One way to describe Salesforce’s flagship product—their customer relationship management tool, in which salespeople keep track of who they’re selling to and how those deals are going—is that it is a bunch of lists. People make this point sometimes: Salesforce is just a database with a point-and-click user interface on top. Businesses buy Salesforce to help their sales teams sell stuff, and when a business wants to sell stuff, they need to maintain a bunch of interconnected lists: Of companies they’re selling to, of their clients’ email addresses and phone numbers, of the products they sell, of the messages they've sent their prospects and the meetings they’ve had with them, and of various other things that you might imagine salespeople wanting to write down. The database part of Salesforce maintains those lists, and the interface part of Salesforce lets you look at and update those lists. But you could, as people sometimes suggest, also do all of this in a big spreadsheet.
This isn’t exactly wrong—Salesforce does have a lot of lists, and most of the time, people log into Salesforce so that they can read or change the lists. Still, reliably keeping track of lists can be very hard. Keeping big, dynamic lists in Google Sheets would be complicated and brittle, and Salesforce probably does a better job of it than Jeff. So people pay Salesforce $37.9 billion a year, for a bunch of lists.
But another way to describe Salesforce is as a playbook. Selling stuff to people—especially big stuff; especially if you’re selling it to large companies; especially if many people are involved in closing a deal—is a complex operation that requires a lot of coordination and collaboration. Salespeople have to identify potential buyers; they have to find ways to get introduced to the those buyers; they have to have pitch meetings, and prep meetings, and debrief meetings, and meetings at fancy steakhouses. They have to make materials for those meetings; they have to follow up, circle back, touch base, check in. There are motions to all of this, and various sales experts have developed recommended frameworks—BANT, CHAMP, FAINT, NEAT, SPICED, SPIN, YHTMAAAIYP—for how to manage it.
And Salesforce, via the product’s vocabulary, features, and defaults, is a loose encoding of those recommended processes. To close a deal in Salesforce, you have to create an “opportunity,” which is associated with an “account,” and pass that opportunity through a series of default “stages.” Certain important fields, like the amount that a deal is worth, have to be filled out before you can add it to Salesforce’s list of contracts. There are default forecasting categories that are considered industry standards. So, sure, Salesforce helps people sell stuff by keeping lists, but it also helps people sell stuff by keeping opinionated lists. The lists have specific names; the lists are governed by pedantic rules. The lists help you sell stuff by existing, but they also help you sell stuff by telling you, roughly speaking, how to sell stuff.
Lots of software follows the same pattern.1 It helps you accomplish something—sell stuff, manage tasks, remember your thoughts, date people—through both a mechanical utility—it maintains lists, logs to-dos, stores your notes, shows you pictures of people you might like—and some sort of embedded expertise. The builders of Salesforce were (or talked to) expert salespeople; they decided that the best sales teams follow a particular process when they’re out in the field; they then built lists in Salesforce that guide people towards that process. Linear, a task management application, did the same with product teams and engineers; now, Linear not only keeps track of your tickets, but it also tells you how you should prioritize them. Roam, a note-taking product, believes it’s more useful to imagine your notes as a network of ideas rather than a filing cabinet of documents, so Roam encourages its users to interconnect their notes through backlinks. Hinge, a dating app,2 wants to prevent people from mindlessly swiping, so they require users to like specific parts of someone else's profile.
But this is all somewhat rough and indirect. First, Salesforce isn't made for your sales team; it's made for a sales team, for the median sales team. Your sales team may not want to use MEDDIC, a popular sales discovery framework, to evaluate deals; they might want to use MEDPICC. But MEDDIC is more common, so Salesforce’s defaults will prefer that.3 Second, Salesforce has to paraphrase their expertise through product features. Advice like “make sure to define a compelling event when you’re trying to close a deal” could get reflected in Salesforce through rules that require salespeople to define a target close date when they create a new selling opportunity. But people might work around that by always choosing a close date that’s three months in the future, just so that they have something to put in form. Nudges can encourage good behavior, but nudges can be ignored. And finally, Salesforce’s rules are blunt instruments. From Matt Slotnick:
To move a prospect between sales, a lot happens. But the way it’s reflected in application is quite simple, necessarily. The seller will do a ton of work (research, meeting, emails, powerpoints, etc) that ultimately gets recorded in the system in a fairly simple way… from Stage 1 to Stage 2… Prospecting to Qualification… with a few mandatory fields to fill out.
But the texture, the granularity of what happened, is largely lost because there’s no way for the system as designed to comprehend or make use of it.
So. If you wanted to sell stuff, what would be better than buying Salesforce? One answer might be to buy a very customized version of Salesforce that is designed to fit exactly what you need. And for a long time, people have tried to mimic this, by hiring Salesforce consultants to tailor their versions of Salesforce to fit the way they want to sell.4
A second answer could be to build an entirely new version of Salesforce, just for you. Don’t buy a Salesforce; vibe code your Salesforce:
LLMs…will drive the cost of creating software to zero. What happens when software no longer has to make money? We will experience a Cambrian explosion of software, the same way we did with content.
Vogue wasn’t replaced by another fashion media company, it was replaced by 10,000 influencers. Salesforce will not be replaced by another monolithic CRM. It will be replaced by a constellation of things that dynamically serve the same intent and pain points.
But there’s third—and maybe even better?—way to replace Salesforce: Don’t use it all. What if, rather than buying Salesforce’s product, you just hired Salesforce's sales experts and had them manage all your lists directly? Instead of buying a very approximate version of their expertise, delivered through oblique suggestions in a piece of software, could you just buy the experts themselves?
If you have the experts, there’s no need for opinionated lists. They sit between you and the lists, and they can organize their lists however they want, or enforce whatever pedantic rules they like. You don’t need to know about any of it. You just tell them what’s going on, and they figure out how to keep track of it. If you want to add a new prospect to your lists, you tell them to do it, and they record the details in the proper places. And if they don’t like how you’re doing something—if you try to create a sales opportunity without identifying a compelling event, and that troubles them—they can just tell you that, and what you should do to fix it.
Compared to Salesforce, the experts with lists are more directly helpful, because they don’t have to translate their recommendations into product features.5 They’re more precise, because they can tell you, per Slotnick’s concern, when things are halfway between stages or when something doesn’t fit into an exact taxonomy. And they’re more flexible, because they can adjust their advice to your business and circumstance. Maybe this deal doesn’t need an amount attached to it, because it’s for a partner account. Maybe this deal should be forecasted as “Commit,” despite missing some of the usual qualification criteria, because they know the sales rep has a close relationship with the buyer. Maybe this deal shouldn’t be recorded at all, because it's a bribe. Salesforce doesn't allow for these exceptions. People with spreadsheets and good judgement do.
Of course, this is all somewhat impractical. You can’t hire Salesforce’s employees directly.6 And even if you could, a couple of people, no matter how good they were at sales stuff, couldn’t keep track of everything a sales team does. But if you could hire an infinite number of experts, each of whom had an unwavering attention span, an unrelenting attention to detail, and the ability to read really fast…
You can see where this is going.
Rather than building opinionated lists to help people sell stuff, you could imagine Salesforce building a completely different product to do the same job: AI bots explicitly instructed on how to run a good sales process, and a bunch of spreadsheets. Every morning, the bots check their spreadsheets. They tell you who they think you should call. Before each meeting, they tell you where the deal currently stands, and share some facts about it from their lists. After each call, you tell them how it went and what they should change. They meticulously update the lists, in accordance with whatever best practices they've been told to follow, while also making reasoned judgements about where to allow for exceptions.
Rather than being a database and a UI that’s inspired by a suggested operational workflow, Salesforce would instead be a database and an explicit description of that workflow. It is the expertise of Salesforce’s team, written down as words. The product is the prompt, and the prompt is the last sales guide you’ll ever need.
Which, admittedly, might sound uncomfortable. If it’s not in Salesforce, it didn’t happen! We need to see the lists! We need to check the lists, to touch the lists, and know they’re real! We can’t put robots between us and our lists!
But can we? Our reliance on seeing lists in Salesforce (or in Linear, or Roam, or any other app) feels more like a security blanket than a genuine need. We compulsively look at the lists because we haven’t had any other way to run a sales team, so we begin think that compulsively checking the list and running a sales team are the same thing.
But consider, for example, email. If you don’t care that much about managing your emails, you use Gmail, do nothing, and end up with 606,646 unread messages in your inbox. If you are moderately important,7 you buy Superhuman and it does various things to cajole you into organizing all of your emails.8 But if you are actually important, you hire an executive assistant, and they manage your emails for you. And you don’t care if they use Superhuman, or Gmail, or run an email server on a Gameboy. You don’t care how they tag and triage emails, or if they keep your inbox at 0 or at 606,646. You hire them because they are experts at managing and filtering lists of emails, and you let them do that however they want. And if you trust your EA, you stop checking the list.9
Could the same trust not extend into other workflows, and around other things we do? If the best way to manage our emails is not a fancy email client, but an email expert with access to our inbox, could the best way to manage other parts of our lives be an “expert” with a spreadsheet?
Anyway, there’s a new dating app:
Today’s dating apps bank on the speed of onboarding and having millions of options. Users create profiles within seconds by uploading photos and answering simple questions. The apps then rely on basic info and feedback from users’ swipes to find them potential matches.
Sitch aims to take a more thoughtful approach with its onboarding process and uses large language models (LLMs) to bring a human matchmaker’s expertise to the dating app experience, helping people find potential matches without swiping.
The startup was co-founded by Nandini Mullaji, whose knack for the dating market comes from her grandmother, also a matchmaker.
…
Essentially, Sitch built an AI version of Mullaji that helps users onboard by asking them details using almost 50 questions, which they can answer through text or voice.
After the dater’s profile is set up, the AI matchmaker displays its suggested matches. If both users agree to match with each other, the bot adds them to a group chat with the AI. At any point in time — even after their real-life dates — users can provide feedback about their matches to improve the AI’s personalization.
Right, exactly—most dating platforms are a database of people and their interests, and an app that tries to encourage similar people to talk to each other through things like suggested matches and how profiles are presented. Sitch is an expert with a spreadsheet. It’s a model that is told Mullaji’s matchmaking secrets and each user’s preferences,10 and then tries to assess if each person on one side of the dating pool is a good match with each person on the other side. And if it works—if the suggestions are good, which, to be seen, I suppose—then Sitch’s users probably won’t care how the expert manages its lists.
In fact, this is probably the ideal version of an app—or, more generally, the ideal version of software. It’s a thing that I tell what I want, and it manages the rest. It’s an inverted version of SaaS: Not software as a hosted service, nor a custom piece of software that I have to build, but a service—a matchmaker, an EA, a sales operations team—that is replicated by software, and software that acts like a specialist and a spreadsheet.
As Randy Au points out, expertise isn’t just embedded in software. It’s also in physical products and even physical spaces:
Most people want the guidance afforded to us by the various cues built into a space. Those cues, leftover shelves on a wall, the placement of a vent, outlet, rug, or even a door, are essentially the built up opinions of the original designer and anyone else that used the space before. Those ideas may not be the absolute best use of a given space, but they're at least a sign of something that had worked for someone in the past. That sort of signal has some weight to it.
A question I have long wondered: How did dating apps become The Apps? Like:
A: “Do you use the apps?” / B: “Yeah, it’s much easier to call a car than hail a taxi.”
A: “Do you use the apps?” / B: “No, I prefer to eat out than order DoorDash or Uber Eats.”
A: “Do you use the apps?” / B: “I do, my doctor’s office finally let me start booking appointments through them.”
A: “Do you use the apps?” / B: “Bro. I’ve been YOLOing crypto options since Bush’s presidency;* my Robinhood user ID is four digits; I’ve used the apps for a decade.”
All of these could’ve made sense! There are so many apps! Ride-sharing apps, food delivery apps, booking apps, trading apps! How did dating apps end up as the apps?
Don’t take this example too literally. The point here isn’t about MEDDIC versus MEDPICC—which is somehow a real thing—but that teams will sometimes want to operate in specialized ways that cut against the grain of a product that’s made for a more general audience. Ready-to-wear clothing can have plenty of stylistic opinions, but it will never fit as well as something made-to-measure.
This is so popular that the market for Salesforce consultants who build these customizations, at $18 billion a year, is about 50 percent as large as the market for Salesforce itself.
They’re also general experts who likely know things about how to design a workflow that you may not. That’s the downside of vibe-coding a custom app. You know yourself, but you may not know the field as well as other people. It’s like making your own clothes—you know what you like, but you probably don’t as good of a general eye as a real tailor.
Here’s a question I have, after this acquisition: How will Superhuman deal with grammatical typos now? On one hand, it’s now owned by Grammarly, whose whole thing is fixing bad grammar. On the other hand, sending rushed emails with bad grammar to show everyone how busy you are has always been Superhuman’s whole thing. Will people want Grammarly recommendations inside of Superhuman? Will they want it to suggest bad grammar, to make them look even more rushed and important? If I type “Thanks, Benn” inside of Superhuman, will it tell me to change it to “thx, b”?
This is a bit of an exaggeration; you might still check your email. But the inbox that you look at can be pretty basic. The same is true of Salesforce—people will still probably want to look at some lists, but they wouldn’t need all of the operational infrastructure around them.
Another notable difference between Sitch and traditional dating apps is that Sitch apparently has a relatively long onboarding process. But that makes sense: In the specialist-with-a-spreadsheet model of software, you have to “onboard” the specialist in the same way you’d have to work with a real matchmaker or executive assistant. And people are generally perfectly happy to do that, provided that the training makes them good at their jobs.
brilliant breakdown, i really appreciate how you unpack saas as not just a tool but as a codified form of expertise.
Great read 🙏