Skip to main content

This is a prototype vision of how a future government service could work. It's not a real service yet, but we're exploring what it could look like. Your feedback will help shape the real service.

Walkthrough Complete - Planning Application AI

You’ve completed the Planning Application AI walkthrough

Walkthrough complete

You've experienced AI-powered document extraction for planning applications

Congratulations! In just 8 minutes, you've seen how AI can eliminate hours of manual data entry from planning application processing. What would have taken a traditional proof-of-concept weeks to demonstrate, you've experienced hands-on right now.

What you've learned

🤖 Automatic field extraction

AI read your planning PDF and automatically extracted applicant name, site address, and development description. No templates, no configuration - it just worked.

Value: 45 minutes of manual typing reduced to 15 seconds

📊 97-99% accuracy

Confidence scores showed 97-99% accuracy on typed documents. Better than human transcription, with zero typos or misread postcodes.

Value: Eliminated transcription errors and costly corrections

🏗️ Format independence

AI worked with any planning application format - householder, full, outline. No template matching or custom configuration needed.

Value: Works with applications from any planning portal or agent

⚡ Instant processing

8-20 seconds to extract 5-6 fields from multi-page PDFs. Scales to hundreds of applications with no slowdown.

Value: Officers focus on decision-making, not data entry

Committee-ready talking points

Use these points when presenting to decision-makers:

The problem

"Planning officers currently spend 45 minutes per application manually entering applicant details, addresses, and descriptions into our planning system. With 200 applications per year, that's 150 hours of valuable officer time spent on data entry."

The solution

"AI document extraction automatically reads planning PDFs and populates our system in under 15 seconds, with 97-99% accuracy. This frees up 112 hours annually - equivalent to nearly 3 weeks of officer time - to focus on actual planning decisions."

The value

"Annual cost savings of £2,812 in staff time, plus faster application processing for residents. Officers report higher job satisfaction working on complex planning issues rather than repetitive data entry."

The risk

"Low. Pay-per-use pricing means no upfront investment. Trial with 50 applications costs under £20. Fields with confidence below 95% automatically flag for human review, ensuring accuracy."

Next steps

Generate Evidence Pack

Create your business case with what you've learned. Perfect for committee papers.

Generate Evidence Pack

Return to Planning Application AI

Review deployment options, costs, and technical details.

Back to scenario

Try Another

Explore more AI scenarios for local government.

Browse scenarios
Important Remember to clean up your AWS resources

Clean up your resources

Your Planning AI resources will automatically delete after 2 hours, but you can delete them now to stop any further charges:

  1. Go to CloudFormation console
  2. Find your stack: ndx-try-planning-ai-[timestamp]
  3. Select the stack and click "Delete"
  4. Confirm deletion - this takes 2-3 minutes
  5. S3 bucket will be automatically emptied before deletion

Estimated cost so far: Less than £0.30 for an 8-minute evaluation (Textract charges per page processed)

What if I want to test with more documents?

If you want to continue testing document extraction:

  • Resources will auto-delete after 2 hours total (from deployment)
  • Cost: approximately £0.10 per planning application processed
  • Maximum cost: £10 (enforced by template configuration)
  • You can upload your own planning PDFs (max 10MB)

You can redeploy the scenario anytime to start fresh with new sample documents.

Production considerations

If you're considering deploying this for real planning applications:

Integration with existing planning systems

Production deployment would integrate via:

  • API endpoints to send extracted data to your planning database
  • Webhooks to trigger processing when applications submitted via planning portal
  • Direct integration with Idox, Ocella, or other planning systems
  • Queue-based processing for batch imports of existing applications
Handling low confidence extractions

Production systems would:

  • Set confidence threshold (e.g., 95%) for auto-approval
  • Flag low-confidence fields for officer review
  • Show original PDF section next to extracted value for easy verification
  • Track accuracy over time and retrain on edge cases
  • Allow officers to correct and provide feedback
Data protection and security

Production deployment includes:

  • Encryption at rest for S3 buckets (customer-managed KMS keys)
  • Encryption in transit (TLS 1.2+)
  • VPC endpoint for Textract (traffic never leaves AWS network)
  • IAM roles with least-privilege permissions
  • CloudTrail logging for audit compliance
  • Automatic PDF deletion after configurable retention period
  • GDPR-compliant data processing agreements with AWS
Cost at scale

Real-world cost examples:

  • 200 applications/year: ~£200/year (£1 per application)
  • 500 applications/year: ~£450/year (£0.90 per application due to bulk discounts)
  • 1000 applications/year: ~£800/year (£0.80 per application)

Costs include Textract (per page), Lambda (per execution), S3 storage, and API Gateway. No upfront licensing or infrastructure costs.

Questions or feedback?

We'd love to hear about your experience with this walkthrough:

Return to homepage

Need help?

If you're stuck, check the troubleshooting section or contact us.