Go Slow to Go Fast: Software Building and Investing
What this covers
This is a conversation between host Patrick and early-stage software investor Chetan. It moves between how enterprise software companies get built and how an investor should judge them. The examples are Salesforce, Workday, Veeva, MuleSoft and MongoDB. Chetan keeps returning to the same questions: who the first customers are, how long a product takes to mature, and how much timing matters. Databases classically take about ten years to mature, and MongoDB's first code commit in 2007 became production-ready around 2015 or 2016.
The investing half starts from an Asymmetric Payoff: a loss is capped at one times capital, while a missed winner can cost 100x or more. Chetan says an investor who never misses isn't taking enough shots. He argues for seeing the present clearly rather than forecasting. His MongoDB example is a 2012 bet that developers who didn't want to plan a schema upfront would drive adoption, which the best developers, judging by MySQL-style checklists, missed. Two claims are marked contested: that cheap mistakes make that database appealing, and that a similar company started in 1992 would have failed.
The building half covers five to ten lighthouse customers and the point where features stop generalizing. It also covers the argument for professional services early on: Workday and Veeva drew nearly half their early revenue from services. Salesforce's $12,000 average contract in 2004, against Siebel's $400,000, explains why the two rarely competed. Chetan also covers capital efficiency, using MuleSoft's growth on small spend. Patrick contributes a view on being default open, and Chetan makes the case that software is still early.
Chetan Puttagunta argues that durable software businesses are built by going slow to go fast: serving a few lighthouse customers, staying cash-efficient, and relying on product quality and trusted relationships rather than early scaling, formula-driven growth, or competition-driven selling.
- Early-stage investors should see the present clearly rather than predict the future, and back non-consensus insights (MongoDB's developer-first bet in 2012)
- Generalizable products built for a few lighthouse accounts, plus professional services early on, create the trusted vendor relationships that later convert into subscription revenue
- Mature products drive customer acquisition cost toward zero, while premature scaling and spreadsheet-driven growth break businesses
Enterprise software and its demand are early, under-penetrated, and broadening beyond knowledge workers.
- Oracle built a large database business and then moved up into the application tier, serving CRM, HR, and ERP-style applications such as PeopleSoft on top of its database technology.
“Oracle did was built a great database business and then moved into the application tier”
- Three indicators show enterprise software is early: every large enterprise now expects software to power its business, developers are a tiny fraction of knowledge workers, and roughly a trillion dollars of annual global IT spend is still largely on-premises rather than cloud.
“a lot of it is still contained in the on-primed world. So, the move to the cloud, we're still on the very, very early innings of that cloud”
- Software markets have turned out to be orders of magnitude larger than most people anticipated; Salesforce, which had $13 billion in revenue, still claims less than 20 percent penetration of its core markets, a dream about software that was still not big enough.
“Salesforce itself has $13 billion of revenue, and Salesforce would argue that they're less than 20% penetrated into their core markets.”
- Oracle, founded in 1977 and a database leader from the 1980s, sustained dominance for decades, which is unprecedented in technology for a single company behind one core technology concept; SQL's pervasive use by developers and analysts propelled it.
“That's fairly unprecedented in technology that you had one company behind one technology concept that had that long of a run of dominance.”
- Once a general-purpose platform like Salesforce exists, vertical applications can be built on it to create multi-billion-dollar companies, as Veeva did with a healthcare-specific CRM.
“Veeva built a CRM system for the healthcare vertical and created a multi-billion-dollar public company out of it, on top of Salesforce.”
- Early-stage companies must say no to good ideas, such as a $200,000 custom widget request, and this is easier when the company has not set an artificial revenue goal that pressures it to accept off-strategy deals.
“It's much easier to say no to $200,000 to build a custom widget if you haven't set an artificial revenue goal for 2020.”
- A general-purpose software product for a use case increases the efficiency of every customer that uses it, which is better than each company building custom point-to-point systems, as Salesforce's cloud CRM beat every firm building its own CRM.
“Salesforce building a cloud-based CRM system was way better than every single company building a CRM system on their own”
- Every large enterprise across industries now expects its future to be powered by software, so non-traditional IT buyers such as Caterpillar, pharma, financial services, and shipping companies are increasing the number of software buyers.
“Companies that you may not have thought of as being big consumers of IT are now big consumers of IT.”
- Because every business wants to become bigger and more profitable, enterprise software that increases a customer's revenue or cash flow will always find adopters open to new ideas.
“if you have a solution that will either increase revenue or increase cash flow, whatever it is, you will find adopters”
- MongoDB did not generate meaningful revenue until around 2011 and 2012, several years after its first code commit in 2007, which shows how long product maturity takes before revenue.
“they didn't really have real revenue until like 2011 and 2012.”
- The number of software developers in the world is still a tiny fraction of the number of knowledge workers, implying software is deeply under-penetrated relative to the workforce.
“the number of developers in the world is still a tiny fraction of the number of knowledge workers in the world”
Lighthouse customers and early adopters drive early growth and compounding expansion revenue.
- At the start, a software company should focus on a small number of relevant lighthouse or focus customers, perhaps five to ten, and build product to meet their needs only up to the point where features generalize to the whole focus market, avoiding custom work that becomes professional services.
“that could be as few as five, or as many as 10, and you're addressing the specific needs of those customers”
- In 2004, Salesforce's average contract was about $12,000 per year with around 15 licenses, while Siebel's average contract was about $400,000 with around 1,000 licenses, so the two rarely competed for the same deals despite both being in CRM.
“Salesforce's average customer contract was $12,000 per year and they were on average selling 15 licenses at the time. Siebel at the same time had an average contract value of $400,000”
- Customers have a job to do every day and will abandon a tool that does not make them faster; product builders should therefore design for the job the customer needs done rather than expecting users to master the system.
“their job is to accomplish that, not to become an expert in your system.”
- Offering professional services to early customers, even though services revenue is often dismissed as low quality, creates a trusted vendor relationship and training path that can later convert into subscription software revenue, and services can produce reusable migration tools.
“a services contract is actually a commercial engagement with that customer that then allows you to build a trusted vendor relationship”
- In their early years, both Workday and Veeva drew nearly 50 percent of revenue from services because they were displacing entrenched on-premise systems like PeopleSoft and home-grown CRMs, and the migration tools built in those engagements sped up later customers.
“in the early days of both companies, nearly 50% of their revenue was from services”
- Salesforce's average revenue per customer compounded about 15 percent per year over fifteen years, and combined with new customer acquisition, this helped revenue grow from about $100 million to $13 billion, illustrating how a 15 percent annual expansion rate can compound into a large business.
“the average revenue per customer has compounded at a 15% annual rate”
- Customers rarely run bake-offs between two solutions until contracts reach roughly $1 million or $10 million, and at $10,000 to $20,000 contracts, customers evaluate a startup against whatever custom solution they already use.
“until later stages when you're going into bake-off for a $1-million contract or a $10-million contract, it's very rare that a customer is baking-off two solutions.”
- Entrepreneurs with a novel insight, often from fresh eyes or from deep incumbent experience, find early adopters who agree almost immediately when asked why a market is done a certain way, and those adopters become proof points for an industry standard.
“it's almost like you ask the question, 'Why is this being done this way, shouldn't it be done this other way?'”
- Sarah Tavel's consumer software engagement framework applies directly to enterprise software when used through a lighthouse-account lens, because deep engagement, and the kind of engagement customers exhibit, matters across software categories.
“if you read that blog post from an enterprise software angle, you actually find that everything she's talking about is very applicable to enterprise software.”
Cash-efficient, long-horizon growth compounds into durable software value over time.
- Cash efficiency is what ultimately matters as a software business scales because it yields a logical return on invested capital against R&D and sales and marketing, the assets that compound over long periods.
“cash efficiency is ultimately what matters as you scale a software business because that then gives you a logical equation on return on invested capital”
- Sustaining 15 percent annual expansion for ten years requires solving fundamental business problems for customers rather than point solutions, which makes the vendor a long-term business partner, and this requires patient capital and long-term thinking.
“so I'm not solving point solutions for them, I'm solving fundamental business problems so that I'm actually their business partner for the next 10 years.”
- MuleSoft grew from about $100 million to just over $200 million in revenue on roughly $8 million of spend, then from $200 million to $300 million on about $4 million, and Salesforce acquired it for $6.5 billion, while Salesforce's own growth accelerated because of a capital-efficient product-led go-to-market system.
“They've earned something like $8 million to go from $100 million to a little over $200 million of revenue, and then burn $4 million to go from $200 million”
Cloud and enabling infrastructure still growing fast make now a strong time to start a company.
- Microsoft Azure grew 59 percent year over year in the latest quarter off a huge base, which suggests the infrastructure layer of cloud is still growing very fast and the total addressable market remains large.
“we just saw the Azure earnings release from the last quarter and it was 59% year-over-year growth.”
- Building a general-purpose database usable for any application is extremely hard because security, auditing, and surrounding infrastructure must wrap the database, so specialized databases have historically been easier to build than general ones.
“building this general used-case, applicable to anything you want to build is a really, really hard problem.”
- Containerization extends virtualization from whole servers down to individual workloads or applications, so containerized applications can move easily across clouds and data centers, which is why Kubernetes is an enabling technology.
“if you have containers that contain a data or an application, or part of an application, moving those around across servers and across clouds actually becomes really easy.”
- Infrastructure and enabling technologies are moving so fast that it is a great time to start a company, because more businesses see software as core and seek vendors that help them do their jobs better.
“it's a really great time as an entrepreneur to start something”
- AWS, which became public around 2006 and matured through 2007 to 2009, transformed development by replacing rented colocation cages and personal servers with on-demand infrastructure, which massively accelerated innovation in the technology economy.
“what Amazon enabled with AWS at the earliest stages of application development was truly remarkable”
- A Wall Street Journal article reported Microsoft at 17 percent cloud market share, which Chetan cited as evidence that Microsoft has become a serious competitor in cloud infrastructure.
“the latest Wall Street Journal article on this has said that they're at 17% market share.”
Developer-first databases win through agility and market timing that incumbents miss.
- A database that does not require upfront schema planning lets developers make cheap mistakes and learn cheap lessons, because a failed consumer app wastes little time when no schema has to be designed first.
“that allowed you to have very cheap mistakes and learn really cheap lessons”
- Databases classically take about ten years to become mature; MongoDB's first code commit was in 2007, and it became mature enough for production use with millions of users and transactional data by 2015 or 2016.
“classically, everybody says that it takes about 10 years to develop a really mature database”
- In 2012, betting that developer agility and speed of iteration would be a primary driver of a developer-focused database was non-consensus, because the best developers judged MongoDB by MySQL-style feature checklists and missed the class of developers who did not want to plan schemas upfront.
“But you missed the entire need of the entire class of developers that wanted something where they didn't have to think about schema”
- MongoDB succeeded because web and mobile changed how users and applications interact with data, and unstructured, unpredictable data volumes grew sharply; market timing matters so much that a similar company started in 1992 would likely have failed.
“if you invested in something that was started in 1992, it wouldn't have worked.”
- Because of the cloud and developers' openness to new database technologies, databases have shifted toward specialized open-source options built for specific use cases, such as globally scalable relational databases (CockroachDB) and time-series data for IoT (TimescaleDB).
“because of the cloud and developers being so open to understanding new database technologies, it's now become very much about, okay, I'm building this specific used case”
Software must mature as a product before growth metrics, or it slides into spreadsheet failure.
- Software businesses work best when products are mature enough that customers keep renewing, lifetime value is effectively unbounded, and customer acquisition cost approaches zero, which requires a long time to build the product.
“if you're building great software, LTVs should be infinite. And so software businesses end up working really well when you build mature products”
- Many founders start to build technology but gradually shift into spreadsheet-driven businesses optimized for formulaic growth metrics, which is an unhealthy failure mode that makes it hard to re-engineer the company later.
“they go from a what I call technology-driven business to a spreadsheet-driven business.”
- Companies that justify hiring fifty sales reps using CAC, LTV, and payback-period formulas often find those assumptions falling apart within three years, leaving a high burden that is hard to re-engineer out of.
“let me hire 50 sales reps and let's push this into the market, and three years down the line all the assumptions that went into that formula of growth just start falling apart”
- Early-stage software companies usually do not face direct competitors because the new system is unlikely to be built by anyone else; the real obstacles are objections to an incomplete product, and founders should avoid chasing incumbents until the product is mature.
“if you're building a brand-new software system, it's unlikely that someone else is building what you're building.”
- Bill Gurley's writing on the traps of LTV and CAC is widely applicable to all technology businesses, because every consumer and enterprise company can justify inorganic growth with a formulaic LTV-to-CAC ratio and then run the business by spreadsheet rather than by actual customer needs.
“every type of consumer companies have this trap where you justify CAC or inorganic growth based on some formulaic LTV”
Openness and open ecosystems strengthen developer communities, companies, and cloud competition.
- The developer ecosystem around open source is much bigger than most people assume, and developers constantly seek the best solutions, so founders should join developer communities, share their work, and publish talks, as Netflix has done, to get feedback and recruit talent.
“the developer ecosystem is a lot bigger than what I think most people assume.”
- As the number two and three cloud providers, Microsoft and Google have embraced open ecosystems and partnered heavily with startups, offering a catalog of best-in-class open-source options with their backing, which is a smart strategy to win share from Amazon.
“because they're number two and number three, they have really embraced this idea of an open ecosystem”
- Proprietary processes are no longer reliably defensible because innovation is fast; durable advantage comes from the customer relationship and being a trusted vendor, and openness such as publishing papers, as Google has done, strengthens rather than weakens an organization.
“What ends up being defensible over time is actually what we talked about earlier which is the customer relationship and being the trusted vendor”
- Google Cloud gained popularity for machine learning and analytics workloads partly because Google open-sourced TensorFlow, and its aggressively low prices for services such as BigQuery and Bigtable made it attractive.
“part of that has been driven by them open-sourcing TensorFlow”
- Having three strong cloud infrastructure providers, Amazon, Microsoft, and Google, is good for the innovation economy because it prevents one company from owning compute and storage for the whole technology economy.
“I think having three really strong infrastructure providers is amazing.”
Scaling depends on a structured, process-driven hiring funnel rather than luck.
- At scale, a software business is fundamentally a people business, and every failed hire or need to re-recruit is a failure of the company's process, not bad luck; the only way to scale is with great people.
“It's actually a people-driven business ultimately.”
- Startups can gain an advantage in recruiting by treating talent as global and distributed across the United States, instituting from day one a remote-friendly culture in which everyone joins meetings from their own computer and decisions are documented in wikis, which is more about culture than tools.
“we only do meetings where everybody is on their computer.”
- Recruiting works like sales: a company must build a large funnel of candidates, run a structured predictable process so candidates feel good, and reach a double opt-in hiring decision, with references checked for past performance.
“there's a lot of parallels between recruiting and sales, frankly, you have to build a really big funnel”
- Elastic's founders hired complementary executives across product, engineering, and sales who scaled the company from single-digit millions to hundreds of millions in revenue by continuing to learn and by being given resources and freedom.
“They scaled from single-digits revenue to hundreds of millions of dollars in revenue”
- Giving talented people resources, freedom, and trust creates a flywheel in which they bring on even more talented people, which in turn builds trust across the organization.
“there's trust in the organization, they then bring on even more talented people”
- Hiring is a double opt-in process requiring a large candidate funnel and structured process; each time a company must re-recruit to replace someone, or makes a bad hire, the fault lies with the company's process rather than the candidate.
“every time you have to go back out to market, to get somebody or to replace somebody is actually a failure of that process”
Learning comes from networks, open communication, and asking experts good questions.
- Companies that write detailed wikis about their thinking and how they reached decisions become remote-friendly, because asynchronous participants can catch up and add to the reasoning.
“you write a lot of content about your thought process around how you arrived at a decision”
- Learning in public on Twitter and across networks can produce informal mentors who sharpen an investor's ideas, and being open-minded to many people is a key source of professional growth.
“learning in public and finding brand-new connections through Twitter that then become sort of informal mentors”
- Benchmark has five equal general partners with no junior partners or associates, so its Monday partner meetings can run open-ended discussions without an agenda cutoff, and each partner's network can be tapped to answer questions quickly.
“we don't have anybody else. We don't have junior partners and associates”
- Learning efficiently requires asking subject-matter experts the right questions and synthesizing complex information into digestible ideas, and a partnership with a deep collective network makes it likely any question can be answered quickly.
“being able to talk to other people that are subject-matter experts and be able to asked them the right questions to learn from them very quickly”
Early-stage investing is a double opt-in relationship built on networks and focused partners.
- Early-stage investing is a double opt-in relationship in which both the company and the investing partner must resonate about the long journey ahead, unlike public markets where an investor can buy a stake without any relationship.
“it's a double opt-in relationship, when you are embarking on this really long journey of company building.”
- When a partner group has a question about an investment, members can call the three most relevant people in their networks to get an answer quickly, so the group's collective network effectively answers almost any question.
“there's probably not a question that we can't answer, and if we can't answer it, we know how to find the traps”
- Benchmark invests in five to ten new companies a year, with each partner doing one to two deals and taking a board seat each time, and it avoids a thematic wish list because the best insights come from entrepreneurs and require an open mind.
“if you tell me what are the 10 start-ups that you want to invest in this year, or next year, or the next 10 years... It's zero”
Investors should watch present ground truth to avoid missing winners.
- Duffel's founders from the payments industry saw that airline global distribution systems were convoluted and, using APIs and microservices, proposed a simpler way to sell tickets that airlines themselves quickly endorsed.
“they looked at the travel problem and said, 'Why is it being done this way? Couldn't we solve this?'”
- Missing a winner costs an early-stage investor far more than a bad investment, because a loss is capped at one times capital while passing on a winner can forgo 10x, 100x, or even 1000x outcomes, so an investor who does not miss is not taking enough shots.
“if you make a mistake you lose one times your capital”
- An early-stage investor's job is to see the present very clearly, watching ground truth in the world, rather than predicting 10 or 15 years ahead, because forecasting the future tends to produce bad timing.
“our job as early-stage investors is to see the present very, very clearly and not predict the future”
Publishing technical problems and seeding culture early, plus offline marketing, supports operations.
- Publishing the technical problems a company is solving attracts people who want interesting work, making open engagement an effective recruiting channel.
“people want to work on really interesting problems, and if they see the interesting problem that the companies working on, they'll come and engage”
- Company culture is a set of shared stories transmitted to new employees, and when headcount doubles yearly, half of employees have under one year of tenure, so culture must be seeded early to propagate and compound.
“a company culture is just a set of shared stories that then get transmitted on”
- Offline inventory such as train stations and airports is surprisingly efficient for enterprise marketing because it targets people in a specific city during idle time, which is why enterprise software ads appear in airports.
“offline ad inventory is actually really efficient.”