Jump to section
Overview
In early 2025, Norfolk County Council needed to deliver emergency financial support quickly to residents experiencing sudden financial shock, using the Household Support Fund (HSF).
The HSF ran from October 2021 to March 2026 before being replaced by the Crisis and Resilience Fund. Supported by the UK government and managed by councils, it helps residents with high living costs who apply using a form and bank statements.
“The focus was on getting people food and fuel quickly, without judgement, and without creating a big, expensive system.”
Funding was confirmed in February for an April start, leaving six weeks for setup. With unpredictable demand, the council aimed to direct most funds to residents rather than administration.
To address this, the projects and digitisation team in the finance directorate created an automated artificial intelligence (AI) process using existing Microsoft tools, speeding up application handling and promoting fairness without buying new systems, keeping costs low and implementation quick.
Challenges
- Extremely tight timescales: Funding confirmation came in February for a service starting in April, leaving around six weeks to design and launch.
- Highly variable demand: Application numbers could spike without warning, making staffing to handle the demand difficult to plan. Schools also promoted the offer, including on social media, which increased demand.
- Governance and confidence: This was the first operational use of AI within Power Automate at the council, so the team needed clear governance and reassurance around fairness and decision-making.
- Data sensitivity: Bank statements and personal data, submitted as part of the HSF application process, needed to be handled securely and in line with information governance requirements.
Approach
Rationale for using AI
AI was used to extract consistent factual information from unstructured bank statements at scale, addressing the main bottleneck of manual checks for basic details such as name, address, and closing balance. This enabled applications to be triaged in seconds and processed consistently without recruiting additional staff. All possible rejection decisions and accountability remained with council staff, with AI deliberately constrained to factual extraction only.
Which AI solution was used?
The projects and digitisation team designed an end‑to‑end process using Microsoft Forms, SharePoint Lists and Power Automate, with AI restricted to tightly defined tasks. Using existing Microsoft tools and rule‑based criteria kept costs low, strengthened security, and enabled faster processing with human oversight.
Implementation
- Applications were submitted by residents via Microsoft Forms.
- Supporting bank statements were sent by email to ensure virus scanning before processing.
- AI within Power Automate was used only to extract specific fields from bank statements (name, address, and closing balance).
- Extracted data fed into a rules-based workflow that applied predefined eligibility criteria consistently. AI did not interpret spending behaviour, assess affordability, detect fraud, or make rejection decisions.
- Any uncertainty, low‑confidence extraction, or non‑standard document was automatically routed to a human reviewer rather than being rejected.
- Eligibility rules, including residency checks, household circumstances, and maximum award thresholds, were agreed in advance by the service team and applied consistently across all applications.
- Rather than a long testing phase, the service launched softly, with close monitoring and rapid improvements made during the first week before wider promotion.
Governance, assurance and transparency
Oversight of the use of AI came from Information Governance and the council’s AI governance board, with design changes made in response to their feedback.
The residents’ application form stated that AI was used to extract limited factual information from bank statements and was not the final decision‑maker. The council’s privacy notice explained data use and retention.
Data handling and retention
Bank statements were received via a controlled mailbox to ensure virus scanning before processing. Documents and extracted data were handled entirely within the council’s Microsoft environment.
Retention periods were defined with Information Governance oversight. Bank statements were retained for a shorter period than application records, in line with operational and audit requirements.
Human review and business‑as‑usual arrangements
The projects and digitisation team supported and maintained the solution as part of business-as-usual operations. Applications that did not meet the automated eligibility criteria, or where data extraction confidence was low, were routed to staff for manual review, reducing the risk of unfair or incorrect decisions.
Where all automated checks were met, including those based on AI-extracted data, awards were issued automatically. Human oversight was maintained through spot checks of automated awards to ensure the system was continuing to perform as intended. As a further control, award outputs were reviewed before being sent to partner organisations, allowing any unexpected variations, such as unusually high total award values, to be identified before vouchers were issued.
Clear fallback arrangements were in place throughout. The team could revert to manual processing at any point if required, including in the event of system failure. Data captured from the initial application was structured and retained within the council’s systems, ensuring a clean dataset was available as a backup. Each stage of the workflow could be completed manually if necessary.
“AI does the factual checks, not the judgement. That was really important for trust.”
Outcomes, benefits and risks
Return on investment and operational impact
The automation delivered a significant reduction in processing time. Some applications were processed automatically in under a minute, with vouchers issued within 24 hours.
The approach enabled the team to scale without increasing staffing. In one peak week, four members of staff processed 1,664 applications. Based on historic manual processing rates, this volume would have required up to 32 full-time equivalent staff to manage within required timescales.
Faster processing meant residents could access food or fuel at the point they needed it most, rather than waiting days or weeks for support.
Benefits
- Fairer and more consistent decision‑making: Removing subjective judgement from bank statement reviews reduced the risk of bias in crisis decisions.
- Staff capacity protected: Automation freed staff to focus on higher‑value support, such as referrals to welfare rights, debt advice, and assistance with essential household items.
- Reuse and flexibility: The same model was quickly adapted to support other schemes, such as school uniform assistance, with minimal changes and no additional recruitment or new systems.
Risks and mitigations
Lessons learnt
- Start simple: A clear minimum viable product helped meet deadlines and reduced risk.
- Be explicit about AI’s role: Clear communication internally and with applicants helped build trust and confidence.
- Engage governance early: Early involvement of information governance and AI oversight avoided delays later.
- Human oversight matters: Human review for edge cases (such as shared housing) was essential to avoid unintended exclusions.
- Use automation to support fairness: Applying consistent factual checks helped keep crisis support focused on immediate need.
In future iterations, the team would invest earlier in staff engagement and allow more time for confidence-building alongside delivery.
What’s next?
Building on this work, the council will continue to use the system to deliver the Crisis Resilience Service. The service brings together emergency food, fuel and heating oil support with AI enabled automation, alongside Power Apps case handling for more complex needs such as budgeting support, household goods and debt advice, all accessed through a single application form.
How to replicate in your organisation?
This approach is designed to be replicable for councils that already use Microsoft 365. Key prerequisites are:
- Skills and capacity: A service owner, a Power Automate builder, and an information governance lead; plus, access to operational staff for rules definition and user testing.
- Technology and licences: Microsoft Forms, SharePoint Lists and Power Automate, plus access to an approved AI capability within the organisation’s Microsoft environment (and an inbox/secure storage location for attachments).
- Governance and assurance: Agreed eligibility rules, a DPIA (or equivalent), retention/handling rules for bank statements, prompt/configuration change control, and a human-in-the-loop process for exceptions and sampling.
Typical timeline: 1–2 weeks to define rules and data handling; 1–2 weeks to build the flow and dashboards; 1 week for soft launch, monitoring and iteration.
Useful links
Visit the following resources, also produced in collaboration with Norfolk County Council:
