Why Many On-Demand Platforms Fail
themacro.com
themacro.com
Retention is the most common driver. Even good virality can't overcome bad retention.
A rough advance estimate of retention is the "toothbrush test": do people use it as often as a toothbrush. Services used rarely are more likely to be forgotten by the next use. If they need to find you again on Google, your competitor can buy them with Adwords.
I saw a startup once for a mobile app to arrange funeral services. I had to chuckle because that may be the lowest possible score on the toothbrush test.
We've all had bad taxi rides but overall the variability in possible service quality for Uber is pretty low. But Homejoy died because so many people didn't like the quality of the service.
I honestly question the premise that people want more on-demand services, given that the major companies that are succeeding in this space were already on-demand (hotels, taxis, food delivery).
What market has shifted from not-on-demand to on-demand?
Ugh - I have heard this before - its such an unhelpfully high level that it fails its purpose as target setting advice. Almost nothing is a "twice a day" service. Whether its Uber or Pizza, the vast majority of users do not use them twice a day despite their love and loyalty. For the few businesses that are e.g. FB, Slack etc. its because they are communication oriented rather than task oriented so they are more like monitoring a feed than using a tool.
Sure, its wise to be sceptical about services that are once per life "AirB&B for gravestones" or once per year "Uber for Christmas decorations" but your once per month billing/invoice system is likely to have high retention once you get customers locked in.
Note how sales velocity and retention are similar, can you retain customer interest long enough to close a sale?
There are successful occasional use products. It's not meant as a threshold or black and white test, its just one factor to consider when choosing which business to start.
That fails to capture the power of recurring services. If a user needs to think of your service then sure, it needs to be a toothbrush. When the customer schedules your service, enjoys the benefits and pays without thinking about it you don't need to be a toothbrush (a lot of the SaaS world).
Then there's the insane profits between those two models like TurboTax. The toothbrush analogy is a nice way to categorize certain businesses, but it's only one of many profitable consumer software models.
Servicemaster operates in that space. Some workers are employees and others are franchisees who may have employees of their own. That works, and Servicemaster, after 85 years of operation, has a market cap of $4.5 billion. Not a get-rich-quick scheme, but a success. HomeJoy tried to compete with them, not too successfully.
Then there are the lead generation services. Plumbers have these; most of the ads in what's left of the Yellow Pages go to the same call center. They don't do much and they don't make much.
Operating somewhere between those points seems to be difficult. It's not clear yet if Uber makes it; they're still losing money, and eventually, the investment money stops flowing in and they have to cut costs. (Look at Twitter. Growth is not enough.)
1. I am talking about Uber's current business. If they can solve the automatous driving problem before they run out of cash (or fools) then they will be a good investment.
The largest issue is failing to understand the difference between a commodity and a service, and how easy a substitute is to find.
Others vs uber(and uber like companies) are different
Uber solves an extremely hard problem that looks easy, it also has very few alternatives. The other services solve a problem that is less urgent, had many substitutes and is largely less valuable.
Uber provides transportation when and where it is needed. If you simply need (using the most well known detivative business) to borrow a skill saw from your neighbor for a tool sharing site you have many options. This project may be planned in advance, you may be able to purchase a used tool, and you may be able to borrow kne free.
The issue is how variable a service is. Uber can generate transportation for where and when you need it. However it may appear that the commodity is transportation, but that is the variable. Finding a user and providing them transportation from an unknown location, to an unknown location, immeadiately is hard. A ride from Logan Airport to south station is much different than a rake/hammer.
Each ride has a large amount of logistical constraints. Out toolsharing example has to solve all the same problems except a tool is always the same tool. A ride is never the same ride. It makes it much less valuable, and much easier to compete as there are often many substitutes and less expediency. Therefore coverage & on demand has a sup/demand musmatch, is prohibitively expensive and can only take a very small slice if value.
Uber is fundamentally unlike all of these other "Uber for X" startups. First of all, @paulsutter made a great point: the CAC is > LTV. How many times in your life are you ever going to have to hire a...home decorator or something? A painter? The reason why Geico advertisements are on all media literally all the time (they have like five completely different mascot strategies!) is because you almost never make a choice about car insurance, and when you do, they need to be top-of-mind. This is the same problem that affects a huge swathe of terrible startup ideas, not just on-demand services: the whole category of travel management, niche social networks, event-oriented tools, and yes, most "Uber for X" products. It works for Geico because the LTV of a Geico customer is high enough, because Warren Buffet is an amazing capital allocator and amazing at identifying and retaining operational talent, so Geico runs at roughly break-even on the insurance product, but frees up a large chunk of float for BH to deploy and earn high returns. That business requires doing like 10,000 impossible things right and they do it. An app that lets you summon a guy who trims trees to your house the one time you realize you need a tree trimmed has none of these qualities.
A lot of comments here are about how 'the jury is still out on Uber' because it loses money. But cabs don't lose money, collectively. Uber's aggressive money-weaponization growth strategy is certainly high-risk, but there have been tons of businesses that direct requests-for-transit-via-car to independent contractors and clip a fee: virtually every taxi dispatch company, say (contractor vs. employee depends heavily on the history and regulatory structure in a given market). The idea of having for-hire cars distributed through the city and used on an on-demand basis is also profitable for zillions of people. The idea of adding on an app that summons the cars more efficiently, handles payment, and clips a fee is not that hard to believe in.
OP frames the issue about the lack of variance in quality of an uber ride vs...what? Hiring a plumber? Do any of us have the ability to identify a quality plumbing job vs. not quality? I mean, the toilet flushes now, or it doesn't. I think this is _close_ to the issue, but the issue is more about 'what adds value': being close to me right now, and able to get to me immediately, on a pre-determined price schedule trumps everything in ride-hailing most of the time (sometimes you want a limo, or something special, and maybe you won't use Uber. Maybe you want to hire a driver for a month while you are in India. You probably won't use Uber. Different parts of the value equation are changing value). Most of these on-demand apps operate in markets that don't have this quality. Other things are important.
The post also goes into the 'Trust' question. This is a classic trap for technology entrepreneurs, right up there with 'building tools that makes it easier for non-technical people to build apps'. So many zillions of dollars and hours of smart-people-work have gone down the drain building web of trust, chain of trust, and other trust management products, and they basically don't work at all. What works is "Does this thing have a lot of five-out-of-five star ratings from real-seeming people compared to how many one-out-of-five-star ratings it has?"
This is essentially 'word of mouth' online. The ratio of 5:1 star ratings over a threshold. This is basically how all real-world functional trust systems work: ebay, amazon, yelp, etc. Yes they have enormous problems. Yes they are really stupid. Yes they can be gamed. Yes it weights all sorts of idiots equally. But guess what? It works well enough for most people to make purchase decisions. Every fancy smart thing we've ever come up with that's more sophisticated than this is basically useless (e.g. the global tls cert web of trust, the various pkis, every product that builds a FOAF trust web underneath arbitrary objects).
Afaik, on-demand 'uber for x' products work one of two ways, mainly: - as a marketplace, where they fail because of all the normal chicken/egg problems, but basically because the ltv of a customer is very low and the cac is very high (once you run out of VC cash to subsidize the service to extremely low prices.) - as an ONO service with employees, where the unit economics supposedly become good with crazy scale, but it's extremely difficult to get there again when you run out of subsidy cash . The classic example here is Kozmo. I'm sure Kozmo's model predicted that eventually, when a delivery person was delivering a candy bar to five people in the same building, the unit economics turn the corner, but it consumes insane volumes of cash to get there, if there's even a 'there' (I'm not aware of any company who has ever gotten to this point. It is possible that Postmates will be the first, if they survive).
Postmates is an interesting play because unlike most of these on-demand services, there's at least a world in which like you could _imagine_ if there was a small but positive way to get the unit economics to work, the LTV outstrips CAC because people need to order food and stuff all the time. Many times a day possibly. Even if Postmates made a penny/order on each active user, you could have a very high order volume over a very long time that would eventually get over the CAC cost. BUT the idea of the unit economics getting there is...hard to see. I have a friend who orders coffee from postmates _every morning_ instead of making it or going out. He just has a postmate drive up the hill to his house from starbucks. Starbucks! It's not even like a fancy coffee place. It's interesting to think about a world in which that postmate is running like, a hundred starbucks orders on an optimized loop, and they're still able to charge $5 or whatever for the delivery. That starts to look _possible_ but insane. As it is, it's twenty minutes up the hill, twenty minutes down, never mind the starbucks time.
The other world of on-demand stuff I think is interesting to think about is grocery delivery because I think grocery delivery is basically doing something different than what it thinks its doing: the on-demand delivery aspect of it is a vehicle for grocery price discrimination.
It's an old joke that rich people don't know how much food costs: George Bush Senior at the checkout counter, Lucille Bluth saying, "it's one banana, michael, what could it cost? Ten dollars?" That 30 rock episode where Jack Donaghy says something about the "grocery concierge" telling you a sack of potatoes is four hundred dollars.
But groceries remain an insanely low margin business. Grocery chains operate at around 2-3% margin. Independent grocery resellers in immigrant neighborhoods are even tighter. Costco's grocery business is run near break-even (they make money on the membership fees. This is also why, long run, Amazon's non-AWS business best comp is Costco.)
I've shopped for groceries on amazon pantry. Sometimes I'll type something in like "soy sauce" or "rice". Is that a good price? I sometimes literally don't know if its 2x the normal price or .5x. There are things I know the price of well: avocados, milk, liquor, coffee. But there are tons of things I don't know, and i've been buying these staples for years!
Grocery stores have very little power to raise prices, but if they could segment the market to provide a service like delivery to a price-insensitive consumer that allows them to charge 20-30% more for certain staples, the excess margin in the groceries completely dominates the cost of the delivery business.
This is the Instacart strategy. It's very interesting. A lot of the unit economics problems of these businesses go away if you just...make things expensive. What if coffee was $30? Sure, postmates would work fine, if anyone would pay. But groceries have larger tickets where a 20-30% increase in price might not be objected to by wealthier price-insensitive consumers. Instacart basically is selling coffee for $30.
I'm very curious to see if this works. I mean, it seems insane, but less insane than say HomeJoy, Rinse, Washio, Homee, Handy, the 'uber for kids', Lugg (this has gotta be the worst idea of the bunch), the one that parks your car, the one that washes your car, the one that fills your car up with gas, the (multiple!) ones for private jets, the massage one, etc.
Yes this is true, but any market that can support this sort of markup is going to attract a lot of competition. There are very few ways of keeping out competition in these markets.
What's your contact details, BTW?
Can you kindly expatiate on why building dev tools for non-technical people requires a web of trust?
For any given individual, evaluating whether such tools work for you should be a pretty quick exercise similar to test-driving a car. Why would you need a web of trust?
SOrry if that was confusing i was writing pretty stream-of-consciousness
a) You might not know what you normally pay for groceries but I think most Americans have a rough sense of what they should pay for their items. Even if they don't know, it will eventually become easy to find out what the market price is (e.g. Google probably will have this).
b) Amazon's retail business can sustain higher margins than Costco because it can sell non-popular items. I agree that mainstream consumer retail items are sold at extremely low margins. But if you want to buy something more specific, the margins go up because your options to buy this item become limited. This is a large reason of why AMZN worked so hard to develop a reputation as the "everything store"
Products like Tableau, R+R Studio, Ruby on Rails and even Excel have been highly successful because they lower the technical barrier needed for to perform computational tasks within a domain.
This isn't a criticism. Interested to hear your thoughts.
My argument is not that these tools are bad, or that there's absolutely no market, or that it's fundamentally impossible. My argument is that this market represents a trap to the kind of people who start technology companies: it looks very shiny and appealing to people with the exact psychological makeup to launch technology companies, and that makes it harder for them to see the challenges.
For every Excel there are a 1,000 Lotus Improvs. For every Ruby on Rails there are a thousand Google App Inventors. For every Hypercard there are a thousand Runtime Revolutions. For every IFTTT there are a thousand Yahoo! Pipes (and I'm not really sure if its fair to call IFTTT 'successful').
The linking feature of technology entrepreneurs is they are people who see problems everywhere that can be solved with technology. They are also generally 'generalizers' or 'lumpers' (which leads to other antipatterns: http://www.joelonsoftware.com/items/2008/05/01.html). These are necessary preconditions to being successful in this line of work: identifying problems, classifying them as tractable with existing or adjacent-possible technologies, generalizing disparate-looking issues into something solveable with one approach.
But almost no one else thinks like this. The stereotypical tech startup founder's favorite toy as a kid was lego. Now, lego is a popular toy. But most people when they get legos build whatever was in the picture on the box. Lego almost went out of business around 2000. They turned it around by getting aggressive on IP development and licensing that does some of the imagining for you.
Let's say you could invent the ultimate 'non-technical person app-maker' tool. It's a magic helmet. You put it on, imagine the software feature by feature, and then it 's in your Dropbox, seconds later. Most people would have absolutely no use for this helmet. This becomes really clear if you do a lot of consulting. People have the most insane descriptions of their problems and really have no idea what they need. It's sort of like Henry Ford's "if I'd asked the customers they would've said they want a faster horse". In technology we've seen this a million times. The most famous examples are probably why didn't GE or IBM own the photocopying industry, or why didn't IBM own RDBMSs before Oracle got out there. The classic explanation of this is the Clayton Christensen model: the new product threatens existing revenue streams and the company's internal culture protects them.
As a quick aside, I think there are a lot of issues with this bit of Silicon Valley religious orthodoxy: The Innovator's Dilemma i think has a lot less explanatory power than we generally accept. There are many, many markets where successive waves of technical development been captured by market leaders. The largest shipping company in the world by TEU capacity _or_ ship count is A.P. Moller-Maersk Group. Founder Moller-Maersk bought his first ship in 1886. This is the ship: http://media-cache-ec0.pinimg.com/736x/48/68/db/4868dbcbadb6... Not only did these guys survive the crazy waves of disruption that have hit the shipping business recently, they started with a break bulk _steamship_. Diesel, containerization, automated ship management...they're the dominant player in international shipping. 130 years later.
IBM owned 100% of the technology, theory, and experience in the RDBMS world then. They could have charged whatever they wanted for it. It was no threat at all to their existing database business. It could have expanded it above or below. Same with photocopying. IBM didn't have a large segment of...crummier...document duplication?
The problem in these cases is the fail to understand what customers need, and to explain it to them. It's much more basic.
But it also gets at the broad Silicon Valley disdain for 'marketing'. Marketing is both incredibly difficult and incredibly important. Packaging up a new, abstract concept in a digestible way that makes people change behavior is as important to driving the adoption of a new technological solution to an old problem as engineering.
Anyway, I'm getting really off-track and burning up my Sunday afternoon here.
The short version of my point is: It's easier to make non-technical people technical than it is to make non-problem-solvers problem solvers.
It is _insane_ what 'non-technical' people manage to get done with Excel, if they have the correct problem-solvey mind. Physics majors who end up as analysts at investment banks and build incredibly complex modeling and information-management tools in Excel.
But if you are the kind of person who can clearly identify an issue and envision a reasonable set of tools to solve it, there are tons of technical people around who can help you solve it, and if you want to learn yourself, its generally not that hard to get started.
The typical technology entrepreneur gets a job doing _literally anything_ and is surrounded by things that make them think "jesus christ! I can't believe we're still doing that X way! We should have a system that takes the foos from the foo queue and automatically bar-izes them..." And they are limited by the speed at which they can build software. Their implementation speed can not possibly keep pace with the tons of ideas they have! But imagine if...instead of solving the specific problem with the foo queue...they solved _the problem of solving problems_! Then they wouldn't need to build a messy salesforce to address the foo queues in this small backwards industry, they would just use a no-touch sales strategy to sell directly to teams who will bring the domain experience to solve any of these problems themselves!
This is why the marketing pages for even the most generic products in the world almost always talk more about 'solutions'. Like, companies don't even know they need a database. They know that they need something to keep track of products. They know that they need to keep track of customers. They know that they need to keep track of resources, inventory, etc. Customers think in terms of 'screens', not the fact that these are all fairly small things built on top of very complex database software.
So you start out inventing the RDBMS or the...SAN storage fabric...durable event queue...whatever it is. But you can't sell it to companies until you explain that no, it's a customer relationship manager. No, it's a way to store all your digital assets. It's a way to capture all your customer interaction data so you can make your digital marketing more effective. It takes an enormous amount of work to go from 'well-engineered technical solution to a general problem' to 'thing people will buy'. Imho, this was one of Marc Benioff's most brilliant moves: the company is called salesforce dot com. It's for your salesforce! If you have a salesforce, get this. We make CRM. Salesforce, of course, is really just a little lego box of hosted database configurators and form-builders, search engines, form hosts, etc. It's kind of like a cloud Microsoft Access that runs at larger scale, but that product would've fallen flat on its face (in fact it has, several times. the most elegant of these was probably Dabble DB) because no one wants a cloud-hosted form-builder + a database. They want a CRM system you don't have to install. They want to see the pipeline view, the contact view, etc. _on the website_. Not a video presentation where a guy at StrangeLoop shows how easy it is to make a Todo List.
It's an old dumb airport business book cliche to say "customers don't want drills, they want holes." These tools tend to be more like trying to sell customers a modular motor-and-battery-and-gear train set that can be used to create any tool: a drill, a saw, a sander..._they're all just battery-driven small motors with different gearing and tool-holders!_
So, the really messed up and interesting part is that the difference in the sort of 'consumer surplus' if you will between "cloud-hosted form-builder + database" and "cloud-hosted form-builder + database + we already made the contact-entry form and the pipeline view" IS FRIGGIN HUGE. No one will pay for the cloud-hostable database + form-builder. Or within epsilon of no one. That's also 99.9999% of the work. TONS of people will pay for the same thing with the contact form and the pipeline view ("the CRM system").
This is known as "The Software Industry". It's a completely messed up place! You see this all the time on HackerNews: commentors complaining that YC/A16Z keep investing in "things that are just skins on a database" instead of "hard tech". It's because no one pays for hard tech anymore. Operating systems are not a business. For decades, they were a multibillion dollar industry, but now they exist only as a loss-leader. Oracle survives because they sell the forms on top of the database (Peopleware, etc). HDFS, Kafka, HAProxy, Hadoop, Hive, Spark, Linux, Memcached, are all free. MIXPANEL costs tons of money.
The venture capital industry essentially exists now to exploit this huge bizarre situation: it costs a very small amount of money to see if you can add the last 5% to the collective pile of software that's free to make a very large amount of money. It wasn't always like this, but it is now. People will point to cleantech funds and health care investments, etc. but as of now, the total return on all invested VC with IT subtracted out is 0. The entrepreneur's tendency is to want to build the free 95%, but the economic return is in the last 5% you can charge for.
Tableau is also very interesting. I'm curious how people are using it in practice. In my experience, there are generally tons and tons of business people who say things like "i want to see the X and Y correlation blah blah blah which customer bought this that also went to one of our events..." and they have no ability to answer these questions. So they pay Business Objects Microstrategy or some big BI vendor to build a 'business intelligence' system...70% of the cost is consulting...and they end up with a handful of fixed reports and no ad-hoc querying.
Then they hire some sort of programmer-with-a-different-title to write reports.
Alternatively, they keep dumping the data into Excel and some clever 'non-technical' person muddles through using macros to clean it up, some combination of index/match/offset to prepare it, and pivot tables to make the final reports.
My curiosity about tableau is, in actual, real-world deployments, how ad-hoc are people's queries really? Also, how do they handle what used to be called ETL? It seems like an ongoing and challenging problem. It would be relatively easy to build a friendly-if-limited query and visualization platform on top of perfect data, but there is no perfect data in the world. In my experience, getting to the point where you can do ad-hoc queries on transactional business data means 90% of your resources will have been spent on ETL-ish problems, and about 10% on the 'fun part'.
But Tableau is obviously killing it so they must have found some magic combination of the right amount of query flexibility and manageable data...management. I haven't like, talked to their sales people or used the product in many years though so I'm not super familiar with where it is now.
OK LONGEST COMMENT IN THE WORLD OVER NOW!!!
I dunno how much of this is:
phycological - people have an optimism about the usefulness and difficulty of their work and deference to abide with convensions.
Stratigic - people somehow lose power and influence if their domain is standardised/automated or they admit high levels of uncertainty.
Lost in translation - You said that many people don't think like programmers in addressing problems, but I also often feel grated when businessy topics come up on HN and clever programmers appear to slightly butcher simple microeconomics. I probably do the same when discussing solutions to theoretical algorithm questions.
There's a value in establishing a common terminology and philosophy within a field because of increased communicational efficiency and agreement. This comes at the costs of alienating the less experienced and creating a clash on domain interfaces. Maybe if programmers and business execs knew more about each other's cultures then more useful tools could be made.
Take web pages as an example. Circa 1995, only a relatively few technical people could weave the HTML magic to create them. Fast-forward 10 years, and 5 year old kids were creating them for fun. The demand for web pages continued to increase and eventually the tools to help the majority of people create them emerged naturally.
IMO, the same will apply to app development tools.
I'm working on a startup that has built a PaaS that helps people with beginner or intermediate tech skills build apps quickly, and I agree that building the platform is like a guiding a mountaineer to Camp IV on Everest. Using the same platform to build a bunch of ready-made apps (that can be easily customized if required) is like helping them reach the Summit.
There is still a ton of value in building the platform because it helps us create a whole bunch of these ready-made apps that address different use cases very quickly. The cherry on top is that customizations are very easy and where the non-technical user has the problem-solver mindset, they can make those changes without us.
So, if you are uber and the only trust factors are solved by technology (i.e. where the driver is and how soon theyll get to you) it can create a new class of workers.
if you are in skilled home services, trust is built by the provider not the platform, so the only current value is lead generation for the worker...which doesnt require a platform.
The argument of this post is that there is an opportunity in home services for a platform to really cater to the service providers in a way that they always want to transact on the platform and thay they can use to manage and grow their steady customers...with on demand being the leadgen side of the platform.
A lot of businesses require subscription or other predictable recurring revenue to truly work. Even golf courses, bowling alleys, heating and air repair companies, bug terminators, advertising, etc. tend to rely on recurring revenue as the major source of revenue.
Even restaurants and bars rely on recurring revenue. You show me a restaurant or bar without regulars and I'll show you a restaurant or bar about to go out of business.
For many venues, events: golf and bowling tournaments. I don't think that works for pest control though....
Any service where the costs are dominated by labor and the enforcements of labor regulations is weak (i.e house cleaning services) is doomed from the start. The only question is why the "smartest guys in the room” missed this with homejoy.
As a consumer, they suck bogwater.
Yes, they ran out of cash because CAC > LTV, but that's a bit like saying an AIDs patient died of the flu. There's an underlying cause.
I'm co-founder of a p2p marketplace for dog sitters and dog owners and most of our sitters are not "pros" (as in people who do it just for money and full time).
IMHO that's the basic difference between a truly sharing economy platform and just an "on-demand service".
These two kind of companies tend to fall under the same umbrella, often because the ones that are just on-demand services use the "sharing economy" label to protect themselves against regulators.
On the other hand Uber for healthcare for example is impossible because you can't predict who will get sick when and where. It's impossible to know how many doctors you would need on a given day. So it's a lot of wasted money and doctor time.
More than anything, what seems to be lacking is long-term vision for the platform as a business.
Finding a someone a place to stay in jakarta, indonesia is much different than a place to stay in Boston, London, ect. The person, service, lication and time are all variables. This is complex, hard and valuable. The same goes for uber/lyft. There is expediency and significant variables to account for as each pickup and drop off is dynamic so while it appears to be a commodity; it isnt.
To your point, some ideas sound great and get funding but are not complex, urgent nor particularly valuable. On demand "lunch" competes with either direct ordering, a local option or brown bagging it. Eat 24 provides a directory & information search engine not physical delivery(in most cases). So while it seems like they violate what I said, they dont have huge infa costs. Other "similar" providers will go get you the food and deliver it. A user is not willing to pay a huge premium for that in most cases, and the volume must be significant amd clustered. Eat24 takes a small percentage for a software solution, the other type needs a large percentage for a phys solution.
So they comvince vcs to back them until they get "big" which they wont and achieve economies of scale. However, they cant, because expenses scale pretty in line in those models