Hackney / Xantura Early Help Profiling

The Early Help Profiling System was software built by Xantura, an analytics firm, and run for the London Borough of Hackney until 2019. It matched data from across council services to flag the families it judged most at risk. Each month it sent social workers written summaries on those families, not a numeric score.

What the system did

The system matched records from across the council's services. They included social-care referrals, school attendance and exclusions, youth offending, reports of domestic abuse and antisocial behaviour, truancy, and benefits status. Its aim was to flag families for early help, meaning support offered before problems grow serious.

Inside, it gave each family a risk score. Each month social workers received a short list of the families it judged most at risk, about 20 at a time. The list was written as summaries explaining why each family was flagged. The social worker decided whether to offer support.

What the families were told

The families were not told directly that their data was used this way. Notice came only through a general online privacy notice. The privacy impact assessment recorded no option to opt out. The council argued that disclosing details of the system could prejudice possible interventions.

The council withheld the method as commercially sensitive. It refused requests for the data-sharing agreements. No independent evaluation of the system's accuracy or harm was ever published. So the council could not fully say what it had bought.

How it was bought

By October 2018 Hackney had paid Xantura £361,400 since 2015, perhaps not all for this system. It procured the system through London Ventures, a programme run by London Councils and the accounting firm Ernst & Young. London Councils represents London's 32 borough councils and the City of London. £91,400 came through London Ventures.

How it ended

The council dropped the pilot in 2019. It reported that the available data was more limited and more variable than expected. The system had flagged about 350 families, but it surfaced 7 children the council did not already know about. The council judged that it could not deliver enough new insight to justify further investment.

This was a rare kind of ending. There was no public scandal. The council made a cost and benefit judgment that the system did not work well enough to keep.

Why a small pilot matters

Hackney matters because it is small. Most AI in public institutions is not a statewide system. It is a pilot bought from a vendor by a stretched team. At that scale, buying the system is the main point of control, monitoring is informal, and quiet failure is the likely end.

A pilot that cannot say what data it uses, or how families would ever learn of it, has already failed a governance test. That holds however accurate it is. What finally stopped this one was not the model. It was the records beneath it.

So the useful tools here come before any change to the model. They are purchase terms with teeth, keeping less data, and someone with the authority to ask who approved profiling families without telling them.

What this network is drawn from

This network follows the pattern the case file describes. It is not a reconstruction of the actual system. It shows the profiling system, the council data it matched, the council's privacy and information governance, the social workers, and the family profiles.

What the available tools can and cannot address

A failure pathway is a link between two parts of the network where a mistake made by one part can be passed on to the other. Closing a pathway means mistakes stop passing along it. The work along it goes on.

This case has a budget of 12 units. Each tool costs the same at every target level, except Understand the system, which pays for ongoing study of the deployment. It costs 3 units under Explore (No Targets) and Service Targets Only, and 4 under the two higher levels. While it is on, Store less data, Vet connections, Review on schedule, and Pause AI on alarms each cost 1 unit less. With its stronger setting, Deep research, they cost 2 less, but never less than 1.

Before any tool is used, four failure pathways are open. They are Council data into the profiles, Monthly report to social workers, Family history read by workers, and Records scored for next report. The network is at a tipping point, where mistakes could start building on one another. It is not self-correcting, meaning it does not clear mistakes on its own. The service is strained.

Explore (No Targets) sets no targets. Under Service Targets Only, the targets are not met before any tool is used. That level asks for the network to be self-correcting and for the system to be helping the work. Escalate checks meets them on its own for 2 units, and so does Mark AI-written records. Assign a challenger at its stronger setting meets them alone for 3. With lingering effects off, Vet connections does too, for 3. Lingering effects is a setting in which damage outlasts its cause, on by default. In all, more than 600 different sets of tools within the budget meet them.

Under Service and Safety Targets and All Governance Targets, the targets can be met. Both levels ask you to close every failure pathway and keep the service from being strained, among other targets. The cheapest way is Escalate checks with Mark AI-written records, for 4 units.

Escalate checks has social workers check the monthly reports more closely when monitoring flags trouble. It closes Monthly report to social workers. Every set of tools that meets these two levels' targets includes it. Pause AI on alarms also closes that pathway, but it takes away so much of the system's benefit that no set including it meets the targets.

Mark AI-written records closes the other three pathways. In the Lab it marks machine-written content in the records so readers can weigh it. Here it stands for marking which entries the profiling system wrote, and where each record came from.

Vet connections with Understand the system closes the same three instead. With Escalate checks, that way costs 8 units under these two levels. More than 100 different sets of tools within the budget meet the targets at each of the two levels.

More is not better here. Using every tool on offer, each at its strongest setting, costs 35 units, nearly three times the budget. It meets the targets at none of the three levels that set them, because the system then adds too little to the work.