The friction of connecting AI document processing platforms to enterprise business systems has long stalled automation projects in regulated industries. A new tool from Melbourne-based Affinda aims to remove that barrier, allowing organisations to describe integrations in plain language and have the software write the code automatically.

Affinda's AI Integration Agent connects the company's Intelligent Document Processing (IDP) platform with downstream business systems - the vendor claims compatibility with more than 2,800 applications - without requiring developers or custom API builds.

The agent writes, refines, and tests integration code based on plain-language instructions provided by users. Among the supported enterprise platforms are Xero, Excel, Salesforce, Hubspot, Dynamics 365, OneDrive and Power Automate.

The tool targets two distinct audiences: organisations with in-house developers who want to rapidly prototype automation workflows, and smaller organisations without technical staff that have previously found integration costs prohibitive.

Affinda Head of AI Andrew Bird said the agent allows organisations to automate data export to "virtually any system - whether it's ERP, CRM, or other databases - using just natural language instructions."

Integration complexity has been a persistent barrier in IDP deployments. Organisations have typically faced a choice between rigid, pre-configured connectors or expensive custom API development - both of which add time and cost to automation projects before any documents are actually processed.

The Integration Agent attempts to address this by generating the integration code itself.

Affinda General Manager Charlie Bellingham cited a specialist lending firm as the type of organisation previously locked out of automated document and records workflows.

"The set-up time is drastically reduced from days or weeks - to something that can be done in 15 minutes," Bellingham said.

The broader context is a competitive shift in the IDP market. Traditional machine learning approaches required hundreds of training documents and ongoing retraining cycles.

Newer platforms - including Affinda - use large language models (LLMs) and retrieval-augmented generation (RAG) to reduce configuration time and handle greater document variability.

While this lowers the technical barrier to adoption, it also introduces new considerations around prompt governance and the traceability of AI-driven decisions in regulated environments.

“These are considerations core to our offering. Every output is traceable back to the source document, so data are auditable and defensible – which is exactly what regulated environments demand," Bellingham adds.

https://www.affinda.com