The Client

Certora Training is the UK’s leading provider of equipment operator training, qualifications and assessments. The company delivers hundreds of training sessions every month collecting delegate feedback through paper-based course evaluation forms at the point of delivery.

The Challenge

Certora’s feedback collection process is entirely manual. Tutors distribute and collect paper forms, which are later returned to Certora head office. Administrative staff manually input the data into a spreadsheet and archive the paper documents for up to two years.

The process was:

• Time-consuming and labour-intensive
• Prone to human error
• Physically dependent on storage space
• Inflexible and difficult to analyse feedback at scale

Certora wanted to modernise this process by leveraging Artificial Intelligence (AI) whilst maintaining data privacy and platform choice by staying within the Microsoft ecosystem.

Our Approach

Magnifica was engaged to assess, prototype, and recommend an AI-based solution that would automate the processing of hand-written course feedback forms using Azure AI Document Intelligence and SharePoint:

• Digitise and process handwritten feedback forms
• Eliminate the need for manual data entry
• Store responses digitally for easier access and analysis
• Provide structured, validated output ready for reporting

We began by analysing the current workflow to understand where AI could best support the team. It was clear that the data entry element was the part most suited to AI.

Certora are heavily embedded in the Microsoft ecosystem and use SharePoint for much of their document and data storage and retrieval. This was the natural destination for the data once extracted by AI. It also made sense for the paper feedback forms to be scanned to SharePoint where the AI extraction would be initiated.

Model testing and feasibility

It was established that there was a strong use case for AI. Next we tested the feasibility of AI tools including ChatGPT, Claude.AI, and Microsoft Azure AI Document Intelligence to establish how effective each was at reading and extracting hand-written content from the forms.

Our testing followed a test plan that used success criteria and a scoring matrix to determine how the models responded to the task. Reading of hand-written text can be challenging however the models we tested were able to return results with sufficient accuracy and in a structured format.

The results met the criteria we set meaning a system that used AI to extract data from scanned documents was viable.

The next step was to consider how the AI would fit into the organisation’s feedback gathering workflow.

Workflow integration

The initial data collection would remain paper-based with the documents being returned to Certora headquarters as physical artifacts. These documents would then need to be digitised by some method, assumed to be via a scanner, and then stored ready for AI to process. Once the data is extracted the documents and their associated data would need to be stored for later retrieval and analysis.

The extracted data would then need to be saved into the master spreadsheet, also stored in SharePoint.

Prototyping

A basic prototype was developed that monitored a SharePoint library for new scanned feedback forms being uploaded. This was successful. We now had the components of a solution that covered the following:

• Input – Scan / document storage
• Process – AI model for text extraction
• Storage – Save extracted data to spreadsheet and move processed document to SharePoint archive library

Now it was clear that we could scan the documents, store them in SharePoint, process them using AI and save the extracted data back to SharePoint. Next we developed a proposed solution.

Proposed solution

The results from the model testing and feasibility, workflow integration and prototyping phases allowed us to design a solution that would meet the brief set out at the start of the project.

The solution we proposed would have a dramatic impact in terms of efficiency, access to data and reduced physical document storage requirements.

The basic workflow of the system follows an Input > Process > Storage approach. The Process part of the workflow is where the AI is leveraged.

System overview

System detail

The core of the solution will be implemented as an Azure Function (cloud hosted serverless scheduled job). This function will run on a nightly schedule to get all files from source location, process them using AI model, and save to output spreadsheet and file repository.

The steps are as follows:

• Files scanned to shared location (SharePoint) as individual jpg image files
• Files processed using AI
  • Data read, extracted and processed (date parsing etc)
  • Data validated and scoring
• Row added to spreadsheet (stored in SharePoint)
  • Url to source file added to new column
• File moved to storage repository (SharePoint list)
• Final destination will be organised in a meaningful directory structure
  • Log files added for each run
  • Provision for “retry a batch” built into the flow

AI scoring

To account for the fact that no matter how impressive the results from AI can be its not perfect. So, we need a way to score the quality of the data extraction phase. A new column will be added to the results spreadsheet where a rating will be saved based on the quality of the response from the AI model.

The Results

Magnifica delivered a detailed report serving as an action plan for the development and implementation of a production ready solution. A proposed infrastructure approach, AI model selection, workflow diagrams and outline model costs were all included.

Certora now benefit from a plan to implement a fully automated, scalable, and secure process for handling their evaluation forms. The AI model ingests, validates, and archives data automatically, cutting down hours of weekly admin and reducing human error.

Once live this solution will deliver:

• 🔐 Full Microsoft ecosystem compliance – no public model usage
• 📥 Automated nightly processing of all scanned forms
• 📊 Structured data output directly into SharePoint spreadsheets
• 🔍 Data confidence scoring to allow quick human QA when needed
• 📁 Digital archiving, eliminating the need to store physical documents

A targeted consultancy project that helped Certora Training to go from idea to project plan in a couple of weeks. We’re looking forward to getting this AI solution implemented.