and you get post logs back from whoever you’re buying TV from, down to the minute. If you look at, call it a seven minute window around the time of spot errors, the lift in traffic above what their baseline traffic should be at that time gets attributed back to TV, and then you could apply conversion assumptions, and that’s who comes in. That’s gonna make your TV look very expensive, because that’s ignoring any halo effect or kind of lift on other channels.

And so for us, when we wanted to decide, hey, how much do we want to scale up TV? We ran, and we were public when we did this, we were getting a bunch of analysts covering us, hey, Stitch Fix on OpTV, are there problems? But we ran an eight week holdout test where we actually paused national TV advertising, and we would always buy nationally because it was a lot cheaper from a CPM standpoint to do. We paused national for eight weeks.

We picked five regions, and we went and go bought local TV spots, And we measured the overall lift in those markets and the interaction between other channels that we were running and came up with awesome insights on actually what was the interplay from TV on our Facebook ads and what was actually the true incremental CPA. And when you understand incrementality, you could actually go back into whatever attribution model you’re using, whether it be last click or whatever, and go and apply incrementality multipliers to them. And then you actually are getting the sense of precision that a lot of growth marketers are looking for.

But without incrementality testing, you could be drawing very wrong conclusions on how your channels are working.

Amarage? Most teams should actually have some return on ad spend, either ROAS or kind of payback threshold. And you could be tight with payback. If you’re a subscription business, you’re not paying back within a year, that’s probably problematic. Ideally, if you’re ecommerce, you’re paying it back on first transaction because the year is making all these conversion and repeat assumptions that you might not have that point. But I do think ultimately payback period is probably the guiding metric that you want to make that decision. At Stitch Fix, from a contribution margin payback standpoint, a year was our final threshold.

We had enough data on what was repeat where we were able to be relatively tight on that. The how much kind of ultimately, I think, should boil down to whatever payback makes sense for your business. Typically, I’d say they’re gonna be first transaction or some period of time.

So there’s two primary flaws with it. I think one is just this notion of lifetime, which is the L in LTV. Early in a company’s journey, it’s rare to have a real idea of what a reasonable customer lifetime is. If you’re two years in, how are you gonna say you have a five year lifetime? That’s obviously flawed. And then it’s also too coarse. I think many companies will look at this on a blended basis and disregard that there’s actually a bunch of variance in customer quality by channel, keyword, audience, etcetera.

The better method, which I was getting at, was Payback Period, which removes the notion of lifetime, but also it could be calculated on a more granular level. And Payback’s the however month of growth margin, it costs to pay back the initial incremental CPA. I think you want to make sure your calculating this paid over paid, so use paid CAC, not blended CAC, and then with scale you could get to more granularity, so not only looking at it on a channel level, but actually a keyword level.

With Stitch Fix, you should have different basically CAC thresholds for someone who’s actually searching for high priced dresses versus discount shoes. And that’s the important thing. One caveat on all this stuff, Bill Gurley was on our board at Stitch Fix, I remember he once wrote an article people should read called The Dangerous Seduction of the LTV Formula. I remember I first learned about this the night before presenting in my first board meeting there, where I was presenting our new paid acquisition strategy as a guy that came in to, like, scale up our ad budget, and I was just terrified.

But he’s right. One of the important points he made there is this is a tool and not a strategy. Too many performance marketers use this formula as a way to justify more budget. So many people out there are competing on the size of budgets that they manage, and there’s these weird kind of power dynamics and stuff. I think at a certain point, having a deeper understanding of user psychology and being great at creative provides more of an edge than some of this mechanical LTV CAC optimization.

I just think this notion of lifetime is the main one, but also the granularity. You’re able to get at that better if you have a payback kind of framework.

you no longer believe? That growth solves all problems, which is something we used to say back then. But I think growth might mask certain problems, but there’s always second order effects and subsequent hills to climb. It’s important to be deliberate and strive for control versus having a growth at all cost mindset. What’s the one word that will be on your tombstone? It’s two words, lifelong learner. I think much of what I say about growth models and my own desire for exploration and discovery, I think this extends in what I hope to model for my two kids, having a growth mindset, desire to learn, evolve, reinvent themselves throughout their lives.

I like to think that I’m very receptive to feedback, have a genuine desire to learn and grow because of it. And this extends deep into different pockets of my personal life too, not just the growth stuff that we’re talking about here. I’m always trying to find better systems to learn, fail, iterate, etcetera. So, yeah, think lifelong learner would be it. Andy Weissman submitted.

surrender Well, today Andy found a way to squeeze in a reference from the best BAM in the world. Outside of reading the Helping Friendly book, I would just say conversations with smart, curious people from a wildly diverse set of backgrounds leads to eternal joy and never ending splendor. I’ll leave it at that.

you say is the biggest mistake founders make when hiring growth teams? I’ll give two. Not referencing deeply enough to contextualize what growth means in that hire’s company. Hiring a head of marketing to go run a product function might not work out. The second is probably assuming that growth is just the growth team’s job and setting your org up that way. The person’s just destined to fail.

The meta lesson at Tilt was about focus. But if I go down to growth, I would say the main lesson for me personally was around over incentivizing usage. We knew that we had a product that grew contributor acquisition through organizer retention, so that was our kind of growth model. We often incentivized organizers to come back. Once someone gets hooked on a promo, even if it’s small, it’s very hard to go back to non incentivized behavior. I took this lesson into Stitch Fix. We never discounted any product there, which is rare for e comm actually, and was a core part of the company’s discipline.

So it would come down to over incentivizing kind of usage. What would you most like to change about the world of growth? People assuming that it’s a set of micro optimizations versus a holistic system, can include big bets.

I hope we see more, and I also hope we see more heads of product and heads of marketing that actually just understand these principles and roll that into their functions. I think this discipline is making its way out there. Don’t I know whether that shows up in titles, but I do think it will continue to accelerate. You are great at x, name x. Why do other people fail at x? Taking feedback, I’ve conditioned my mind to absorb it all as inputs to my own personal growth model that I could use to iterate and improve myself.

Others might fail at this because they’re afraid to seek and hear the truth. They might be closed off, defensive, overreact to it. I don’t know. And the truth is oftentimes uncomfortable, so people might rather seek comfort. That’s probably what I would say.

I’ll reference Fare another time. I think their b to b referral mechanic was insanely powerful. And early on, like, that was the only thing that was working for growth and that continued to carry them this far. I think referral systems and referral engines, typically, there’s always more juice from a referral program to squeeze than companies realize, but they typically are thought of in a consumer context. I think Fare’s b to b marketplace pulling in cross side referrals, that was huge. I totally admired the way that they built that and scaled it out over time.

That’s probably the one that I go to.