// CUO & portfolio
Your portfolio view is as fresh as your slowest bordereau.
You answer for loss ratio, appetite, aggregation and rate adequacy, on data that arrives late by construction. That is an architecture problem wearing a reporting problem's clothes, and the fix is the arrival path rather than another dashboard.
// The situation
The layer beneath you cannot ingest automatically.
Only 19% of carriers automatically ingest external data into pricing models, and 47% cannot price optimally because of integration issues between new and legacy systems.
hyperexponential · State of Pricing · undated in our source · vendor research.
If pricing runs on a lag, the portfolio view above it inherits the lag.
So the first project is the arrival path: what lands, when, in what shape, and from whom. That scoping happens before anything is visualized.
// The stack you actually run
N delegated books, N reporting formats.
Every delegated relationship is its own schema. Column names drift between months, valuation dates are inconsistent, and a late file is indistinguishable from a partner who wrote nothing — which is exactly the signal you needed.
Casualty appetite is conservative for a reason: $62 billion of adverse development in commercial liability across the period from 2015 to 2024.
BCG · development years 2015 to 2024 · publication date not stated in our source.
Development that size is written long before it is discovered. What a build actually reaches is how early the book tells you it is moving, and secondary-peril losses have run hot for three consecutive years while reinsurance repriced against record capital.
Your treaty conversation is a buyer’s-market conversation, which moves the question from securing capacity to deploying it profitably. That is a data-freshness problem before it is a strategy problem.
// The build
Four builds, and the first one is unglamorous on purpose.
Bordereaux to portfolio in days, not weeks
If you carry a meaningful delegated book, your portfolio visibility is gated on a monthly or quarterly reporting cycle rather than a data feed. We normalize inbound at receipt, so the aggregate updates continuously and late reporters are visible as late rather than absent.
Surface: the delegated-reporting intake — ACORD AL3 where a reporter supports it, mapped workbooks where they do not. The mappings are yours to own.
Aggregation with shared-dependency detection
Accumulation by peril, region and program, plus the dependencies most portfolios cannot enumerate: cloud region, security vendor, managed service provider, file-transfer product. An inventory first and a model second, because you cannot model a correlation you cannot enumerate.
Surface: an exposure inventory in your warehouse, built from bound-policy records rather than submission data.
Appetite drift monitoring
What you said you would write, against what actually got bound — measured continuously by class, geography, limit and attachment, and surfaced when the drift starts. Appetite drift does not surface in quarterly reviews. It surfaces in loss ratios.
Surface: bound-policy records from the policy system, effective-dated and versioned so an endorsement does not restate history.
Rate adequacy against bound terms
Filed rate is not achieved rate once schedule credits, IRPM and negotiated terms are applied. We reconstruct achieved rate from what was actually bound, by segment, and track it against your indication.
Surface: the rating and bound-terms records, reconciled against the filing.
What we measure
Days from bordereau receipt to aggregate refresh. Share of delegated premium visible within a fixed number of days after month end, on one cohort definition both sides. Enumerated dependencies against modeled ones, counted. Achieved rate change by segment against indication. And drift alerts raised in period against drift discovered at the quarterly review.
// We ran one of these
Appetite drifts and the loss ratio arrives later. Neither is readable until the data underneath the portfolio view is fixed, so that comes first.
The operating record →// What's hard about this
Two things we tell you in week two.
The ingest comes before the dashboard, and that is a duller first project.
A portfolio view assembled from stale, manually keyed inputs is a confident picture of last month. Most requests that arrive as “we need better portfolio reporting” are a data-arrival problem two layers down. So week two names which of the two you have, and the arrival path is scoped and priced first.
The number everyone quotes in this space is marketing.
The widely quoted $100 million to $180 million in annual opportunity cost per $1 billion of premium, from quote-capacity constraints, comes from a vendor with a product to sell. The underlying characterization is plausible; the dollar figure is not evidence. So the calculation gets run on your own submission, quote and bind counts, and the result goes in the report whether or not it supports a build.
Vendor-published · vendor research.
// What ships with it
Responsibility is non-delegable.
New York DFS states it plainly: an insurer cannot rely solely on a third party’s claim of non-discrimination. So the aggregation and drift models leave with a model inventory entry and risk tier, data lineage naming the provenance of every external source and every delegated reporter’s file, pre-deployment testing results, drift thresholds with remediation triggers, and a human-override specification naming who can act on an alert.
Roughly half the states have adopted the NAIC AI Model Bulletin. Connecticut requires an annual AI compliance certification attested by a named officer, and Iowa formally defines bias and outcomes testing. A portfolio model without lineage back to the inbound file is a model that cannot be certified.
NAIC and state bulletins · current at September 2026.
What the governance file contains →Fix the arrival path, then build the view.
A portfolio view is only as current as the slowest file feeding it. The first project is the one that makes every subsequent number worth reading, and it is the one most proposals skip.