When the internal tool is better than your actual product (best ever pivots)

On 12 August 2026, Indie Hackers ran Lukas Böhler's story (opens in a new tab). He built a tool to run his own agency, BoehlerBrothers, and it now does over $83k a month as Gleap. Six days later they ran Cecilia Razak (opens in a new tab), who built something at a previous business to make remote work less awful and grew it into Slides With Friends at over $40k a month. Six days before Böhler, Sergiu Chiriac (opens in a new tab), who built a tool to submit his own products to launch directories and now runs a portfolio doing over $10k a month, mostly from that one.

What these all have in common is the tool someone built to get their real work done turned out to be worth more than the real work.

The famous version is Slack. Stewart Butterfield's games company, Tiny Speck, was building a browser MMO called Glitch with an exec team split across Vancouver, San Francisco and New York, so they built themselves an internal messaging tool (opens in a new tab). Glitch was killed in November 2012. The messaging tool opened as a preview in August 2013, went generally available in February 2014, and Salesforce announced it was buying it for $27.7 billion in December 2020 (opens in a new tab). Butterfield had done it once already, since Flickr was a tool built from a failed game called Game Neverending.

The list is longer than you might think…

37signals, 1999. A web design consultancy. Jason Fried brought in a Danish contractor, David Heinemeier Hansson, at $15 an hour to build an internal project management tool, because nothing available handled client projects cleanly. Clients saw it and asked for their own. Basecamp went on sale in February 2004, and within about eighteen months it was out-earning the design business (opens in a new tab), so they quit web design. Hansson pulled the framework out from underneath it and released that separately in July 2004, which is where Ruby on Rails came from. (An internal tool that produced a second internal tool that became a web framework is greedy, frankly.)

Rocket Science Group, 2000. Ben Chestnut and Dan Kurzius ran a web design agency. Clients kept asking how to send email newsletters and everything available was either enterprise-priced or unusable (opens in a new tab), so in 2001 they built one. By 2007 the side service was ahead of the agency and they wound the agency down. Intuit announced it was buying Mailchimp for $12 billion in September 2021 (opens in a new tab).

2ndSite, 2003. Mike McDerment ran a four-person design studio, saved over an old invoice in Word, and spent about two weeks coding a replacement (opens in a new tab) from his parents' basement in Toronto. That became FreshBooks.

Snowdevil, 2004. Tobias Lütke and Scott Lake wanted to sell snowboards online, found the ecommerce software clunky and expensive, and built their own on Ruby on Rails in about two months (opens in a new tab). Other merchants who saw the store working asked to license the software. Shopify launched in 2006.

Amazon, 2002-2006. Bezos issued the API mandate telling every internal team to expose its services as though they would be sold to outsiders. S3 shipped in March 2006, EC2 that August, and AWS booked around $21 million of revenue that year. In Q2 2026, AWS did $42.2 billion of revenue and $16.6 billion of operating income (opens in a new tab), against $27.5 billion of operating income for all of Amazon. The internal plumbing now throws off about 60% of the profit of the shop it was plumbing for.

Geni, 2006. David Sacks' genealogy startup built an internal Twitter-ish tool so staff could talk to each other, used it for about six months, then spun it out in 2008 as Yammer (opens in a new tab). It won TechCrunch50 that September. Microsoft's $1.2 billion acquisition completed in July 2012.

Invoke Media, 2008. Ryan Holmes' agency was running client campaigns across Twitter and Facebook, and Twitter had no support for multiple accounts. Seven of Invoke's 21 staff went onto building one (opens in a new tab). It shipped as BrightKit on 28 November 2008, got its name from a $500 crowdsourcing competition, raised $1.9 million from Hearst and Blumberg, and spun out over the turn of 2009 into 2010 as Hootsuite.

Crew, 2013. A marketplace for creative talent could not find a decent photo for its own site, hired a photographer, and had leftovers. Ten of them went up on a $19 Tumblr theme three hours later (opens in a new tab), with a link back to the homepage. Mikael Cho's numbers: 20,000 visitors, then 50,000 ten minutes after that, then the top spot on Hacker News. Unsplash outgrew Crew and was sold to Getty in 2021 on undisclosed terms.

Hammer & Chisel, 2013-2015. A tablet games studio raised $8.2 million led by Benchmark and shipped Fates Forever to good reviews and few players. While building it the team noticed that voice chat on mobile had barely moved in a decade (opens in a new tab) and started a skunkworks on it. Jason Citron bet the company on that skunkworks. Discord opened to the public in May 2015.

Nine companies. Three of them sold for a disclosed number, totalling $40.9 billion, and a fourth went to Getty for a price nobody published. In every case the winner started life as overhead.

Why the internal tool keeps winning

Some interesting reasons, and luck doesn’t feature as heavily as you may first think.

• The demand was already paid for. An internal tool exists because someone was carrying a pain that felt real. McDerment lost an invoice. Lütke was paying for ecommerce software he hated. Invoke was logging in and out of Twitter accounts by hand across client campaigns. That’s straight-up willingness-to-pay evidenced in wasted hours, before anyone was asked to build anything to fix it.
• The feedback loop has no latency. The builder is the first user, so the gap between noticing a problem and fixing it is however long it takes to open the editor. A normal discovery cycle puts many more steps in that gap.
• Distribution arrives attached. 37signals' clients asked for Basecamp because they were already inside 37signals' projects. Merchants asked Lütke to license his software because they could see Snowdevil working. Crew put a link home on the Tumblr and got the front page of Hacker News. The service business is the distribution channel, which is why agencies and consultancies produce so many of these. A captive audience with the identical problem, who have already paid you and therefore already trust you.
• Scope stays fenced. A speculative product has no natural edge, so it grows features to cover cases nobody has hit yet. An internal tool gets built up to the point where the work stops hurting and then stops. Böhler's, Razak's and Chiriac's were all small enough to finish, which is pretty much the only reason they exist.

The case against

Dan Norris has argued for years that "scratch your own itch" is bad advice (opens in a new tab), because you’re signing up for a product where you have no way to gauge demand from a sample of one. N. Taylor Thompson made a version of the same case in Harvard Business Review in May 2014 (opens in a new tab). Norris' own worked example is an analytics dashboard he built for a problem he had invented in his own head.

They’re kind of right; there are far more examples than not of internal tools that stayed internal, because nobody writes about these.

But quite honestly, in the age of AI development when we can knock together an MVP in a weekend - who cares?

What separates the ones that got out

Looking at these nine examples against each other, two signals show up in all of them.

Someone outside asked for it, unprompted, before it was for sale. Clients asked 37signals for a copy. Merchants asked Lütke to license Snowdevil. Not "would you use this", which people answer politely, and not a survey. An actual request, arriving before anyone pitched.

And the tool solved a problem the buyer had for reasons unconnected to your business. Böhler's agency tool works for other agencies. Invoke's scheduler works for anyone with several Twitter accounts. Where the tool only makes sense inside your particular way of working, you have improved your own operations, which is worth doing, but is a different thing.

Slack clears both. Other companies who saw Tiny Speck using its internal messaging tool asked for access before Slack existed as a company.

The bit to remember

Butterfield's second act is one of my favourites. Tiny Speck raised money and hired a team to build Glitch, and Glitch is what failed. The messaging tool had no roadmap, no budget, and nobody whose job it was to build it. It won anyway.

If a tool nobody was assigned to build beats the product everyone was assigned to build, you might wonder what the deliberate product got wrong that the accidental one got right. I believe it’s that the accidental one was never allowed to be speculative; it was only ever permitted to solve a problem that had already happened, to someone who was already in the conversation.

And the magic is that there’s probably no way to make that deliberate without ruining it, which is why the pattern keeps producing accidental runaway successes.