When Redactive AI was tasked with building a chatbot to double productivity at PEXA, Australia’s digital property settlement giant, the challenge was clear: make it happen with ‘purposeful, responsible’ AI. Here’s how they pulled it off.
When Fred Herselman wanted to incorporate an AI system into the PEXA (Property Exchange Australia) enterprise to drive employee productivity, securing sensitive data was top of mind.
“We were specifically looking for an AI partner that uses permission-aware vectorization of the data,” Herselman tells Forbes Australia in an exclusive interview. “We couldn’t take the first step into using an LLM without first solving the vectorization.”
Herzelman is PEXA’s general manager of Decision Intelligence and AI. The ASX-listed company digitises more than 75 per cent of property transfer settlements in Australia and is valued at $2.5 billion.
The company’s focus on closing property transactions requires that some employees have access to buyers’ and sellers’ sensitive, confidential information. Melbourne-headquartered PEXA recently launched in the UK market and has plans to continue international expansion. So when Herselman wanted to incorporate an AI chatbot into employees’ workflow, he knew it had to be secure.
“We have 1,100 employees globally, approximately 60% are in the Aus, and 40% are in the UK. We were observing that AI capabilities were accelerating innovation in other organisations, but we are very risk averse. We really need guardrails around data,” says Herselman.
PEXA’s cloud provider Amazon Web Services suggested Australian-founded Redactive AI might have a solution that could help. Herselman was intrigued. His first goal was to develop an AI chatbot that employees could use to double productivity. If all goes well, PEXA may roll the AI system out to assist in closing property transactions.
“The system we have developed effectively gives every employee an assistant,” says Herselman. “A financial analyst or someone working in HR can securely use AI capabilities to remove some of the mundane repetitive jobs that they are facing.”
The AI employee system was built in partnership with Redactive AI, as well as AWS, and technology consulting firm Thoughtworks. Company data can be accessed via Redactive’s permissions-aware retrieval engine, and uses Anthropic’s Claude 3 Sonnet model on Amazon Bedrock.
Andrew Pankevicius, is co-founder and CEO of Redactive AI.
“The standard industry framework uses a shortcut to access information, effectively copying from one source and pasting into an LLM. We don’t want to create a new document or ‘paste’ anything, which may change the security risk. Our system is different, it is built to respect fine grain permission of data,” says Pankevicius.
It does that by enabling permissioned-Retrieval Augmented Generation, or RAG in AI parlance.
“You can confidently use the LLM in conjunction with your sensitive documents without the risk of data leaks,” says Pankevicius.
Questions that PEXA employees can ask the AI chatbot
Herselman says that PEXA Sales and HR departments can now use the employee AI system.
An example of a prompt that an HR employee might ask the chatbot, is to “analyse this CV against the job description.” Rather than copying and pasting the CV and job description into the LLM, Redactive AI has designed a system where the LLM brings the documents to the employee, together with the requested analysis.
Not ‘pasting’ the document into a system that may have security flaws, enables PEXA to maintain confidentiality, and uphold ethical standards.
Another example of how the chatbot can be used is in the procurement process.
“An employee could ask the chatbot to analyse a statement of work, and a proposal, and to contrast and summarise the differences between them,” says Herselman.
PEXA’s Chief Technology Officer Eglantine Etiemble says the tool integrates cutting-edge technology into the product, while also ‘fostering a culture where employees can thrive and engage with new tools confidently.’
In addition to her work with PEXA, Etiemble serves on the board of directors for the Tech Council of Australia. She says she is driven to ‘Innovate for Good,’ and is pleased to spearhead technology that has responsible AI at its core.
“PEXA’s AI Assistant is a significant step forward in the company’s journey toward an AI-powered future, reflecting its ongoing commitment to innovate responsibly and effectively across the business landscape,” according to a company statement.
Layers of security
Given the amount of sensitive and confidential information PEXA handles, Pankevicius says that embedding multiple layers of security protection into the PEXA AI platform was essential. One of those layers was built in through AI developer platform Amazon Bedrock.
“Security is a fundamental requirement for PEXA, as well as for many other organisations – especially in regulated sectors like financial services,” says Jamie Simon, director of AWS ANZ financial services.
The multiple layers of security are a unique proposition, and were designed by Redactive alongside technology consultancy firm Thoughtworks.
“Unlike existing models that often rely on outdated architectures and lack fine-grained access control, PEXA’s solution ensures robust data security at both the model and data levels, retrieving live information from within the organisation. The platform operates using a permissioned Retrieval Augmented Generation model, which controls what data is passed to the AI in small, secure segments,” according to Redactive AI.
Look back on the week that was with hand-picked articles from Australia and around the world. Sign up to the Forbes Australia newsletter here or become a member here.
