History & Projects
Our AI work is built on 36 years of asking how data should be stored when the people who need it don’t know in advance what they’ll need.
1989: the question that started it
Azquo’s development began over 35 years ago. Bill Cawley was given the task of defining the basis of an expert system for marketing — this was the mid-1980s, when expert systems were considered the future of computing.
Two things became clear early in the research. First, marketers rarely know at the outset of a project exactly what data they will require — and that data is likely to change as the work develops. Second, the spreadsheet was already the dominant analytical tool for business users, and any solution that ignored it would be ignored in return.
Bill concluded that the prerequisite for any expert system was a database fundamentally more accessible than the relational databases of the time: one that would accept any data at any time without needing to call in IT, and that could transfer data to and from a spreadsheet seamlessly.
The longer account — how Rupert Armitage’s phone call started it, and what we built next — is in our blog post Ambit Research.
Early customers — proof at scale
The architecture proved itself with the toughest possible first audience: De Beers. Ten times a year, around 300 wholesale customers flew to London for a “sight.” Ark allocated up to about £5 billion of stones a year across them, matching every customer’s specifications against the available supply by size, colour, shape and hundreds of other attributes. A single buyer might take £3 million of stones at one sight; others much less. The data shape varied with every sight, yet the same database accepted it without reshaping. HSBC and Wimpey ran their own deployments on the same foundation in the same period.
One dataset, every report shape
Through the late 1990s, UK Pupil Results were collected at national scale and reported back to each Local Education Authority on the LEA’s own Excel workbook — its layout, its preferred breakdowns, its headline figures. The same canonical pupil dataset drove dozens of bespoke LEA report workbooks. The lesson stuck: spreadsheets are not the enemy of a structured database; they are the natural front end of one. We have designed for that ever since.
Dormant, then rebuilt
The underlying technology of 1988 — disk-bound storage, limited memory — eventually constrained what the architecture could deliver at speed. The founder turned to other businesses through the 2000s, including a successful trampoline-and-pool import company and then Feefo, the customer-feedback service still running today. By 2013, RAM was cheap enough to make an all-in-memory database routine. Azquo was relaunched on that footing — JSON for persistence, the original storage restrictions shed — with the same architectural thesis on modern foundations.
Lloyd’s of London — the durable proof
Our original Lloyd’s project was designed for the most difficult contracts at the syndicate: bordereau feeds where each broker’s data arrives shaped differently, with up to 80 distinct premium and claim classes per contract year, reported back through the workbooks the underwriters already used.
Since 2017 it has run smoothly. Years of continuous operation, no architectural rewrites. The clearest day-in-day-out proof that “any data, at any time” is not an aspiration but an operating mode.
Now, an AI layer on the same foundation
The two tasks that always required Azquo-specific knowledge — designing the import templates and writing the report queries — are precisely the tasks AI handles well. We have built an MCP server through which any MCP-capable AI agent can drive the database directly. In a recent test, Claude built a complete US insurance loss-cost database from a 168-page filing covering 73 territories, unaided. The architecture didn’t need to change to accommodate AI; AI turned out to be exactly the right collaborator for the architecture.
“I have always believed the method of storing data was valid. I was waiting originally for humans who agreed. I have found agents.” — Bill Cawley
Find Out More
The Team
Azquo remains a small, focused team: Bill Cawley and his son Edd, together with two colleagues, Ed, who liaises with customers, and Nick, who specifically understands insurance, based in Ludlow, Shropshire.
The team works directly with clients, building and adapting Azquo deployments to fit each organisation’s specific data challenges. We are actively looking for businesses that can exploit this unique database — organisations where conventional data tools are too rigid, too slow to adapt, or too dependent on IT specialists.