Magical thinking and maintenance

On Excel-gate and the UK government’s “quantitative revolution”

Row 65,536 — the last row in an XLS file

In 2014 I was working at a service-design agency, doing discovery interviews at a big utility company — talking to stakeholders and finding out what could be done to make their billing system compatible with the “smart” devices being rolled out by the marketing team.

After a few conversations, it became clear that all roads led back to two system administrators — two women who sat in an unbeloved office at the back of the building, the only people able to do the magic required to spit billing data out of the system. The systems architecture was made up of, roughly, three parts:

  • On the left, a much-modified instance of some enterprise software that was installed c.1995, which couldn’t be updated because it would stop all the subsequent modifications working
  • On the right was a tool that someone who had now left the company had made that could (eventually) transform data from the left into information that could be understood by the billing system
  • And in between were the two women in the unbeloved office, and the incomprehensible series of procedures they needed to do to keep the information moving between the left and right

I’ve poked around in dozens of such systems over the years, in all kinds of contexts — from museums and theatres to banks and auction houses and hospitals. They are Heath Robinson bits of back-office complexity that one or two people understand, often regarded as too nerdy/difficult/boring (delete as appropriate) to benefit from the jazz hands investment and innovation narrative that gets money moving and stakeholders excited.

But, inevitably, at some point, they become important.

And that is usually when something breaks.

Anyone surprised by Public Health England’s use of Excel has perhaps never encountered such a system: a system in which legacy bits of technology are, at first, temporarily and incrementally lashed together because the time and money needed to update to a more modern, manageable system has never been found; a system in which DIY oddities become the basis for whole work flows; a system that always sort of works, supported by rafts of jobs, which is always just out of scope for digital transformation.

These kinds of systems are the meeting point of hard-edged, inbuilt obsolescence and the messy shifting ways that life happens and requirements change. They are what happens in a world where software updates can be postponed and modifications are possible, where the work of ten years ago is quaint and antiquated, but the modern and shiny is increasingly expensive and complex.

As Donna Haraway writes in Staying with the Trouble, “It matters what compostables make compost” — which is perhaps another way of saying, “Garbage In, Garbage Out”. Enterprise systems are not often created to break down into good compost, but they are created to eventually break down, so that new licenses can be sold and versions upgraded.

We live every day in the mulch of these legacy technologies but, politically, the importance of repair and maintenance is often denigrated in favour of innovation and novelty. Digital technologies — with their referred associations to speed, progress and financial success — provide a good deal of that excitement; meanwhile, the VC-friendly hyperbole of the entrepreneur — tailored for people who want guarantees of fast financial return — is well-suited to the recent heady speed of the election cycle.

Sky journalist Ed Conway described Public Health England’s Excel workarounds as “a little like putting together a car with sellotape”, and that kind of repair is familiar to anyone who has been asked to do more and more with a system that can’t quite cope, and which has not benefited from the minimum necessary investment. The widening gap between narrative and reality means that legacy technologies are often expected to fulfil the promises of an imaginary technical future — so it is not surprising their repair might be hasty, incomplete and not fit for purpose.

Alejandro De Coss Corzo’s paper, “Patchwork: Repair labor and the logic of infrastructure adaptation in Mexico City” explores how “the endurance of urban infrastructure and of urban modernity requires the ad-hoc work of patchwork and of adaptive repair labor”. De Coss Corzo followed engineers working for the Mexico City water authority, SACMEX, and observed how their ingenuity was constantly called upon to compensate for a lack of systemic investment, and how that ingenuity — comprised of “practical knowledge and embodied expertise” — contributed to a constantly evolving system.

“In the messiness of breakdown, embodied expertise allows workers to know where leaks and other problems are and enables them to patchily fix them.”

This is the work of living with technology, of patient innovation, but the focus of the effort is driven by what De Coss Corzo calls “repair politics” — pressure to prioritise certain repairs over others. If what gets measured matters, what gets mended matters even more.

The repair politics of digital public infrastructure is often neglected. This is in part because technology is often used as a proxy for interest in and occupation of the future, while repairing the work of the past is seen as mundane, and probably unachievable within an election cycle. I was on a panel with Marc Warner, CEO of Faculty AI the other day, in which he talked about the cost of not innovating — and this is a familiar refrain. Anxiety for the unknowable, unmissable future creates more heroic propulsion than anxiety about the repairable past.

In their book The Innovation Delusion, Lee Vinsel and Andrew L. Russell talk about how — although innovation is important for both improving economic growth and quality of life — what they call innovation-speak is actually:

“a sales pitch about a future that doesn’t yet exist… [it is] the rhetoric of fear. It plays on our worry that we will be left behind: Our nation will not be able to compete in the global economy; our businesses will be disrupted; our children will fail to find good jobs because they don’t know how to code… . Innovation-speak is a dialect of perpetual worry.”

Current UK technology policy is written in fluent innovation-speak. The ministerial foreword to the National Data Strategy sets out a phoney war in which “data and data use [are] opportunities to be embraced, rather than threats against which to be guarded”. The BEIS R&D Roadmap and the current Covid-testing strategy both speak of “moonshots”, terminology repurposed from the ARPA of the late 1950s, that — as Sharon Weinberger’s Imagineers of War sets out — was as focussed on quelling anxiety about Soviet Russia as it was on creating true scientific gains. Perhaps this fear of the future is meant to stop us all being so scared of the present day, but it also creates impatient Parliamentary consumers and a demand for quick fixes.

But quick fixes are not, as yet, quite the job of our public digital services.

They are, however, a specialism of the US company Palantir, whose recent SEC filing sets out a strategy perfectly suited to filling this narrative gap. It speaks of stale processes, obsolete systems and the glory of perpetual beta, with an average install time of 14 days, compounding FOMO in governments everywhere with the phrase:

“Our most critical institutions cannot wait a year or longer for a promised application or bespoke solution to be developed. Those options are often obsolete before they are even delivered. Our partners need solutions now. And we have built them.”

This is the Field of Dreams at Warp Speed 10, a ready-made, inhabitable future that can be initially bought for as little as £1 and installed in a fortnight. To put it another way, a public body can have an instance of a Palantir tool set-up to deal with critical data more quickly, and much more cheaply, than most people can get a new sofa. Why do the hard work of maintenance when Palantir has already made the future?

Inevitably, this kind of purchasing at speed is also bad news for governance.

The recently updated UK Government Data Ethics framework is by no means equipped to deal with this — it’s a lengthy checklist with no formalised escalation process or commitment to resolving red flags, that ends with this rather leisurely exhortation:

If you have scored a 3 or less in any of the principles, this could indicate the need for additional checks and potential changes to make your project more ethical. Please explain the reason for the score and consult the outcome with your team leader, organisational ethics board or data ethics lead to advise on the specific next steps to improve the ethical standards of your project.

I had a demo of Foundry — one of Palantir’s flagship products, currently being used by NHSX as part of their “single source of truth” — from some of the UK Palantir team earlier this year. This is probably far too simplistic, but from what I can see the platform allows multiple data structured and unstructured data sources to be combined; customers can then generate insights by adding queries and algorithms. Palantir’s website says that Foundry:

enables users with varying technical ability and deep subject matter expertise to work meaningfully with data. With Foundry, anyone can source, connect, and transform data into any shape they desire, then use it to take action.

Without a rigorous culture of information governance, these really are weapons-grade tools. It is probably faster to install Foundry than it is to raise a government data ethics query, and while some governance features are built in, the ability to know how to use them is not automatically installed in the people who will use it. To put it simply: not all information is meant to go together, not everything is related, not all patterns are meaningful — and some are outright dangerous.

Through the myriad technology SNAFUs of 2020, I am not sure the UK government has shown it is qualified to operate in this data-driven future. An administration that lacks the care and attention required to update to a more recent version of Excel — to repair an existing critical system — needs to demonstrate it has the necessary digital maturity and understanding to use the kinds of easy-to-use, hard-to-repent tools that Palantir has to offer.

Unfortunately, that probably won’t stop the announcement of a shiny new initiative. After all, why should a failure with an XLS file get in the way of a “quantitative revolution”?

Feminist. Responsible technologist. Reading and writing on equality, automation and climate crisis. On sabbatical-ish. Formerly @doteveryone .