Research · Grant compliance management

Consolidating scattered requirements in research projects

A research institution manages around 1,200 projects a year with complex requirements. Scattered funding notices led to missed deadlines.

Science
Data analytics

Initial problem

Requirements and deadlines scattered across projects

Solution type

Agentic workflow for requirements

Complexity

Transformation

Key metric

Reduce missed requirement deadlines

User group

Research administration and project leads

Implementation time

3-6 months

Operating model

On premises or private cloud

Starting point

A non-university research institution with around 600 employees oversees roughly 1,200 ongoing projects each year. Around 300 new projects are added annually, many of them subject to requirements from funding notices, ethics approvals or internal committees. Today, the administration documents these requirements in spreadsheets, emails and notes spread across different folder structures. Each project carries an average of five relevant requirements, such as reporting obligations, interim acceptances or documentation evidence. Three employees in research administration together spend around 1,200 hours a year recording requirements, tracking them and resolving queries.

1,200: ongoing projects per year
5: requirements per project on average

The challenge

Research administration has to ensure that requirements from funding notices, ethics approvals and internal committees are met on time. In practice, notices sit in different repositories as PDFs, emails and minutes. Project leads often pass requirements on in their own words, making it hard for the administration to tell which obligations have already been fulfilled. Around 100 deadlines are checked manually each month, often only shortly before they expire. Queries to project leads and specialist departments take time because context is missing and documents have to be tracked down. Missed deadlines lead to follow-up demands from funders and additional reporting effort.

“We often only notice shortly before the deadline that a requirement is still open”

“I spend too much time searching notices and minutes for specific wording”

“We can't reliably tell which requirement belongs to which document”

“When covering for colleagues, we lose track of which deadlines are really critical”

Why this remained unsolved

For a long time, the institution accepted the problem as an unavoidable consequence of complex funding conditions. A dedicated IT project seemed costly because requirements come from very different sources and change frequently. Spreadsheet solutions and calendar reminders helped only to a limited extent, as they had to be maintained manually and had no connection to the original documents. For IT, a classic workflow system was hard to justify because the requirements were difficult to formalize and free text from notices did not seem reliably expressible as rules.

Deadline calendar

A central deadline calendar was introduced but remained incomplete because entries had to be maintained manually.

Spreadsheet templates

Standard spreadsheets for requirements were provided but used inconsistently and rarely linked to the original notices.

An agentic workflow was introduced that captures requirements from notices, minutes and emails in a structured way, assigns them to projects and tracks them across their entire lifecycle. Incoming documents are stored centrally, analyzed automatically and turned into requirement entries with a deadline, responsible person and type of evidence. Research administration reviews and completes these suggestions and releases them to the deadline register. Project leads get a consolidated overview of their requirements and can upload or link evidence of fulfillment directly. The workflow sends reminders about upcoming deadlines, bundles queries and documents decisions in a traceable way.

Where generative AI comes in

Generative AI analyzes notices, minutes and emails, identifies the requirements, deadlines and conditions they contain and creates structured requirement entries with a short description. It suggests matching projects and responsible people and condenses complex passages into wording suitable for administration, which the administration then reviews and approves.

Building blocks

Central intake of incoming notices, minutes and relevant emails with automatic assignment to projects and funding lines.
Document analysis to derive structured requirement entries with deadlines, responsibilities and required evidence.
Agentic workflow that monitors deadlines, triggers reminders and links evidence of fulfillment to the requirements.

Current-state analysis
Survey of existing repositories, deadline lists and communication channels in research administration, and selection of a pilot area.
Data integration
Connecting the document management system, the project administration system and the central email inboxes to the new workflow.
Requirements model
Joint definition of categories, fields and status values for requirements, including roles and responsibilities.
Pilot operation
Gradual deployment in the pilot area with parallel manual checks, rule adjustments and staff training.
Rollout
Expansion to further departments, fine-tuning of the reminder logic and establishment of governance processes.

Metric	Value	Basis
Missed requirement deadlinesKey metric	Reduced by around 70 percent	System measurementmeasured
Processing time per requirement	Reduced from 60 to 30 minutes	Time tracking before/afterestimated
Correction rate of AI suggestionsQuality metric	Around 15 percent of suggestions require corrections	Samplemeasured

Effort class

Standard (60-150 person-days)

Payback period

6-12 months

Main lever

Time savings

Ongoing costs

Ongoing costs arise mainly from hosting, maintenance, model updates and occasional adjustments to the requirements model.

Limitation

The solution depends on the quality of document filing. If notices or minutes are kept outside the connected system or filed late, the workflow does not capture requirements completely and deadlines go unnoticed.

Connected systems

Integration with the existing document management system, the project administration system and the central email inboxes of research administration.

Operation

Operated in the institution's private cloud or in the parent organization's data center, with separate environments for testing and production.

Compliance

Access rights follow the existing role models, sensitive project data is processed only within the institution, and logging supports internal and external audits.

Labeling

Interfaces clearly label AI-generated requirement suggestions and show whether an entry has been reviewed or is only a suggestion.

Accountability

AI suggestions are non-binding; legally relevant decisions and deadline approvals remain with research administration.

Escalation path

Unclear or disputed requirements are escalated to designated contacts in legal or project management, including a reference to the document.

Traceability

Every change to requirements, deadlines and status is versioned, including timestamp and responsible person, to secure audit trails.

Operational roles

Research administration is responsible for content and approvals, IT operates the platform, and departments provide feedback on requirement categories and workflows.

Transferability

The solution suits organizations that need to coordinate many requirements and deadlines from external notices and internal committees. Wherever obligations are currently scattered across emails, minutes and spreadsheets, an agentic workflow can structure requirements, monitor deadlines and link evidence. This is especially relevant in regulated environments with recurring audits, where traceability and complete documentation are essential.