Clouseau: A Postmortem

How I vetted and dumped a startup idea in ~20 hours and for under $1000.

The cheesy logo for the project

As the head of Google’s Rapid Rollout Lab, I help direct our explorations as to how our team can impact how quickly and cheaply Internet infrastructure can be designed, built, and operated. I’m also an investor in a number of startups (e.g. via my drone.vc and neuron.vc syndicates) and I thought it would be interesting as a weekend exercise to combine a “rapid prototyping” style exploration with a business concept not remotely related to my day job. To emphasize: all of the below exploration was done on my own time and budget and was not part of any Google exploration.

The overall thought here was that it would be interesting to help jumpstart a small business on the side and have one of my friends run it. Even if it never became giant, it could potentially generate meaningful returns via dividends for a solitary investor (me).

The Problem Statement: When I toured Northern Europe with my wife and young son last summer, one of the things that surprised me was that even the fancier hotels didn’t do a very good job making it dark in the rooms. Actually, a number of the cheaper hotels did a better job. It didn’t seem obvious to me how to pick hotels that were going to afford a good night’s sleep or allow it to be dark enough mid-day to allow my son to nap. So I started thinking about a data product play that could offer an objective third party service to measure important hotel data, such as the “darkening power” of the shades in a room to see how dark you could get it. It seemed there really was no trusted authority on such a thing and that there was effectively a natural monopoly that could be created in offering such a service.

So what if we could create such an objective service that measured all of the different properties of a room that contribute to the quality of one’s stay? We could even get layouts and 3D models of the hotel to predict shading and sun exposure. Wouldn’t it be cool, I thought, if you could have individual rooms in different properties compete against each other? Maybe the north-facing room at the somewhat less fancy hotel is actually a much better overall experience than the poorly-shaded south-facing room at the nice hotel that noisily overlooks the pool? Hotels could properly monetize their assets and could improve revenues by actually investing in factors that are known to materially improve sleep quality. Slam dunk! You frame the whole thing as being the defenders and arbiters of sleep quality. Everybody knows how important sleep is and loves sleep, and that is most of the experience that a hotel is supposed to be selling you anyhow! I got excited.

I decided to start with light metering; use a light meter to measure how bright a room is with lights off and door closed but curtains open, and then take a second measurement with the curtains fully drawn. These two numbers should tell you a lot: the ideal room is very bright and airy during the day with the curtains open but is incredibly dark with the curtains shut. This was in theory easy enough to capture and could lead to follow on data products like shower pressure, water temperature after two minutes set to full hot, Internet quality, air quality, noise dampening, etc.

So I bought three light meters (Dr. Meter LX1010BS; $18.99) and started signing up my friends to go measure hotel rooms. A surprising number of them bailed out when they found out that I was asking them to talk their way into seeing a room. I had my lawyer draft up a simple NDA / IP release form that I sent out via a HelloSign template and saved into a Google Drive folder. I used Google Drive to organize the concept, progress spreadsheets, everything; it may just be that I’m used to doing this with projects at Google but it was a very quick way to manage a shared exploration with media, presentations, spreadsheets, etc.

I also thought it would be fun to come up with a clever code name for the project. Since we’d be “inspecting” hotel rooms, I decided “Clouseau” would be a fun project name. I registered clouseau.co ($30; Google Domains), @ClouseauCo, FB/Clouseau.co, and angel.co/clouseau and got a cheesy vector logo designed ($16; Fiverr) because hey, logo!

The light meters showed up next day thanks to the magic of Amazon Prime. I tested a few different measurements in my bedroom and quickly found that the fact that it couldn’t discern between kinda-dark and really-dark (all were <1 lux) meant it wasn’t all that helpful. Whoops. So I bought a fancier light meter (Extech LT300; $124.24)— just one this time — and found that we could take useful sub-lux measurements with that one.

My friend Chris Matthews ended up becoming my cofounder in this exploration and took on the challenge of charging into various local hotel rooms. We didn’t want to lie but we also didn’t want to have to actually book lots of rooms so we used a script of “I have an executive I’m working with for whom sleep quality is important, so I’d like to measure how dark your rooms are — do you have a moment to show me an available room?” worked pretty well.

We noticed that the fancier light meter had functionality to measure the lowest and highest light measurements in a time window and averaging these produced more self-consistent results but that there was still fairly large variation in measurement (though smaller, even percentage-wise, with darker measurements). My next door neighbor and friend Jonathan rigged a little analog light meter for us with a photodiode to illustrate what would likely be measurable / constructible as a measurement device. We got a few water measurement devices but didn’t find them terribly consistent or usable and they required more than casual calibration to function properly.

I licensed a database of all of the US hotel properties ($47, US Business Data) and in parallel had a team of remote workers from Upwork put together a county-by-county database of the lat/long, name, address, and star rating of each hotel in the county for a number of California counties. One of the big surprises was how few properties there actually are; while there are a few million hotel rooms in the US, there are only ~50k distinct hotel properties! I had four workers do a few hours each and assembled data for several California counties to judge approximate data acquisition cost and quality. I ended up spending a total of $496.24 for these acquisitions or $0.17–0.82/hotel (depending on the worker). Not too bad.

We brainstormed a little around creating an app or phone accessories that could measure some of the data we were talking about — this could let us capture a much larger swath of data in theory, like finding out when exactly it gets how bright in room 537 at the Hyatt downtown or how noisy the rooms on the first floor across from the pub get on a Friday night.

But before jumping into the next round of measurement and calibration we decided to dig back into the business validation side of things; what would it be like to license this data? How actionable could it be for people looking to rent a room? I decided to reach directly out to people in the industry and share the concept.

I called my cousin, who had worked for Hilton for a decade before moving on. She walked me through how important it was for hotels to do room assignments morning-of; having to hold certain rooms was a tremendous impediment to flexibility and efficiency since it’s quite difficult for them to predict which guests are going to show up when and even which guests are going to check out which days sometimes. So it’s a huge hassle for them to book a particular room for a particular guest and they’ll generally only do it for really important constraints (e.g. a handicapped person needs one of the handicap-friendly rooms) and not on a lark. Everything else just gets recorded as a “note” on the guest folio to help guide the same-day allocation process.

This was quite a setback because it meant that any per-room data we collected wouldn’t actually be actionable on the part of the traveller since they would be unable to book a particular room. We could only give general guidance about the quality of an overall property and not realize our vision of baking off individual rooms in different properties against each other.

I then spoke with an acquaintance who had spent time in the industry doing a startup that provided room-level hotel data and he confirmed the same thing — that doing even trivial deals like acquiring floor plans could take as long as a year for a property, and that a vision of per-room advance bookings was not likely going to be realizable. (We did note that the HHonors app allows members to pick a room at time of checkin on their mobile device, which is neat.) He also pointed out TrustYou, which aggregates property reviews and semantically parses them for positivity and specific feedback in several different categories and provides this feed to a number of companies. While it’s not “100% accurate” it’s good enough to extract actionable insights for a property. Sadly, the incremental data does not apparently produce large conversion lift — price seems to be the only thing that can swing a large number of people to one property versus another.

So even if we could ourselves directly, objectively, usefully measure room properties and make it available as a data service, it would only very minimally likely affect consumers’ decisions to book at one property versus another. There are a limited number of parties who you could license the data to: Google, Kayak, TripAdvisor, etc., so your total addressable market size would be pretty darn small. And given the low impact on conversions, it’s questionable how valuable the data really is.

Those conversations really made me pause; while nobody is collecting and offering the data, if it’s not actually going to be actionable, doesn’t drive decision-making, and isn’t going to prod properties to improve, then it might be a solution looking for a problem. While manually collecting room quality data for at least one room in every property in the US would likely cost less than a million, it might take years and the business isn’t interesting enough to pursue as-is. So I decided to hit the pause button on it, less than two months after beginning, and spending probably a grand total of about twenty hours in the exploration, plus another two hours writing up this essay.

Overall, I tried to be pretty time and cost efficient; leveraging existing resources, third party firms, friends, public data, and industry insider feedback to avoid wasting time on something unlikely to yield fruit.

I’d love your thoughts here about the process, whether you feel I came to the right conclusion, and what you might have done differently. I figured openly sharing “failed startup ideas” could be useful for giving some visibility into what is typically a fairly opaque process for people who are prone to try and hide both early explorations and their known-bad-ideas.