Your cart is currently empty!
At Fylde, we’ve spent the past few months quietly designing, building, and deploying a new kind of AI solution. One that not only works in real-world council environments, but does so with governance, purpose, and scalability at its core.
Unlike off-the-shelf tools or bolt-on copilots, our approach puts local government control, trust, and explainability at the centre. It doesn’t just answer questions. It understands context, remembers outcomes, works alongside staff, and does all of this transparently, within the boundaries of data protection and compliance.
The Challenge: AI That Works for Councils, Not the Other Way Round
We saw a clear pattern emerging across local authorities. While many had invested in Microsoft 365, Power Platform, or even Copilot licences, few had managed to turn those capabilities into meaningful service transformation. Prompts were being written. AI was being “trialled”. But it wasn’t delivering operational value, and certainly not in a way that frontline teams could trust.
We knew the problem wasn’t the technology. It was the lack of structure: no memory, no orchestration, no governance. Just prompts, bots, and good intentions.
So we set out to build something better.
The Concept: A Modular Agentic Orchestration Layer
Our goal was simple. We wanted to create a system that enables local authorities to deploy real AI agents: specialised digital assistants that work across inboxes, policies, systems, and data, while remaining fully accountable and human-centred.
We called it an Agentic Orchestration Layer. This is a modular AI fabric that can ingest complaints, route requests, draft responses, escalate when necessary, and log every decision made.
But it’s not just agentic in name. It is designed to operate like a real assistant, with context, tools, structure, and clearly defined boundaries.
The Layers That Make It Work
Here’s how we built it:
🔹 Language Model (GPT-4-turbo via OpenAI)
We use GPT-4 for reasoning, summarising, classification, and tone-aware drafting. It is wrapped in structured prompts with role-based context and fallback logic.
🔹 Orchestration Layer (Python + FastAPI or Power Automate)
This controls the flow: receiving emails or form data, passing tasks to the agent, managing memory, and triggering actions like sending responses or saving to SharePoint.
🔹 Input and Output Interfaces (Microsoft 365)
Everything connects into the council’s existing systems: Outlook shared inboxes, SharePoint libraries, Teams approvals, and Power Apps for oversight.
🔹 Memory and Logging (SharePoint + Dataverse)
All decisions and summaries are stored in SharePoint or Dataverse. This ensures they are auditable and available for reporting, including the agent’s reasoning, draft responses, and policy references.
🔹 Governance Controls
Usage thresholds, duplication handling, rate limiting, and escalation logic are built in from the start. Data never leaves the organisation’s tenant unless explicitly permitted. We support both OpenAI API and Azure OpenAI pathways.
Why This Goes Beyond Agentic AI
While agentic AI is often discussed in abstract terms such as multi-agent systems, autonomous workflows, or AI-as-a-colleague, we believe the real breakthrough lies in deployability.
This isn’t theoretical. We have designed agents that:
- Classify and route complaints
- Draft FOI triage summaries
- Read council policies and provide evidence-backed replies
- Operate under strict request caps with human review loops
These agents have purpose, tools, context, and safety nets. They improve with each interaction.
They’re not simply chatbots with better prompts. They are embedded, transparent digital assistants that are built for real local government services.
A Platform, Not a One-Off
While we began with complaints handling, a common pain point for almost every council, we have designed this to scale. The orchestration layer is modular, so new agents can be added easily. Examples include:
- FOI Assistant
- HR Policy Lookup Agent
- Emergency Comms or Continuity Bot
- Planning Enquiry Router
Each follows the same pattern: clarity of role, traceability of decision, and respect for organisational boundaries.
We have also modelled usage using GPT request blocks, allowing councils to scale incrementally, just as they would with compute. Where appropriate, councils can bring their own Azure OpenAI key to maintain control and manage compliance more directly.
What’s Next?
This is more than an AI project. It is a blueprint for how AI can genuinely support local government work in a way that is ethical, transparent, and operational.
We are continuing to pilot, refine, and expand the platform. If you are a council team, tech partner, or supplier interested in building AI that actually works for public services, rather than simply selling the idea of it, we would love to talk.
This isn’t about replacing people. It is about designing digital assistants with accountability, structure, and empathy.
Leave a Reply