Lendlease Data Automation
At Lendlease, hundreds of site visits went unrecorded, so management assumed they weren't happening at all. Solo, in two to three weeks, I built a system that captures the team's existing email habit and uses a custom, LLM-powered Claude agent I designed and trained to read, standardise and surface every visit. It saved a 16-person team 30+ hours a week.
Role
Digital Experience Analyst
Industry
Construction / IT (enterprise)
Team
Solo (16-person team served)
Timeline
2–3 weeks
Problem
Management assumed zero; reality was hundreds
Lendlease runs hundreds of sites, each needing constant upkeep, IT support and commissioning or decommissioning. Those site visits carry real operational, compliance and reporting weight. But none of them were being recorded anywhere. The work happened through an informal habit: someone would email the team, "heading to this site, offline today," and that was it. Nothing was captured, standardised or surfaced.
So management had no visibility, and worse, a false picture. The organisation believed site visits simply weren't happening, when in fact they were happening constantly. Decisions were being made on an assumption the data would have flatly contradicted. The gap wasn't just missing reporting; it was an operational blind spot with compliance exposure, and no one could see it to fix it.
Process
Capture the habit, don't change the behaviour
The core decision was to meet people where they already were. Rather than impose a new form, I built an email catcher that formalises the informal habit: the "heading to site" email the team already sent became the trigger. Zero friction, no trade-off on how anyone worked, and adoption was immediate.
I designed the Microsoft Lists architecture first, so there was a clear structure to capture into, then designed and trained a custom, LLM-powered Claude agent in Power Automate to read each email, extract the visit details, and look the site up against our records. Training it meant grounding it in real examples, shaping the extraction to the schema I'd defined, and validating its output against known site records until the edge cases held. That discipline is what made the extraction trustworthy rather than approximate.
From there every visit flowed into Microsoft Lists as the single system of record and into Power BI dashboards for management: a clean pipeline from an offhand email to governed, visible data, running across every site in Australia.
Outcome
From zero on record to full visibility
The system turned an invisible process into governed, visible data. More than 200+ site visits have been captured and standardised so far, and a visit now reaches a manager's dashboard in around four minutes instead of never. It covers every site in Australia.
The business impact was immediate. The automation saved the 16-person team more than 30 hours a week, and just as importantly, it gave management operational visibility they had never had. The data didn't just fill a gap; it overturned a false assumption, proving the visits were happening all along and adding value both inside the team and beyond it.
30+
Hours a week saved (16-person team)
4 min
From an email to a manager's dashboard
500+
Site visits surfaced (none were on record)
Reflection
Adoption came from changing nothing
The clearest lesson was that the best automation often changes nothing about how people already work. The system succeeded because it sat on top of an existing habit instead of asking anyone to adopt a new one, which is why adoption was immediate and there were no trade-offs against current processes.
The other lesson was about data as an argument. Surfacing the visits didn't just improve reporting; it overturned a confident, organisation-wide assumption. It reminded me that the highest-value thing a system can do is make the invisible undeniable, and govern it well enough to be trusted.