The Hardest Thing

4 min readSep 24, 2015


When Matt Clifford interviewed me for the not-an-accelerator he co-founded with Alice Bentinck, I asked him the most common reason early stage startups fail.

His answer was that the number one reason for failure was making something nobody wants. I agreed. Then he added that even among the people who know that fact to be undeniably true, it is still the number one reason they fail.

In a way, that’s natural. Ideas are personal, and make us vulnerable. Our instinct is to protect them from the real world.

It’s even worse if you know how to build things. Building things at the beginning feels like work. As anyone who has ever done the dishes instead of their tax return knows, activities that feel useful are the most pernicious form of procrastination.

I’ve made something nobody wanted more times than I can count. At the Guardian, users used to routinely tear our prototypes apart. And I’ve had more than one side project take up months of my life without anyone ever using it.

What has made me more effective as I have gone on is that I now take much less time doing it.

Matt and Alice’s program focuses a lot on how you avoid doing this as a B2B company. They rightly put a lot of faith in deep defensible tech, and a great way of monetizing deep defensible tech is to find a company to sell it to. The basic strategy is to find around five companies who will pay for your thing before you have written a single line of code. Hopefully, these are going to be, in Marty-Cagan-speak, your reference customers.

First, let’s be clear what it means to be a reference customer. This is a real customer (not friends or family), that is running your product in production (not a trial), that has paid real money for the product (it wasn’t given away to entice them to use), and most importantly, they are willing to tell others how much they love your product (voluntarily and sincerely).

B2B is great, but I want to talk briefly about how to attack this problem of building something people want from a consumer internet perspective.

In particular, how do we validate demand if we are building something consumers will probably not directly pay for? People are famously terrible predictors of their own future behaviour, so the next question is useless in both B2C and B2B contexts:

“If this existed would you buy it?”

The trick if you’re selling to a company of course is to get them to buy it in advance, or at least get them to commit to buying it. In a lot of cases consumer facing businesses don’t have that option.

The only proof that a consumer facing product isn’t something nobody wants is that people use it. Which is tricky, because as we already know the worst thing we could do is build it.

The Product

My current obsession is creating news feeds from scraping and modelling data. Rather than writing articles about everything, I want to gather information algorithmically and serve up the relevant facts to the people who need to know them.

On paper, a great place to apply that thinking is local news. What if we could show you all the cool stuff happening around you in a way that was really concise, informative and personally relevant? It’s certainly a market in desperate need of something new.

Does anyone want that?

The thing to search for is the core assumption that can make or break your business. Here, the assumption is that people actually want information about where they live. How do we find that out with as little code as possible?

I started on paper. I invented a fictitious local news product called Stokey News, which catered to my area of London, Stoke Newington (colloquially called “Stokey”). Yesterday morning at 6am, I manually scraped about ten information sources and wrote by hand what this product might have looked like that morning:

Over the rest of the day, I got hacking. By the evening, I had a functioning site,, with an HTML rendering of my handwritten scrawl, and an email sign up box.

The next step is simple. I set up a Facebook ad, leveraging their terrifyingly good targeting rules to show our promotion only to Stoke Newington residents. The next morning, rise bright and early and once again write the news, by hand, for that day.

Variations of this method can test just about any consumer facing idea inside of 48 hours.

The key output is a set of conversion rates. Of those who saw your ad, you know the click-through. After click-through, you have bounce rate and time spent to quantify engagement. And you have your email sign up box, so you know how many people are seriously interested.

This kind of approach can kill most assumptions in a day. Of course, Stokey News isn’t likely to look much like our finished thing, if we ever do build it. But if it proves or disproves our core assumption, it will have been well worth the few bucks it cost to set up.




product human. formerly @FT @CondeNast @theguardian