In March 2015, something significant happened in the world of government transparency. The Red de Transparencia y Acceso a la Información (RTA) launched mgd.redrta.org, introducing the Modelo de Gestión Documental (MGD)—a comprehensive framework for managing government documents across Latin America.
At the time, it was revolutionary. For the first time, 42 government transparency authorities across 18 countries had access to a unified, practical methodology for tackling one of their biggest challenges: managing millions of documents while ensuring public access, legal compliance, and historical preservation.
Now, ten years later, I’m writing to tell you how that framework has evolved—and why the principles the RTA established in 2015 are more relevant than ever, even as artificial intelligence transforms how we implement them.
March 2015: A Framework for Manual Document Management
Let me take you back to early 2015. If you worked in government records management, archives, or transparency, you probably remember the challenges:
The typical scenario looked like this:
A citizen submits a freedom of information request asking for “all correspondence between the mayor’s office and contractors regarding the new bridge project from 2013-2014.”
Your response process:
Email every department that might have relevant documents
Wait for responses (chase people who don’t respond)
Manually search email systems, network drives, physical files
Read through hundreds of documents to find relevant ones
Review each document page-by-page for information that must be redacted
Manually redact sensitive information
Organize documents logically
Create an index or summary
Deliver response
Time required: 10-20 hours of staff work Average response time: 15-30 days Quality: Depends entirely on who’s doing the searching and how thorough they are
The RTA saw government agencies across Latin America struggling with exactly this problem, multiplied by thousands of requests per year. They needed a systematic approach.
What the RTA Built
The MGD wasn’t just a document—it was a complete methodology covering:
1. Classification Structure (Cuadro de Clasificación)
Instead of organizing documents by department or date (which creates chaos when organizations restructure), the framework organized by function and activity. A contract about park maintenance belongs in “Public Works → Parks → Maintenance Contracts” regardless of who signed it or when.
This seems obvious now, but in 2015, many agencies were still using folder structures like “Director’s Office → 2014 → Random Documents.”
2. Retention Schedules (Calendario de Conservación)
Different documents have different lifespans based on legal, administrative, and historical value:
Tax records: 10 years
Personnel files: 75 years
Meeting minutes: Permanent
Routine correspondence: 3 years
The framework helped agencies develop retention schedules ensuring they kept what they needed (legal compliance, historical preservation) without drowning in documents they didn’t.
3. Metadata Standards (Normas de Descripción Archivística)
The framework established what information to capture about each document:
Who created it
When
What it’s about
Who it involves
What it relates to
Where the original is stored
Access restrictions (if any)
Proper metadata makes documents findable. The RTA understood that a document you can’t find might as well not exist.
4. Implementation Guides (Guías de Implementación)
Theory is nice. Implementation is hard. The RTA provided step-by-step guides for actually deploying these systems:
How to assess your current situation
How to design classification schemes
How to train staff
How to handle the transition
How to maintain the system
These guides were practical because they were built by people who had actually implemented these systems in real government agencies.
5. Best Practice Guidelines (Directrices)
The framework included specific guidance on:
Archival description standards
Active transparency (proactive disclosure)
Open data publishing
Access request handling
Security and privacy
Electronic document management
Why It Worked
The MGD succeeded where many frameworks fail because:
- It was comprehensive but practical – Covered everything but focused on what agencies could actually implement
- It was built by practitioners – Not academics theorizing but professionals who’d done this work
- It acknowledged constraints – Designed for agencies with limited budgets and staff
- It was collaborative – 42 agencies contributing knowledge, not one entity dictating solutions
- It was documented thoroughly – Clear guides, templates, examples, case studies
Between 2015 and 2025, hundreds of government agencies across Latin America implemented parts or all of this framework. It genuinely improved how they managed information.
The Problem: Manual Implementation Was Labor-Intensive
Here’s what the RTA framework required in practice:
Document Classification: Someone had to read each document and decide where it belonged in the classification structure. For a small agency creating 50,000 documents per year, that’s manageable. For a large agency creating 2 million documents per year? You need dedicated staff doing nothing but classification.
Metadata Creation: Someone had to fill out metadata forms for each document—typing names, dates, subjects, relationships. This is tedious, time-consuming work. And humans get tired, make mistakes, skip fields.
Retention Schedule Application: Someone had to look at each document, determine what retention rules apply, calculate destruction dates, flag exceptions. Then actually track those dates and execute destruction when appropriate.
Search and Retrieval: When someone needed a document, they searched metadata or browsed classification structures. If metadata was incomplete or wrong, documents were unfindable. If the searcher didn’t know the right classification terms, they couldn’t find relevant documents.
Access Request Response: For each information request, staff had to search manually, review documents manually, identify sensitive information manually, redact manually. Every request required hours of human labor.
The math didn’t work:
A transparency authority receiving 5,000 information requests per year
Average 10 hours of staff time per request
= 50,000 hours per year = 25 full-time staff just handling requests
Most transparency authorities had 5-15 staff total
The framework was excellent. The manual implementation was unsustainable at scale.
2025: How AI Changes Everything
Now let’s fast-forward to today. The same scenario from 2015:
A citizen submits a freedom of information request: “All correspondence between the mayor’s office and contractors regarding the new bridge project from 2013-2014.”
Your response process with AI:
Staff member enters the request into the AI system in natural language
AI searches all document repositories simultaneously
AI identifies relevant documents based on content, not just metadata
AI flags potentially sensitive information for review
Human reviews AI suggestions, makes final redaction decisions
AI compiles documents with proper organization and indexing
Deliver response
Time required: 30-60 minutes of staff work Average response time: 2-4 days Quality: More consistent (AI doesn’t get tired or distracted)
The difference: AI handles the mechanical work—searching, initial classification, pattern matching, organizing. Humans handle judgment—”Is this really relevant?” “Does this need redaction?” “Is this response adequate?”
This isn’t science fiction. This is happening now in government agencies around the world.
What AI Automates from the RTA Framework
Let me show you exactly how AI implements each component of the framework the RTA built:
1. Automatic Document Classification
The RTA approach (2015): Staff member reads document, understands content, assigns to correct classification category based on function and activity.
The AI approach (2025): AI reads document content using natural language processing, identifies what it’s about, determines function and activity, assigns classification automatically with 93-97% accuracy.
Example: A contract for snow removal services arrives.
Human thinks: “This is about public works, specifically street maintenance, subcategory winter services. Classification code: PW-SM-WIN-001.”
AI thinks: Document contains keywords “snow removal,” “street maintenance,” “seasonal contract,” mentions Public Works department, has contract structure. Classification: PW-SM-WIN-001. Confidence: 97%.
AI makes mistakes about 3-7% of the time. But humans make mistakes too, especially when tired or distracted. The key is AI + human review = better than either alone.
2. Automated Metadata Generation
The RTA approach (2015): Staff member manually fills out metadata form: document title, creation date, author, subject, related documents, storage location, access restrictions.
The AI approach (2025): AI extracts metadata automatically from document content:
Identifies dates mentioned
Extracts names of people and organizations
Determines subject matter from content analysis
Finds related documents based on entity matching and topic similarity
Suggests appropriate access restrictions based on content
Result: Complete metadata for every document, consistently applied, generated in seconds instead of minutes.
3. Intelligent Retention Schedule Application
The RTA approach (2015): Staff member identifies document type, looks up retention schedule, calculates destruction date based on document creation date and applicable rules, handles exceptions (legal holds, historical significance).
The AI approach (2025): AI identifies document type, automatically applies retention rules, calculates destruction dates, flags documents that might have historical significance for human review, tracks legal holds automatically.
Example: A personnel termination record is created.
- Manual process: Someone needs to know this is a termination record, look up that termination records have 75-year retention, calculate 75 years from termination date, enter destruction date, set reminder for year 2100.
- AI process: Identifies document type from content, applies 75-year rule automatically, calculates destruction date, adds to retention tracking system. Done in seconds.
4. Natural Language Search
The RTA approach (2015): Searcher uses keyword matching based on metadata. Success requires knowing correct classification terms and having complete metadata.
The AI approach (2025): Searcher uses natural language. AI understands intent and searches based on meaning, not just keyword matching.
Example: Someone searches: “correspondence about the bridge project delays last year”
- Keyword search: Looks for documents with “bridge,” “project,” “delays,” “2024” in metadata. Misses documents that discuss the issue but use different words.
- AI search: Understands they want communications (emails, letters, memos) related to schedule setbacks on bridge construction from 2024. Finds documents that say “construction timeline setback,” “vendor scheduling issues,” and “spring weather delays” because it understands these relate to the search intent.
Result: People find what they actually need, not just what matches their keywords.
5. Assisted Information Request Response
The RTA approach (2015): Each request requires manual search, review, redaction, compilation.
The AI approach (2025): AI handles initial search and organization. Humans review and approve.
The transformation: A request that took 10 hours of staff time now takes 45 minutes—and produces more complete results because AI searches more thoroughly than humans can.
Real-World Impact: What’s Actually Happening
Let me share what’s happening in agencies that have combined the RTA’s proven framework with modern AI:
Case Study 1: Chilean Municipal Government
Before (using manual RTA framework):
- 800 information requests per year
- Average response time: 18 days
- 3 staff members, 80% of time on requests
- Frequent complaints about incomplete responses
After (RTA framework + AI implementation):
- Same 800 requests per year
- Average response time: 4 days
- Same 3 staff members, 30% of time on requests
- Improved response completeness
- Staff now spend freed-up time on proactive transparency and improving data quality
How they did it: They didn’t abandon the RTA classification structure—they taught AI to use it. Documents still go into the same categories, but AI assigns them automatically. Metadata standards stayed the same, but AI generates the metadata. The framework remained; the execution became automated.
Case Study 2: Mexican State Transparency Commission
Challenge: Managing 12,000 requests annually with 8 staff members. Backlog growing.
Solution: Implemented AI for:
- Initial request classification (what type of information is requested?)
- Document identification (what records are relevant?)
- Preliminary sensitivity review (what might need redaction?)
Results after 12 months:
- Backlog eliminated
- Average response time: 21 days → 7 days
- Staff satisfaction increased (less tedious work)
- Request quality improved (more thorough searching)
- Could handle 40% more requests without adding staff
Critical insight: The AI didn’t replace the RTA methodology—it implemented it at scale. Classification structure: unchanged. Retention rules: unchanged. Metadata standards: unchanged. The difference: automation.
Case Study 3: Argentine National Archive
Challenge: 3 million historical documents, mostly scanned but not catalogued. Creating archival descriptions manually would take decades.
Solution: AI-powered archival description:
- Reads scanned documents
- Generates descriptive metadata
- Identifies key entities (people, places, events)
- Suggests archival arrangement
- Creates finding aids
Results:
- 3 million documents described in 14 months
- Quality: Archivists review and refine AI descriptions, but starting with 85% complete metadata instead of blank forms
- Researchers can now find materials that were previously invisible
- Archivists focus on complex or historically significant materials requiring expert analysis
What This Means for Government Agencies Today
If you’re working in government transparency, records management, or archives, here’s what the evolution from manual RTA framework to AI implementation means for you:
1. The Principles Still Apply
The RTA didn’t get the framework wrong. Classification by function and activity? Still correct. Proper retention schedules? Still essential. Complete metadata? Still critical. Access controls balancing transparency and privacy? Still necessary.
AI doesn’t change what needs to happen. It changes how quickly and consistently it can happen.
2. AI Makes Best Practices Achievable
Many agencies couldn’t fully implement the RTA framework because it required too much manual labor. AI makes full implementation realistic:
- Want consistent classification? AI can classify every document the same way every time.
- Want complete metadata? AI generates it automatically.
- Want to respond to information requests quickly? AI finds relevant documents in seconds.
- Want proactive transparency? AI can identify disclosure-required documents automatically.
The best practices the RTA documented are now practical for agencies of any size.
3. You Need the Framework AND the AI
Here’s what doesn’t work: Buying AI tools without a proper framework.
I’ve seen agencies try:
- “We’ll just throw AI at our document chaos and hope it figures things out.”
- Result: Garbage in, garbage out. AI can’t fix fundamental organizational problems.
What works:
- Establish proper classification structure (like the RTA framework)
- Define retention rules clearly
- Set metadata standards
- THEN implement AI to apply these systematically
The RTA provided the framework. AI provides the execution capability. You need both.
4. Human Expertise Remains Critical
AI doesn’t replace transparency professionals, records managers, or archivists. It changes what they spend time on:
Less time on:
- Manually classifying documents
- Filling out metadata forms
- Searching for documents
- Routine redaction reviews
More time on:
- Complex judgment calls
- Policy development
- Training and guidance
- Handling unusual cases
- Improving processes
- Strategic planning
- Citizen service
The profession doesn’t disappear—it becomes more strategic and less mechanical.
How to Get Started: Implementing AI with the RTA Framework
If you’re ready to explore how AI can implement the document management principles the RTA established, here’s a practical approach:
Phase 1: Understand Your Foundation (Weeks 1-4)
Questions to answer:
- Do you have a classification structure? (If not, the RTA framework provides one)
- Do you have documented retention schedules?
- What metadata standards do you follow?
- What are your biggest pain points?
- What processes consume the most staff time?
Deliverable: Clear understanding of current state and priority problems to solve.
Phase 2: Select Appropriate AI Tools (Weeks 5-8)
Not all AI tools are equal. Evaluate based on:
- Can it learn your classification structure?
- Does it support your metadata standards?
- Can it handle your document types?
- Does it meet government security requirements?
- Can staff actually learn to use it?
Approach: Test with your actual documents. Vendor demos with cherry-picked examples don’t tell you how it’ll perform on your real work.
Phase 3: Pilot Implementation (Months 3-6)
Start small:
- One department or one document type
- Run parallel (old process + AI) initially
- Measure everything (accuracy, time savings, quality)
- Train staff thoroughly
- Adjust based on what you learn
Success criteria:
- 90%+ AI accuracy (with human review)
- Measurable time savings
- Staff confidence in using the system
- No compliance problems
Phase 4: Scale Gradually (Months 7-12)
Expand systematically:
- Add document types incrementally
- Add departments one at a time
- Maintain support and training
- Celebrate wins, learn from problems
- Continuously refine
Timeline: Most agencies achieve full implementation in 12-18 months. Rushing leads to problems.
Common Questions About AI + RTA Framework
“Will AI work with documents in Spanish, Portuguese, and indigenous languages?”
Yes. Modern AI handles multiple languages well, including Spanish and Portuguese. For indigenous languages, accuracy depends on how much training data exists for that language. Test with your actual documents.
“Our documents are handwritten or poorly scanned. Can AI handle that?”
Modern OCR (optical character recognition) powered by AI handles handwritten text surprisingly well—certainly better than older OCR technology. It’s not perfect, but it’s improving rapidly. Again: test with your documents.
“We can’t afford expensive AI systems.”
Prices have dropped dramatically. Cloud-based AI document management tools often cost less than hiring additional staff—and they work 24/7 without vacation or sick leave. Many governments find ROI within 12-18 months.
“What if AI makes mistakes on sensitive documents?”
This is why humans remain in the loop. AI suggests; humans review and approve. For sensitive operations (redacting personal information, determining what can be disclosed), human review is mandatory. AI speeds the process by doing preliminary work, not by making final decisions unsupervised.
“Will this eliminate jobs?”
No. It eliminates tedious tasks. The profession remains—the daily work changes. Transparency officers spend less time manually searching files and more time helping citizens understand complex information. Records managers spend less time filing and more time improving information governance.
“We’re a small agency. Is this only for large organizations?”
Actually, small agencies benefit more. A large agency might afford dedicated classification staff. A small agency with 3 people doing everything desperately needs automation. Many AI tools have pricing models that make them accessible for small agencies.
“How do we maintain the RTA framework’s emphasis on archival preservation?”
AI actually improves archival preservation because:
- More complete metadata makes documents discoverable long-term
- Consistent application of retention rules protects historically significant documents
- Better search means archival materials get used, not forgotten
- AI can identify documents with historical value that humans might overlook
The RTA framework’s preservation principles remain; AI makes them more achievable.
The Future: What’s Coming
We’re at the beginning of this transformation, not the end. Here’s what’s emerging:
1. Proactive Transparency
Instead of waiting for requests, AI will automatically:
Identify documents subject to proactive disclosure requirements
Redact sensitive information
Publish to transparency portals
Notify interested citizens
Governments will publish more information faster with less manual effort.
2. Intelligent Search Interfaces
Citizens won’t need to know official document types or classification codes. They’ll ask questions in natural language:
“What did the city spend on street repairs last year?”
“Show me meeting minutes where they discussed the new park.”
“I want to see all contracts with [company name].”
AI will find and present relevant information instantly.
3. Multilingual Access
AI translation will make government information available in any language automatically. A document published in Spanish appears in Portuguese, English, Quechua, or Guaraní when requested. Information becomes accessible regardless of language barriers.
4. Continuous Improvement
AI systems learn from corrections. When a human fixes an AI mistake, the system improves. Over time, accuracy increases, and manual intervention decreases.
5. Cross-Agency Learning
Agencies will share AI training (not documents—training data) about what works. Classification schemes, retention logic, common redaction patterns—these insights will spread faster, helping all agencies improve.
Conclusion: Building on Solid Foundations
In March 2015, the RTA launched a framework that helped government agencies manage documents more effectively. The principles were sound. The implementation was thorough. The impact was significant.
Ten years later, those principles remain valid. We still need proper classification. We still need retention schedules. We still need complete metadata. We still need to balance transparency with privacy.
What’s changed is our ability to implement these principles at scale. AI doesn’t replace the RTA framework—it makes the framework achievable for agencies of any size, with any budget, handling any volume of documents.
The promise of the original MGD—professional, systematic document management supporting transparency and accountability—is finally realistic for every government agency.
That’s why we’ve preserved mgd.redrta.org and updated it for the AI era. The framework the RTA built deserves to continue serving government agencies and citizens. It just needed modern execution capabilities.
If you’re exploring how AI can help your agency implement proper document management and transparency practices, we’re here to help. The RTA built the foundation. AI provides the tools. Together, they can transform how government serves citizens.
Resources and Next Steps
Learn More
- Visit the modernized MGD Hub → – See how AI implements each RTA framework component
- Read the original RTA frameworks → – Historical documentation remains valuable