Futura AI
it
All articles
  • Public Sector
  • Method

Five public-sector tasks you can automate today

You don't need multi-year projects for a measurable effect: repetitive, documented, high-volume processes are enough. Five concrete tasks for public agencies, with prerequisites and indicators for each.

by Daniele Grotti5 min readUpdated on
Futura AI — Five public-sector tasks you can automate today

You don’t need multi-year projects to get a measurable effect. You need processes that are repetitive, documented and high-volume.

The three conditions matter together. A repetitive but rare process doesn’t justify the setup cost. A frequent but undocumented process needs to be defined first. A process that’s defined and frequent but governed by rules that keep changing requires maintenance that cancels out the benefit.

When the three conditions line up, the path is short and the result is measurable within a fiscal year.

1. Sorting and classifying correspondence

Every incoming communication — certified email, institutional email, registered mail — needs to be read, classified by type and assigned to the right office.

What the system does. It recognizes the type, identifies the sender, extracts the subject and references, and proposes the assignment and classification according to the filing scheme. In low-confidence cases it doesn’t force the choice: it routes to human review.

Prerequisites. An up-to-date classification scheme applied consistently. Read access to the mailbox and write access to the records-management system. A history of already-classified communications, needed to configure and verify the system.

Indicator. Share of communications assigned to the correct office on first routing, measured on a manually verified sample, and average time between receipt and assignment.

2. Pre-filling recurring resolutions and administrative acts

A significant share of an agency’s acts have a stable structure: spending resolutions, payment orders, serial provisions. What changes are the references, the amounts, the parties.

What the system does. It populates the act’s template with the correct data pulled from the source systems, inserts the regulatory references and references to prior acts, flags fields that need evaluation, and reports inconsistencies found.

Prerequisites. Formalized, up-to-date act templates. Access to accounting and case-management data. A clear rule on who approves changes to the templates, because a misaligned template propagates the error to every act generated from it.

Indicator. Time to prepare the draft and number of substantive corrections made before signature. The second is more informative: if it stays high, the template or the source data needs review.

3. Answering citizens’ standard requests

Some of the requests an agency receives concern information that’s already published: the requirements for a procedure, the documentation needed, the status of a case, forms.

What the system does. It recognizes the nature of the request, retrieves the information from the agency’s official sources and drafts a response with references. Where the request concerns a specific case, it pulls the status from the case-management system.

Prerequisites. Published information that’s up to date and consistent, because the system will reflect it exactly as it is. A clear boundary between information and evaluation: the system doesn’t interpret whether a regulation applies to a specific case.

Indicator. Average response time and share of requests resolved without being forwarded to another office.

One clarification: here, human confirmation before sending is warranted in every case, since this is external communication.

4. Searching acts for internal offices

Resolutions, decisions, circulars and regulations accumulate over the years. Finding a relevant precedent today takes personal memory or patience.

What the system does. It lets staff query the archive in natural language, retrieving relevant acts by meaning rather than literal match, with a citation of the reference and date.

Prerequisites. Acts available in text format or digitizable with sufficient quality. Reliable metadata on date, type and current status. Permission management aligned with the sources.

Indicator. Time to locate a precedent, sampled on real requests before and after introduction.

Of the five, this is the one with the lightest prerequisites and the broadest benefit, because it applies to every office.

5. Internal support for staff on regulations and procedures

Questions about how to handle a particular case are currently directed to experienced colleagues. They’re a scarce resource, and in many agencies, close to retirement.

What the system does. It answers questions on internal regulations, procedures and circulars, always citing the source, and states it has no basis when the answer isn’t documented.

Prerequisites. A written body of procedures. This is the most demanding prerequisite: where practice hasn’t been formalized, the first job is collecting it, and that takes the time of experienced people.

Indicator. Number of support requests directed to reference offices and time to autonomy for newly onboarded staff.

What’s not on this list

Three exclusions, for clarity.

Administrative decisions. The system prepares the case, the official decides and signs. None of the five tasks touches this boundary.

Procedures with low repetition or high discretion. The setup cost isn’t justified and the risk of error is higher.

Systems that determine access to essential services or directly affect rights. These fall under a different classification under the AI Act and require a dedicated assessment, to be carried out with legal counsel before any technical choice.

One further limit applies across the board: none of the five tasks produces results on ungoverned archives. If the versions of the acts aren’t certain, the system will correctly cite the wrong source.

One thing all five have in common

All of them require defining, before launch, what happens when the system gets it wrong.

This isn’t a formality. A misrouted assignment needs to be corrected, and the correction has to feed back into monitoring, or the error repeats. A draft act with a wrong reference has to be caught before signature, and that means the review has to be practical: with the data linked to its source, in a time compatible with the office’s real workload.

A channel is also needed through which operators report errors, with the guarantee that reports lead to changes. If a reported error isn’t fixed, or the fix isn’t communicated, the channel closes within a few weeks, and with it the only reliable source of information about how the system actually behaves.

These elements don’t depend on the technology and need to be established together with the office manager, not after rollout.

In closing

The criterion for choosing among the five isn’t the expected benefit but the availability of the prerequisites. It’s worth starting with the one whose prerequisites are already met, measuring, and using the result to take on the next one.

In most of the agencies we talk to, the fourth — searching acts — is the most practicable starting point.

If you have a specific process in mind, we’re available for a conversation about that case: which prerequisites are already in place, what would be missing, and what indicator would be measurable.

If this topic touches a real process in your organization, let's talk about it with a focused AI Assessment.

Request an AI Assessment