About Chancerygate
Chancerygate is a London-headquartered industrial developer and asset manager with a 30-year track record in multi-let industrial and urban logistics property across the UK, Ireland, Spain and Portugal. It manages around £1.2 billion of income-producing and development assets, with around 30 development schemes and 25 asset management schemes serving roughly 700 tenants. In July 2026, Dream, the Canadian real estate asset manager, agreed to acquire Chancerygate, bringing it into a pan-European industrial platform of around 27 million sq ft across nine countries.
The challenge
Chancerygate already had Claude in the business, but much of its back-office work was still manual. An AI workshop with Artefact in spring 2026 found the same pattern across finance, legal and investment teams:
- Around 100 supplier invoices a month extracted, checked and re-keyed into the ledger by hand, costing around 1.5 FTE-days a month.
- Around 1.5 hours spent validating the minutes of each large meeting, some with 30 attendees.
- NDAs reviewed manually, with inconsistent coverage of risk clauses.
- Underwriting models set up by hand, copying tenancy schedules and case facts from PDFs and spreadsheets into a large Excel template.
Leadership wanted to prove value quickly, on Chancerygate’s own Microsoft environment, and leave the team able to build agents without relying on consultants.
A two-week agent sprint
Artefact ran a two-week sprint as one team with Chancerygate. Day one was a hands-on “How to build an AI agent” workshop, where stakeholders built a meeting note taker live. Each agent then had a named owner, a test user and a maintainer from Chancerygate who shadowed the build.
Every agent is built on one reusable framework. Power Automate flows pick up documents from email, a monitored mailbox or a SharePoint folder. A core Power Platform flow then gathers the metadata, the relevant schema and context-specific reference material, and sends them securely to Claude Sonnet through the Claude API for analysis. Claude’s output comes back as a structured dossier for a person to check before anything moves on. The framework holds the integration points in one place and is maintained day to day through SharePoint, so a new agent is mostly configuration rather than a new build.
Meeting note taker. When a Teams transcript is ready, the agent identifies the attendees, has Claude draft the notes and actions, and sends them only to the people who were in the meeting.
Accounts payable agent. The agent monitors the AP inboxes, loads invoice attachments and has Claude populate a schema defined by Chancerygate’s accounts specialists. It categorises each line item ready for onward processing, flags anything it cannot match and reports KPIs to the process owner.
Underwriting model initialiser. The agent loads source files such as tenancy schedules and uses Claude, with OCR where needed, to populate a staging data layer. An existing VBA macro then loads the staging data into the investment or development model, so the agent hands over to the tools analysts already use, and the same staging data can serve other purposes.
NDA review agent. When an NDA arrives, the user saves it to a SharePoint folder and a Power Automate flow sends it to Claude for review against Chancerygate’s guidelines. The user receives a summary of risk clauses, such as liability and termination, before deciding whether the NDA can go forward for signature.
The sprint ran as a side-by-side build, in training-style sessions with each agent’s owner and maintainer. Several Chancerygate team members built their own workflows on their own initiative during the workshops, and the core architecture knowledge was transferred so that Chancerygate’s team could take over as long-term maintainers. All four agents were demonstrated to the wider business on 16 July, then handed over with the reusable framework, prompt templates and handover videos.
Results
- 4 Claude-powered agents designed, built and demonstrated in a two-week sprint
- 80% less time spent processing each invoice, with over 90% of invoice line items matched by the accounts payable agent
- NDA turnaround cut from around 3 days to under 2 hours
- Underwriting model set-up cut from 25+ hours to less than 3 hours
- Validation of minutes for large meetings down from around 1.5 hours to around 30 minutes
- 20+ further agents built by Chancerygate’s own team since the sprint, each now taking less than half a day to build and test
- Around 10 core team members using the agents every day, and around 25 staff trained, who are now training colleagues themselves
- All agents run on Chancerygate’s own Microsoft tenant and Claude subscription, maintained in-house
“The Artefact team was brilliant. In just a little over two weeks, they up-skilled the team on building and managing Claude agents, worked with individual team members to build dedicated agents and left us in a good place to continue building and expanding on our own. As we are ramping up our activities quite significantly, this could not have come at a better time. We are still early in the journey, but we are increasingly adopting AI across all our operations,” says Phil Price, Head of FP&A, Chancerygate.
What’s next
Chancerygate’s own teams are now deploying further agents for lease abstraction, EPC data extraction and investor reporting, all identified in the original workshop. Following its acquisition by Dream, Chancerygate is part of a larger European industrial platform with a growing development pipeline, and is using the same agent framework to scale its operations as activity increases. Next on the roadmap are closing the loop on NDA signing through DocuSign and integrating the accounts payable agent directly with NetSuite, so validated invoices update the ledger without manual entry.
About Artefact
Artefact is an international data and AI consulting firm headquartered in Paris, with 2,500 experts across 36 offices. It helps more than 1,000 companies accelerate their adoption of AI, from strategy to production, and is a member of the Claude Partner Network.

CLIENT CASES




