We build analytics estates with that evidence trail built in: lineage from core banking, policy administration or portfolio systems through to the published report, measure definitions reviewed by the control function, row-level security aligned to segregation of duties, and change control that produces an audit record without anyone writing one.
We work across retail and commercial banking, asset and wealth management, insurance and broking, payments and fintech. The constant is a low tolerance for unexplained variance, and a high value on being able to answer "where did this come from?" in minutes.
Finance, risk and the regulatory reporting team each build their own extract, and the numbers diverge by small amounts nobody can explain. The cause is usually different as-at timing and different treatment of late adjustments, both undocumented.
Month-end reconciliation runs through workbooks owned by individuals, with manual adjustments applied outside any system of record. It works until the person is on leave, and it produces no audit trail.
Client-level and position-level data spread across workspaces with ad-hoc permissions. When asked who could see a given portfolio last quarter, the honest answer takes a week of investigation.
Cost allocation is crude, so profitability by client, product or channel is directional at best. Pricing and relationship decisions then rest on numbers the business privately distrusts.
| Metric | Definition | The trap |
|---|---|---|
| Net interest margin | Net interest income over average earning assets for the period. | Average earning assets must use a consistent averaging basis (daily, monthly or period-end), and mixing bases across reports is the single most common source of unexplained drift. |
| Cost-to-income ratio | Operating expenses over total operating income. | Whether exceptional items, amortisation of intangibles and recharges are in or out must be fixed in the model, not decided per report. |
| Assets under management | Market value of client assets managed, at the measurement date. | Multi-share-class and mandate-level double counting, plus FX translation at the wrong rate date, routinely inflate the number. |
| Combined ratio | Incurred losses plus expenses over earned premium. | Earned versus written premium is the usual error; reinsurance gross versus net needs an explicit, labelled measure for each. |
| Expected credit loss movement | Period change in IFRS 9 expected credit loss provision, split by stage transfer, model change and portfolio movement. | Without the attribution split the number is unexplainable, and attribution has to be modelled deliberately rather than derived later. |
Automated lineage and reconciliation from source systems to the figures submitted in regulatory returns, with variance explanation captured in the platform rather than in email.
Liquidity, concentration, counterparty exposure and limit monitoring with as-at discipline and clear data-currency indicators, so nobody reads a stale figure as live.
Allocation model built with finance, exposing contribution by client, product, channel and relationship manager with the allocation rules visible to the reader.
Complaint root-cause and outcome monitoring, vulnerable-customer indicators, and evidence packs for conduct reporting obligations.
Alert volume, disposition and false-positive analytics to tune detection thresholds, alongside investigator workload and ageing views.
Automated close pack with controlled adjustments, variance commentary captured against the figure, and an audit trail of who changed what.
In all five of our markets, yes, with conditions: documented outsourcing arrangements, exit plans, data location control and demonstrable oversight. Microsoft offers in-region deployment in the UK, Germany, Switzerland and Ireland, which covers most residency constraints. The work is in documenting the arrangement to the standard your regulator expects. We build that documentation alongside the platform rather than afterwards.
Access through directory groups that mirror your control model, row-level security tested from non-privileged accounts, object-level security for sensitive measures, separation between model developers and production publishers through deployment pipelines, and a workspace design where no single role can both change logic and approve its release.
We build the data layer, lineage and analytical reporting around them. We are not an actuarial modelling vendor. The common pattern is actuarial engines producing results that we then land, reconcile, model and report with controlled lineage, variance attribution and disclosure-ready outputs.
Inside your tenant, under your conditional-access policies, with named accounts and no shared credentials. Development uses masked or subset data wherever feasible. Sensitivity labels are applied so that exports remain controlled. All of that is agreed in writing before access is granted.
Services that apply here
A 45-minute working session with an engineer rather than a sales call. You describe the problem; we set out what we would build, in what order, and what it is likely to cost.
Answered within one working day, by the engineer who would do the work