DataMarket

» see datamarket.com as it lived, on the wayback machine

~ the goodbye ~

DataMarket started in Reykjavík in 2008, founded by Hjálmar Gíslason, who had already built and sold the bookmarking site Spurl a few years earlier. The problem it went after is one that has never really been solved and is not glamorous enough to attract many attempts. Public institutions publish enormous quantities of numbers, and every one of them publishes in whatever shape suited the department that produced it.

What the company did was ingest those sources and force them into a single structure, so that a series from Statistics Iceland and a series from Eurostat could be searched the same way, charted on the same axis and pulled through the same interface. By the time it was acquired it held around thirteen thousand datasets, which in its own internal accounting came to something like a hundred million individual time series. The World Bank was in there. So were the national statistical offices of countries whose data had previously existed as spreadsheets on a ministry web page.

Qlik bought the whole company in November 2014 for roughly thirteen and a half million dollars, or 1.44 billion Icelandic krónur, which for an Icelandic startup of that period was a serious exit. The founder's line at the time was that the acquisition put a data supply operation behind an analytics product. What it actually did was change what the thing was for. DataMarket had been a public site anybody could open. Qlik DataMarket was a licensed package bundled with Qlik Sense.

The end was announced as a lifecycle notice rather than a shutdown, retirement of the product and of access to every data package on 30 June 2021, with customers advised to stop upgrading Qlik Sense around the end of 2020 if they wanted to keep using it. Qlik's stated reason is unusually honest for a document of that kind. Customers had grown more sophisticated, and their external data needs had split into the very simple, which they could now fetch themselves, and the very specific, which no general catalog was going to cover.

~ the chronicle ~

~ the pitch ~

The pitch was never the data, because the data was public and free by definition. It was normalization. Somebody had to read each source, work out what its columns meant, decide how its revisions and its missing years should be represented, and then keep doing that as the source changed format without warning. That work is invisible when it is done well and it is the entire product.

I know this category from the wrong side of it. As an engineer you meet public statistics when you need a number for something else, and what you find is a download page, a format from 2004 and a units column that changed definition in the middle of the series. DataMarket was selling the absence of that afternoon. It never crossed my desk and I only read about it years after it had already become a line item inside a business intelligence license.

~ the acquisition ~

Being bought by an analytics vendor is the obvious exit for a data company and it is also the one that ends the public thing. Qlik did not want a website. It wanted a supply of clean external series to sit next to a customer's own numbers inside a dashboard, which is a good product and a completely different one.

The consequence is that everything DataMarket had built as a public resource became a feature of a paid enterprise tool, and then a feature that had to justify itself against every other feature in that tool. Features lose those arguments eventually. A standalone company can be unprofitable for years while it is still the only thing of its kind. A feature only has to become the least important feature once.

~ what is left of the idea ~

Gíslason left and went on to build GRID, which is a spreadsheet product and is very much still alive. The idea he had at DataMarket has not gone anywhere either, it just stopped being a company. Every statistical office now runs its own API, which is both the vindication of the argument and the reason nobody would fund the middleman again.

What did not survive is the part that made it useful, the single index across all of it. Thirteen thousand datasets behind one search box is a different object from thirteen thousand datasets behind four hundred separate APIs, each with its own key and its own idea of what a date looks like. That object no longer exists in a form anybody can open.

@}-->--

              )
             (_)
            .-'-.
            |   |
            |   |
            |   |
            |   |
          __|   |__   .-.
       .-'  |   |  `-:   :
      :     `---'     :-'
       `-._       _.-'
           '""""""

« previous  ·  random  ·  next »