The situation
ClearGlass analyses, validates and benchmarks what pension schemes pay their asset managers. LGPS fund data arrived as an operational bottleneck: deeply hierarchical structures, processed by hand, every collection cycle. The platform that came out of it wasn't on anyone's roadmap — it was hiding inside the ops queue.
note: visuals on this page are redrawn concept illustrations — the real client work stays confidential.
Build, don't staff
The obvious answer to a growing manual workload is more people — and that was genuinely on the table. But scaling the ops team with the workload ties cost to revenue permanently: every new scheme means new hands. The platform case was structural — automate the processing once, and LGPS revenue could grow while the cost base stood still. That framing, margin rather than convenience, is what carried the investment decision.
Under-engineering on purpose
The engineers recommended full reconciliation engines, and the case was compelling — elegant, complete, correct on the whiteboard. We said no, for one reason: there was too much we didn't know. Before a first real collection cycle, a heavy build would have encoded our guesses. So we deliberately built lighter than we could have, shipped into the cycle, and let the cycle teach us what the second version actually needs.
A new client type, priced as one
LGPS funds aren't corporate pension clients with a different logo. The scheme is statutory, funded and defined-benefit, administered locally by dozens of funds with pooled assets and thousands of participating employers — which is why the data arrives as deep hierarchies rather than tidy tables. We treated that as what it was: a new client type. The platform absorbed the structural complexity, and a new charging scheme was built alongside it rather than stretching the existing commercial model over a shape it didn't fit.
Where it landed
A new revenue stream for the business, created from what had been pure operational cost. The platform mapped hierarchical structures, visualised data flows and supported transparent verification — and is now scaling on the learnings of its first collection cycle.
The most senior thing we did was under-engineer on purpose.