Enterprises everywhere are discovering the same thing: large language models are powerful, but they can only operate reliably when they understand the reality of how a business works. Processes, exceptions, workarounds and institutional memory form the detail that keeps an organisation running, and this information rarely exists in a form AI can use.
This context gap is now one of the biggest barriers to dependable AI automation.
At ASOS, we’ve taken a practical first step to close that gap, starting in IT Operations. We partnered with Edra, a New York and London-based AI company, to help us surface and structure the institutional knowledge our teams use every day, but which wasn’t captured in our existing documentation or systems.
Why IT was the right starting point
Our IT service desk manages thousands of tickets each month, supported by a library of hundreds of knowledge articles. Like most organisations of our scale, a lot of the real problem‑solving happened outside those articles, including in resolution notes, in patterns buried across ticket history, or simply in people’s heads.
This undocumented knowledge isn’t just useful, it’s essential for AI to make accurate decisions and automate work with confidence.
What we learned
Edra analysed our full ticket history alongside our existing knowledge base. Within days, we had a complete, evidence‑backed picture of how issues are truly resolved at ASOS. The system identified improvements across roughly half of our articles and revealed hundreds of topics we hadn't documented.
This will strengthen the foundation for automation:
c.30% of inbound questions can now be resolved automatically.
A further c.30% are ready for automation once system‑level actions are enabled.
The remaining tickets benefit from intelligent routing, removing manual triage and getting issues to the right teams faster.
Crucially, the process is fully transparent. Every recommendation is linked back to verifiable evidence, and our engineers remain in control by reviewing, editing and approving updates before anything goes live. This is not a black‑box AI layer; it is a continuously improving, auditable knowledge foundation.
Why this matters for the business
Most AI initiatives struggle not because the technology isn’t capable, but because the underlying knowledge isn’t structured or complete. Organisations that solve the knowledge problem first are the ones unlocking reliable, scalable automation.
For ASOS, IT is just the beginning. The same approach applies wherever operational expertise lives in data but is not formally captured, from customer care to logistics and beyond.
What comes next
This work gives us a blueprint for building AI systems that genuinely understand how ASOS operates. It ensures we’re automating on solid ground, not assumptions. And it helps us move faster, with greater accuracy, across the parts of the business where deep institutional knowledge is crucial.