The Tool That Fixes One Repeated Task in Public Agencies

Start with work chosen by staff

From February 2026, Korea’s Public Procurement Service (조달청), the government agency responsible for public procurement, ran a dedicated AI development team led by its Director General for Digital Fair Procurement and involving 26 employees. Instead of outsourcing all development, it let staff who understood the work select pain points and take part in building solutions.

Its first results, published on September 11, were seven AI services. They focused on work repeated every day: advice on procurement rules, drafts of press releases and speeches, recurring official documents, assigning and searching materials requested by the National Assembly, searching relationships between laws and regulations, and task reminders.

On October 20, it said staff were using 22 tools through its internal Procurement AI Portal: 10 AI assistants and 12 work applications. The number of tools was close to the 26 participating employees. This looks less like one giant system and more like a set of small tools.

The tools also covered estimated-price preparation, bidder eligibility analysis, integrated contract management, construction unit-price comparison, and duplicate checks for recommended products. They are not public-facing services. They are internal tools used by Public Procurement Service staff on its internal network.

The number of tools matters less than the work they actually reduce. The agency did not publish the number of employees using each tool, time saved by task, error rates, or the share of outputs corrected by staff. Their impact still needs to be measured separately.

Fix one person’s blockage, not a giant system

Consider a fictional employee: Lee Su-jin, an assistant manager on the contracts team at a local public enterprise (지방 공기업) with 70 staff. She receives purchase requests from business departments, then checks eligibility requirements, applicable rules, similar contracts, and official-document formats.

Assume she currently copies a project name and amount from a request into a checklist, a contract-management file, and an official-document draft. She spends time finding earlier contracts, checking the latest wording of a rule, and asking the business department again for missing documents.

The bottleneck is not writing. It is scattered information. Rules sit in one place, past contracts in another, and checklists in personal folders. The less experienced the employee, the harder it is to know where to start.

If the assumption that every agency must first buy a giant AI system for all staff is wrong, the likely winners change. Rather than a provider promising to transform the whole agency, the better position may belong to the team that finds one task Lee repeats each week and connects its inputs, outputs, and final reviewer.

On a changed screen, Lee selects “bidder eligibility review” and uploads a purchase request. The tool shows the requirements to check, relevant regulatory wording and effective dates, similar past contracts, and missing information on one screen. It also prepares a draft official document.

But confirming eligibility or an estimated price remains a human job. Staff and approval authorities must interpret exceptions, reject outdated precedents, and approve decisions that affect companies. The tool must not send a conclusion automatically.

So the opportunity is not AI alone. Useful products also include a tool for finding which work to fix first, an answer inbox that shows supporting documents, a record that captures staff edits and approvers, and a measurement tool that compares time before and after adoption.

Where early government efforts started

Singapore’s Pair (페어) is an AI assistant for public officers that began at a government hackathon in 2023. It provides document drafting and summarisation, plus the ability to create and share tailored assistants.

Its official performance dashboard shows 98,527 monthly users in the second quarter of 2026 and an average saving of 81 minutes per task. These are figures calculated by the Singapore government, but they show that a shared tool with task-specific assistants can scale in a large organisation.

The UK government (영국 정부) tested Microsoft 365 Copilot with 20,000 civil servants across 12 agencies from September to December 2024. It started with work shared across departments, including document drafts, meeting summaries, and email writing.

An official report said users reported saving an average of 26 minutes a day, which it converted to roughly 13 days a year. It also found limits when work involved complex materials and policy context, where human review remained necessary. A general-purpose tool alone does not complete each specific workflow.

Leeds City Council (리즈 시의회) tested Zilo Core (자일로 코어) for the narrow task of reviewing planning-permission applications. It helped check applications, search policies and past cases, and draft reports, while requiring qualified reviewers to review and approve every suggestion.

A Local Government Association (영국 지방정부협회) case study described one reviewer saving about one day a week during a three-month trial. It also recorded what the tool did and the reviewer’s edits and approvals. The case separates fast drafting from accountable decision-making.

Four things to build from here

1. A recurring-work bottleneck map

• What it does: Takes staff interviews and existing forms, then turns repeated data entry, document searches, and approval waits into a one-page work map.
• Who uses it: Local public enterprises and government-funded institutions (출연기관) with only a few contract staff and little capacity to plan a new system themselves.
• Why now: Even when staff can make small tools, choosing what to fix first still requires human observation.
• First screen: Five fields: work name, weekly frequency, places where information is copied, reference documents, and final approver.

2. A regulation answer inbox with evidence

• What it does: Answers a question while showing the exact regulatory wording, effective date, and source location. It routes unknown questions to the responsible staff member.
• Who uses it: Administration teams at public hospitals (공공병원) that repeatedly receive email questions about travel expenses and purchasing rules.
• Why now: Both the Public Procurement Service and overseas public bodies chose regulation search as an early task. Showing evidence alongside an answer better supports human review than generating an answer alone.
• First screen: A question field, followed by the answer, supporting text, effective date, and a button to ask the responsible staff member.

3. A final-human-review record

• What it does: Records each AI output, staff edits, reference documents, final approver, and approval time.
• Who uses it: Digital-transformation staff at public agencies preparing to distribute five or more staff-built AI assistants across several departments.
• Why now: As small tools spread quickly, it becomes more important to explain who confirmed an outcome and on what basis.
• First screen: For each item handled today, it shows the original text, AI output, edits, and approval status in sequence.

4. An adoption-impact measurement kit

• What it does: Compares processing time, number of edits, repeat inquiries, and staff satisfaction before and after adoption using the same criteria.
• Who uses it: Innovation staff at a local government (지방자치단체) deciding whether to continue an AI service after a three-month trial.
• Why now: Tool counts and usage counts alone cannot show whether work became faster or errors fell.
• First screen: Select a workflow, then enter time and edit results for 10 cases before adoption and 10 cases after adoption.

What to check in 30 minutes today

Call one person in contracts or administration. Ask for one task they repeated at least three times in the past week, and the item that always required a final human check. If the same inputs and outputs recur, the reference documents are defined, and they can clearly name a final approver, it is worth testing as a small tool.

Why this matters where you are

Check whether public-sector teams in your market are starting with common drafting tools or with narrow, role-specific workflows. The agency rules, approval structures, and records they must retain will differ. Start by finding one repeated task where evidence, human review, and measurable outcomes can be designed into the tool from the beginning.