Housing development in London has a new weapon: AI

Data from an AI learning model that mapped the London Borough of Lewisham suggests the capital’s targets for ‘small site’ development could be radically increased, says Russell Curtis

When the latest iteration of the London Plan was adopted in 2021, for the first time in its 20-year history the policy demanded that the capital’s 35 planning authorities deliver a proportion of their overall housing targets on small sites – that is, with an area of no more than 0.25ha. The figures varied across the city, but the total number of homes to be found on these pockets of land stood at 120,000 – just under a quarter of the overall housing target for the whole of London.

This number was way below that originally proposed by the London mayor in 2018. Alongside a ‘presumption in favour’ of development on small sites close to public transport, the earlier version of his plan compelled planning authorities to find space on small sites for a quarter of a million homes, with the outer boroughs expected to deliver the lion’s share.

The pushback was inevitable, with London Assembly member Andrew Boff claiming, hyperbolically, that this amounted to a ‘war on the suburbs’. With the GLA lacking convincing data to demonstrate the figures were achievable, the targets were slashed before adoption.

But quiet, in the background, progressive boroughs knuckled down and got on with putting plans in place to promote intensification. In 2020 – nearly a year before the final version of the London Plan was adopted – Lewisham Council appointed RCKa and Ash Sakula to prepare dedicated guidance for new homes on its small sites. Just six months after the London Plan became official policy, Lewisham’s Small Sites SPD was formally adopted.

Two years on, as Sadiq Khan looks towards what’s increasingly likely to be a third and final term, he will be considering updates to the London Plan to cement his legacy as the mayor who did the most to tackle the city’s profound housing crisis. Small sites are likely to be a key focus of this work, and it would be a shrewd move to ramp up the small sites targets accordingly. But the question remains over whether there is sufficient data to justify this increase. To counter resistance to suburban intensification (as happened in Croydon) the new plan will need to be backed with robust evidence of the quantity and distribution of these sites.

So where are they? And how many? We tried to find out.

Land Registry data tells us that there are some 66,000 freeholds in Lewisham, and about 85 per cent of these meet the small-site criteria. Armed with our intimate knowledge of the Lewisham SPD and the site “types” it identifies, we set about mapping every one of them. Having built up a vast library of sites, based on the SPD, we trained an AI to categorise a bunch: backland, infill, amenity space and so on. Then, setting our learning model on the remainder of the borough, we created a complete map of Lewisham, including the location, size – and a rough idea of capacity – of every development opportunity from Deptford to Beckenham.

What we found was striking. While Lewisham’s London Plan 10-year small sites target is currently 3,790, based on early outputs from our data we think there might be capacity for two to three times this number. In fact, our AI model shows that there are enough sites to deliver Lewisham’s target on just two types alone. And as we trawl through the data, the AI improves. Ultimately, our plan is to apply the learning model to capture the whole of London.

Now, just because a site is developable it doesn’t mean it will come forward. The AI makes no distinction between public and private ownership, and many of the sites it has picked out will not provide new homes: some are private gardens, others active builder’s yards and occupied garages. But by establishing a policy landscape that makes planning less risky – as Lewisham has done – boroughs can go a long way to meeting these targets.

Extrapolating these figures across the rest of London, we think there’s sufficient capacity for at least 350,000 homes. Backed by our AI, there can be no more arguing over targets when we know not just how many sites there are. We can even point to them on a map. This is a huge opportunity, and those boroughs still lacking a dedicated small-sites policy should be compelled to implement it as soon as they can. It’s time to take small sites seriously.

This article was originally published in the Architects’ Journal.