This is a financial & insurance problem we approach through our Platform to Platform service line.
The problem
Ask your risk desk for the firm’s position right now and watch what happens. Someone opens three systems, exports two spreadsheets, and gives you a number that was true at close of business yesterday, adjusted by feel. The actual answer is being assembled overnight, by a reconciliation job, and it will describe a position you held, past tense.
The gap has a precise cost: every intraday decision — take more of this flow, hedge that concentration, suspend this account — is made against a stale number. Limit breaches are discovered by the reconciliation rather than prevented by a person. And a client active across several books or entities is several unrelated records, so the exposures most worth seeing are exactly the ones the per-book view structurally hides.
Why it’s hard
End-of-day sources cannot be summed into a live number. The obvious build — query every system and add the results — inherits each system’s refresh cycle and answers with the staleness of the worst of them. A live view needs the sources to emit events as they happen, and most systems were never asked to.
Aggregation has to happen at write time, and stay correct. Computing exposure at query time across millions of transactions does not return in decision time; maintaining running aggregates raises the harder question of what happens when an event is amended after the fact. Getting live speed and amendment correctness simultaneously is the engineering centre of the whole problem.
Amendments arrive after the number was shown. Settlements are corrected, results are protested, transactions are reversed. A live surface that cannot represent “this figure is provisional and here is why” either lies during the interim or drowns the desk in churn. Amendment handling is a first-class design problem, not an edge case.
Identity is fragmented on purpose. Books, brands and entities were built separately, often for good regulatory reasons, and each holds its own version of the client. Cross-book cohorts — fraud-shaped behaviour, concentrated high-value flow, systematic abuse of promotions — only exist in a joined view that no individual system can produce.
A live number nobody consumes changes nothing. If the exposure figure is a dashboard and the decisions still run on yesterday’s batch, you have bought a faster way to watch breaches happen. The number has to sit in the decision path.
How we approach it
One reporting surface, fed by events. Live money in, exposure and risk position across every book on a single surface, aggregated as transactions arrive. On the platform where we run this, that surface replaced overnight reconciliation as the operational view of risk; the reconciliation remains as the accounting control, which is the job it was always right for.
An identity spine across books. Clients resolve to one identity across brands and entities, which is what makes consolidated exposure a real number and cross-book cohorts computable at all.
Cohort flagging as the position moves. The same surface flags behavioural cohorts as they form — fraud-shaped flow, high-value clients, promotion abusers — so those populations are identifiable while they are acting, rather than discovered in the month-end pack. Flagged is the operative word: the surface identifies, humans and written policy decide.
Decisions wired to the live number. The strongest version of this pattern puts the live position in the transaction path itself. On the wagering platform, a routing engine decides per bet — against current liability — whether the risk is retained in-house, passed to an external host or shared under a business rule. The transferable mechanism is that retain, lay off or split becomes a per-transaction decision computed against live exposure, rather than a periodic rebalancing exercise against a stale one.
Amendments as visible states. Provisional, declared, amended and settled are explicit states on the surface, so the desk sees a moving number with its uncertainty attached instead of a confident number that quietly changes.
What we do not do. We do not supply the risk appetite: limits, hedging policy and intervention thresholds are yours, and we build the surface and the decision hooks they plug into. We do not provide financial or investment advice. And we would not start with the analytics — if the event feeds and the identity spine do not exist, building the dashboard first produces a beautiful view of numbers that are still wrong.
What it takes
Event access to the source systems. Each source must emit transactions and amendments as they occur, or be honestly represented as stale on the surface. Expect one system in your estate to make this the long pole.
An agreed definition of exposure. Gross or net, by what dimension, valued how, refreshed when. This is a political decision dressed as a technical one, and it has to be made by your risk function before the aggregates are built — changing it later means rebuilding them.
One identity, with an owner. Cross-book identity resolution needs a matching policy, a steward, and an answer for the records that will not merge cleanly. It is the largest single work item in most estates and the one with the least glamorous name.
A decision inventory. Which decisions will actually consume the live number — limit enforcement, flow acceptance, hedging triggers, account intervention — and who owns each. This list is what separates a risk surface from a wall-mounted television.
Where this has been done
Delivered — on a wagering platform, not a financial trading book. On the multi-tenant platform we built and operate, one real-time reporting surface shows live money in, exposure and risk position across many tenants, where each tenant previously reported separately and position was reconciled overnight. The same surface flags fraud, VIP and bonus-abuse cohorts as they happen, and a routing engine takes per-bet decisions — retain, pass to an external host, or share — against the live liability. The multi-tenant consolidation and platform case studies cover it. No performance figures from that work are cleared for publication, and none appear here.
A bookmaker’s liability book and a trading book are close relatives: both are live positions built from many small transactions, moved by amendments, concentrated by correlated events, and abused by identifiable cohorts. The mechanisms — event-fed aggregation, one identity, per-transaction decisions against the live number — transfer directly. The financial-markets deployment is the argument this page makes, not a reference it cites.