National, regional, and local government

AI worker teams for citizen service navigation

Citizens do not know your org chart, and they should not have to. The service failure here is navigational rather than informational — and navigation is precisely what can be automated without touching a single determination.

NATIONAL, REGIONAL, AND LOCAL GOVERNMENTCONSTRAINT ENVELOPE4 sector constraints designed to firstHUMAN AUTHORITY — NEVER AUTOMATED· Every statutory and discretio…· Safeguarding and vulnerability· Enforcement and sanctionACTS WITH APPROVALThresholds and gates on anything with cost or risk attachedACTS WITHIN POLICY3 blueprints start hereAutonomy widens outward only on evidence from the recordearned autonomy
The authority boundary we design to in Government & public sector — autonomy widens outward from the centre, and only on evidence.

Our view

Where we think the real opportunity is

Public-sector service failure is usually described as a resourcing problem. We think a large part of it is a translation problem. A resident describes a situation — a leaking pipe, a missed collection, a benefit they think exists — and the system requires them to know which service, department, and form corresponds to it. The transfers, repetition, and incomplete forms all follow from that mismatch. A conversational layer is unusually well fitted to it, because translation from lay description to correct service is exactly what it does, and it requires no authority whatsoever. That is what makes this the sector where the human boundary is cleanest: every statutory and discretionary determination stays with an official, and the automation never comes near one. The second design commitment is reach — voice and multiple languages widen access rather than narrowing it, which matters when digital exclusion is a live equity concern.

Operating reality

How demand actually behaves here

Volume has a shape, and the shape is what breaks teams. This is what we assume about your operation before we design anything.

  • Residents describe problems, not services

    The mismatch between how a citizen frames their situation and how services are catalogued is the root cause of most transfers.

  • Transfer-driven repetition is the norm

    Each hop restarts the conversation, so a single request is explained three or four times before it reaches someone able to act.

  • Statutory duties shape everything

    Timescales, accessibility obligations, and equality duties are legal requirements rather than service targets, and they constrain the design directly.

  • Everything is potentially subject to scrutiny

    Freedom-of-information requests, audit, and public accountability mean the record of what was said and done is a first-class output.

Where the hours go

The cost we would try to move first

Almost every operation spends its most expensive time on its least valuable work. Naming that precisely is what makes a pilot measurable rather than impressive.

  • Transfers and repeat contacts

    Handling the same request several times across departments is duplicated cost that no single team's budget owns.

  • Incomplete requests bouncing between departments

    A request missing required fields cannot be actioned, so it circulates, ages against a statutory clock, and generates chase contacts.

  • Status-chasing contacts

    A large share of inbound volume is residents asking what happened to a request, which is demand created by the absence of proactive updates.

  • Channel duplication

    Phone, web, and in-person routes maintained separately with different information produces inconsistency and triples the maintenance burden.

The trust envelope

What constrains the design, and what never gets automated

We design to the boundary first. A control added after launch is a control nobody believes — and in this sector the boundary is not negotiable.

Sector constraints

  • Guidance is never a determination

    Eligibility information must be explicitly framed as informational, with no implication that a decision on the resident's case has been made.

  • Data minimisation

    Only fields the specific service requires may be collected, and the workflow enforces it rather than relying on the resident volunteering less.

  • Accessibility and language duties

    Accessible channels and additional languages are obligations, and they should be offered proactively rather than on request.

  • Public-record auditability

    Interactions and access must be recorded in a form that withstands audit, scrutiny, and information requests.

Stays human, permanently

These are structural boundaries, not trust levels waiting to be relaxed.

  • Every statutory and discretionary determination

    Decisions about entitlement, enforcement, and discretion are exercises of public authority and remain with officials, without exception.

  • Safeguarding and vulnerability

    Any disclosure of harm or risk routes to a trained officer immediately, leaving the automated path entirely.

  • Enforcement and sanction

    Actions with legal consequences for a resident require an accountable public servant.

Where to start

Sequencing matters more than scope

The first blueprint should be the one whose success criteria your team already agrees on. Each step below links to the blueprint that implements it.

  1. Start

    Navigation and published guidance only

    Help residents identify the right service and understand requirements. No submissions, no status lookups. This proves navigation accuracy on real language, which is the foundation for everything after it.

    Read the blueprint
  2. Then

    Complete submission for two or three services

    Enable submission for a small number of high-volume services under data-minimisation rules, with caseworker review on every request.

    Read the blueprint
  3. Next

    Status transparency and wider reach

    Add proactive status updates to remove chase contacts, then extend accessible channels and languages based on measured resident demand rather than assumption.

    Read the blueprint

Systems we would expect to read and write

  • Case management system
  • Service catalogue and content platform
  • Identity and verification services
  • Records management and FOI tooling
  • Interpreting and accessibility services

Blueprints that apply

3 operating blueprints for Government & public sector

Each one carries the full team, process, and control detail — the mechanics this page deliberately does not repeat.

Push back on this

Objections we actually hear in Government & public sector

Answered as we would answer them in the room, including the cases where the right answer is that this is not for you.

An algorithm must not make decisions about citizens.
We agree without qualification, and the blueprint is built so it cannot. Determinations — statutory and discretionary alike — stay with officials, and the workers have no authority to assess a case. What they do is translate a resident's description into the right service, explain published requirements as guidance, collect the minimum required information, and report status. None of that is a decision, and the boundary is structural rather than a setting.
This will exclude residents who are not digitally confident.
That concern usually assumes a chatbot on a website. Voice is the flagship channel here precisely because it requires no device literacy, no app, and no portal login — and multiple languages are available without depending on which staff are on shift. The measurable claim is that reach widens; we instrument channel and language usage so you can verify it rather than take it on faith.
How would this survive an audit or an FOI request?
Better than the current position, in our view. Every interaction, access, and action is recorded in a tamper-evident trail, which means what was said to a resident and what was done with their data is retrievable rather than reconstructed from call notes. Data minimisation is enforced in the workflow, so the record also shows what was deliberately not collected.

Other sectors

How the argument changes elsewhere

Next step

Argue with this in a working session.

Bring your own numbers. We will map your version of the operating reality, agree what must stay human, and scope the first blueprint against a measurable win — or tell you it is not worth doing yet.