Why we killed our money-making product 27 days after launch

Startups are hard.

I knew that. Course I did. I’ve worked with hundreds of them. Kev and Jo knew that. They’ve built plenty of them.

Still, the three of us weren’t expecting to be sharing the story of why, less than a month after launching, we decided to close down the product it had taken us half a year to build.

Since we’re always the first to encourage others to share their experience to benefit others, here’s the story of the life and death (and rebirth) of Ricochet — and what we’ve learnt along the way.

We started Ricochet to help B2B SMEs make more sales, by automating activity at the top of their sales funnel. Discovering new business opportunities, qualifying them and then engaging them — it’s a collection of laborious, time-consuming, manual processes. We wanted to make life easier, smarter and faster for salespeople and CEOs.

We couldn’t build everything at once, so we started with the engine that could find and qualify businesses. We used machine-learning and nearly 200 data sources to discover and determine a business’s industry, sector and specialisms; algorithms would identify how that business compared to a client’s current customers, which we’d qualify by comparing it our client’s customer profile.

After several iterations and multiple rounds of user-testing and customer conversations, we added a leads inbox; like email, but it would populate with a handful of qualified leads every day. Our clients were one click away from the relevant sources of information they typically used for due dilligence, like Companies House and LinkedIn. We explained how we qualified the lead and we supplied any contact information (email addresses, job titles, phone numbers etc) that was publicly available.

This was our MVP. We trialled with nine clients for a month, launched publicly two months later, and within two weeks had a total of 24 paying clients. In total, we took ￡2,000 in revenue, without any paid marketing. It wasn’t a difficult sell.

If only it had been that easy. It wasn’t.

We failed to meet expectations…

We assumed that so long as some leads hit the spot, our clients would be happy. If 4 or 5 leads out of 10 looked promising, that seemed reasonable to us — and we could deliver on that. We’d built a sophisticated engine that got results broadly right and, importantly, we could improve iteratively over time.

Our assumption was dead wrong.

When you search for company information on Google, you might receive tens of thousands of results, but even if the vast majority of them are terrible, you don’t complain. After all, Google is only the tool.

However, when you essentially offer to do somebody’s job for them, it’s not unrealistic that they expect you to perform that task as well as they could.

That’s a subtle dynamic we didn’t spot before we launched. We didn’t appreciate that by absolving our clients of all responsibility, by offering to find qualified leads for them, they’d expect every result to be right.

Building an engine required to meet naunced and complex client expectations is a long-term engineering project, and not a place to start from. Very quickly, we began receiving feedback from clients that we were missing the mark. There was friction in our product from day one.

So we experimented. We started augmenting the automated leads with manual results. We suddenly had to find another 20 hours in the week to service two dozen paying clients. Feedback improved. Clients were happier with the results, but were still prone to criticise for arbitary reasons we couldn’t hope to predict.

…because we weren’t where we thought we were…

What did we mean by “qualified business leads”? We meant we’d used machine learning to establish that a business looked similar to your existing customers, and also had a similar need. That’s what we told clients, carefully explaining it on our website, in our onboarding, everywhere we could.

Our clients didn’t neccessarily agree. Partly because people don’t neccessarily read small print on websites, but mostly because a “lead” means different things to different people.

While we’d saved them the effort of looking, they still had work to do. Even though we could provide publicly available contact information, identifying specific stakeholders wasn’t going to happen until the next iteration of the product. Again, we saw this as an iterative process. Again it was causing friction from the start.

We thought we were at the top of the funnel, solving problems at the very beginning of the sales process. Clients felt we were further down the funnel, at a point we should have provided far more value if we wanted to talk about “leads”.

… and we did things that were unscalable.

When Paul Graham talked about “doing things that are unscalable” in the early days of a startup, it’s important to remember the context. Graham referred to undertaking effort that may be unwieldy and inefficient in the short-term only, in order to learn, gain an advantage, or convert those first few customers.

A really good example of this, is how we committed to talk to every new client in-person or on the phone, to learn more about their circumstances and sales processes. Clearly this isn’t practical or even possible if you have thousands of signups, but it’s entirely doable for a small team to make a one-off effort for the first few dozen sales.

Supplementing our daily automated processes with manual effort is a terrible example of this mantra; because the effort required had to be repeated daily for every client, every new sale compounded the problem. Ricochet would have effectively become an agency; our growth would be determined by our ability to recruit staff, rather than acquisition or efficiencies of scale.

The decision

While we likely compounded these issues with our messaging, it was clear we couldn’t iteratively improve our product in daylight over a number of months — and that assertion was underpinning our entire roadmap.

If we’d nailed the execution in our first attempt, it wouldn’t have mattered. But we hadn’t. We had a product that wasn’t terrible, but it wasn’t great, either:

Two weeks after launch, the difficult conversations began. There was a fraught phone conference last Monday — professional, but emotional. We could have blown apart at this point. I’d not slept more than 5 hours a night in over a fortnight, and spent my pre-dawn hours in a silent, panicked mania.

After a few days of distributed working, we were together in the office last Thursday morning. I expected a fight.

It didn’t happen. We resolved any ill feeling within the first 90 seconds. Within 30 minutes we’d broken down the data, the customer feedback and could articulate the problems — and the opportunities. And that’s when we decided to shut down Ricochet as it was; it was a decision based on subjective feedback and objective data, and gut. It wasn’t working, and it wasn’t going to get better.

Before lunchtime last Thursday, we’d started outlining a pivot; the vision remained the same, but we’d start in a very different place, with a product that was deliverable, scalable and aligned with what we’d learnt about our market.

After a full night’s sleep, last Friday morning brought further clarity, brains’ full of ideas and iterations, and a clear path forward. By Friday afternoon we’d started prototyping our new product, contacted our clients to explain our decision to close the service, and drafted an investor update to do the same.

And by yesterday, three working days later, Kev was benchmarking the tech and Jo was sharing her new designs for Ricochet 2.0:

Two critical lessons were reinforced for me over the past week:

• The right co-founders will be the counsellors and cheerleaders and critics you need and deserve, so be the same for them;
• As much as distributed working is vital and neccessary in our own team, it’s difficult to work through emotional situations when you can’t read the room.

What happens next

The last seven months were far from a waste. Significantly, we’re now able to articulate exactly who our target market is; we understand our potential customers are determined not only by their needs and behaviour, but equally by the behaviour of their target market.

And while we’re not quite back to the drawing board, the next few weeks mean more prototyping, more customer conversations, and more user testing.

Startups are hard, it’s true. Going all in on Ricochet is the most harrowing, brilliant, confusing, exciting, despondent, frustrating, hopeful thing I’ve ever done.

But getting it wrong doesn’t mean we failed, this time at least. It means we’ve learnt a way not to do it, and that may just be the key to getting it right.

We’re still building our new product, but if you’d like to be the first to hear about and try our new solution for B2B SMEs, you can join the Ricochet mailing list here.