From the Health & Human Services edition of October 31, 2026
UK StreetLink AI prompts raised the share of rough sleepers found after public alerts from 16% to 23%, now rolled out nationally
UK Ministry of Housing, Communities and Local Government (Homelessness and Rough Sleeping Directorate, Local AI team) with Home Connections · Homelessness and rough-sleeping outreach services · England and Wales, United Kingdom
- Publisher
- MHCLG Digital blog (GOV.UK)
- Original publication
- 2026-10-30
- Source retrieved
- 2026-11-01
What happened
A UK Ministry of Housing, Communities and Local Government (MHCLG) blog post (Oct 30) reports a pilot with delivery partner Home Connections and 10 local authorities that added AI prompts to StreetLink, the public service for reporting people sleeping rough. As people filled in an alert, the system prompted for landmarks, access details and descriptions. Reported results: location descriptions up 45% and person descriptions up 54%; 77.6% of pilot alerts rated average or good; failed outreach attempts caused by poor descriptive data fell from 36% to 18%; the share of people found after an alert rose from 16% to 23%; nearly 89% of users said the guidance helped. All participating councils supported rollout, which is now live across England and Wales. AI-assisted alert quality scoring and prioritisation were explored, with outreach staff judgement kept.
Why it matters
U.S. city and county homeless services, outreach teams and 311 portals take public referrals with the same weak-data problem. The case shows AI used to improve the information coming in, not to decide who gets help. Applicability to U.S. systems is an interpretation.
Evidence and measured results
Before/after operational metrics reported by the program owner (alert content, quality ratings, failed attempts, find rate, user survey). No method section, sample size or independent evaluation.
Limitations and uncertainty
Self-reported by the funding department and delivery partner in a blog post. No sample sizes, time periods, baseline definitions or comparison group are given, and the pilot lasted several weeks. The model or vendor is not named. UK homelessness context; transfer to U.S. continuum-of-care and 311-style outreach referrals is an interpretation.
Put this evidence to work
Lighthouse Advisory interpretation, grounded in this source. Enriched 2026-10-31; this does not change the original publication date. Each role follows the same checklist so takeaways compare across sources.
Sales
Role takeaway
- Customer problem
- Outreach teams waste visits on public referrals with vague locations and descriptions.
- Stakeholders
- City or county homeless services director
- Outreach program managers
- 311 or digital services lead
- Continuum of Care lead agency
- Discovery questions
- What share of outreach attempts fail to locate the person, and why?
- How are public referrals collected today (phone, web, app, 311)?
- Do you rate referral quality before dispatching outreach?
- Value hypothesis
- Bounded hypothesis: guided prompts in the referral form raise the share of actionable alerts and reduce failed outreach visits.
- Potential engagement
- A short referral-quality assessment and prompt design pilot on one intake channel with a measured baseline.
- Unsupported claims
- That U.S. agencies will see the same 16% to 23% change
- That AI should prioritise which people receive outreach
- That the results were independently evaluated
Pre-sales engineering
Role takeaway
- Fit
- Fits web or app intake forms for public referrals; low-risk assistive use at data entry.
- Architecture and integration
- Prompting layer on the referral form, completeness scoring service, and hand-off of structured fields to the outreach case or dispatch system.
- Prerequisites
- Baseline failed-contact data
- Access to the intake form code or platform
- Agreed referral data fields
- Constraints
- Source does not name the model or vendor
- U.S. data-sharing rules for homelessness data (such as HMIS policies) differ
- Phone referrals may not benefit
- Security
- Treat descriptions and locations of people sleeping rough as sensitive; limit retention and access, and keep model providers from training on alerts.
- Proof of value
- Compare failed-contact rate before and after on matched areas
- Audit a sample of prompted alerts for inappropriate detail
- Check quality scores against outreach staff ratings
Delivery
Role takeaway
- Work
- Baseline referral outcomes, design and test prompts, roll out by area, and review outreach feedback regularly.
- Dependencies
- Outreach teams
- Portal or 311 owner
- Data protection review
- Ownership
- City or county homeless services or outreach program.
- Skills and adoption
- Outreach staff report on referral usefulness; portal team iterates prompts.
- Governance checkpoints
- Privacy review of collected fields
- Decision on whether any scoring affects prioritisation
- Safeguarding review
- Acceptance criteria
- Measured drop in failed outreach attempts versus baseline (proposed)
- Increase in successful contacts per alert (proposed)
- No rise in safeguarding incidents (proposed)
- Risks
- Short pilot effects may fade
- Over-collection of personal detail
- Scoring drifting into rationing outreach
Implementation considerations
Lighthouse Advisory interpretation across the operating dimensions a public-sector buyer must settle before this evidence becomes a design. Each note answers the question under its heading for this specific source.
Architecture and integration
What must connect, and where does the AI sit in the workflow?
Inline guidance at the point of entry in a public form, plus quality scoring of submitted alerts behind the scenes. Structured location prompts (including What3Words) feed outreach workflows.
Governance
Who approves, reviews and stays accountable for outcomes?
The team names safeguarding, alert prioritisation and AI quality scoring as areas still being refined. Any prioritisation use needs documented oversight because it affects who is visited first.
Security and privacy
What data, permissions and controls need testing?
Alerts contain descriptions and locations of vulnerable people supplied by the public; prompts that ask for more detail increase the sensitivity of stored data.
Accessibility and workforce
Who is affected, and what skills or accommodations follow?
Prompts were designed to keep reporting quick; outreach staff received more actionable referrals. Accessibility testing results are not reported.
Procurement
What should contracts, pricing and exit terms secure?
The work was funded and supported by a central government AI team with an existing delivery partner rather than a new procurement.
Operating model
Which teams own the service once it runs?
Iterative pilot with regular council feedback, then national rollout; outreach teams keep the decision on response.
Publication history
- 2026-10-31Health & Human Services · Issue 033 resources
Stable resource ID: uk-mhclg-streetlink-ai-alert-guidance-pilot-national-rollout-20261030