Proactive services: when should the state act without being asked?

The OECD lists proactiveness as one of the six dimensions of a digital government, describing a proactive government as one that pre-empts requests, providing answers or solutions through push rather than pull delivery.¹ Estonia’s Information System Authority (RIA) puts the goal in practical terms: bring together the services linked to one life event, so that the complexity of the state stays invisible to the person.²

Few people object to that in principle. The harder question is where to draw the line. A benefit that arrives unasked is welcome; a decision made about you without your knowledge is not. This piece offers four tests. The wider design approach is covered in the guide to life-event services.

One-stop, proactive and no-stop

It helps to separate three levels. A one-stop service gathers everything for an event in one place but still waits for the person to come. A proactive service starts from the state’s side and makes an offer. A no-stop service completes the process with no action from the person at all.

Most of the useful debate is about the step from proactive to no-stop. An offer leaves the person in control; an automatic decision doesn’t.

Test 1: Is the trigger reliable?

Every proactive service starts with an event recorded somewhere: a birth registered, a death registered, a business incorporated. The service is only as good as that record. Where the trigger comes from a well-run register with a clear legal basis, the case for acting is strong. Where it comes from inference, such as a guess that someone has probably moved, it is weak.

This is why proactive services depend so heavily on base registries and data exchange. A wrong entry in an authoritative register now triggers wrong offers automatically.

Test 2: Does the data match the rule?

A register can be accurate and still not answer the question the rule asks. Estonia’s Social Insurance Board makes the point clearly for family benefits: they are paid only if both parents and the child actually live in Estonia, and population register residence data or citizenship on its own is not sufficient.³

The lesson travels. Before automating, check whether the data the service holds is the same thing the entitlement depends on, or only a proxy for it. A proxy is fine for making an offer that someone confirms. It is risky as the basis for a decision nobody checks.

Test 3: Can the person see it and stop it?

Estonia’s approach to personalised event services is that they can be used only after the person gives consent.² That consent step matters less as a legal formality than as a moment of visibility: the person knows the state is acting on their behalf and can say no.

The UK’s Tell Us Once service shows a different version of the same idea. After a death, the registrar explains the service and either completes it with the person or gives them a reference number to use within 28 days, and the service then notifies bodies including HM Revenue and Customs, the Department for Work and Pensions and the local council.⁴ The person chooses to use it, knows who will be told, and can see the limits: it doesn’t cover Northern Ireland, banks or private pension schemes.⁴

Test 4: What happens when it is wrong?

Every proactive service will sometimes be wrong, in two directions. It can act for someone it shouldn’t, or miss someone it should have reached. The second error is quieter and easier to ignore: people missing from a register, or recorded wrongly, may simply never get the offer.

A proactive service passes this test when it has three things: a clear record of what was decided and on what data, a simple way to correct the underlying record, and an ordinary application route for anyone the system missed. Without those, convenience for most people comes at the cost of invisibility for a few.

Where AI fits, and where it doesn’t

It is tempting to extend proactive services with predictions: flagging people who are likely to need support before any event is recorded. That changes the trigger from a fact to an inference, and fails the first test. It also moves the service into territory where AI rules apply. Evaluating eligibility for public assistance benefits is listed as a high-risk use under the EU AI Act, as the guide to AI in public administration explains.

A rule of thumb

Act proactively when the trigger is a recorded fact, the data matches the rule, the person can see what is happening, and mistakes can be put right. Make an offer, rather than a decision, when any of those is in doubt. A confirmation step costs the person a minute and gives them the chance to catch an error before it takes effect.