Cat Model Input Preparation: Clean Cedent Exposure Data
On this page
- What Does Cat Model Input Preparation Look Like on a Real Schedule?
- What Method Turns a Cedent Schedule Into Model-Ready Input?
- Which Controls Catch Exposure Errors Before the Model Runs?
- Which Edge Cases Need a Modeller's Judgement?
- Does an Open Exposure Standard Help?
- Where Does Insurnest Fit in Cat Model Input Preparation?
- What Should the Head of Cat Modelling Set Up First?
- Key Terms
- Sources
- Frequently Asked Questions
Last reviewed: November 2026
The model run looks alarming, and nobody can tell whether the risk changed or the spreadsheet did.
Half the renewal week then goes on finding a duplicate tab and a column in thousands.
Cat model input preparation cleans a cedent's exposure schedule so the cat model receives accurate values, locations and building codes.
It should finish with totals reconciled to the submission and every assumption written down.
What Does Cat Model Input Preparation Look Like on a Real Schedule?
Cat model input preparation is easiest to understand on a small schedule, so here is a hypothetical five-row example with illustrative figures.
The header of the cedent's file says "Values in USD 000s".
What is wrong with the raw schedule?
The raw schedule contains five common problems, and none of them would stop a model from running.
| Row | Occupancy | City | Value as received | Problem |
|---|---|---|---|---|
| L1 | Warehouse | City A | 12,500 | None, USD 12.5m |
| L2 | Office | City A | 8,000,000 | Pasted in whole USD, not 000s |
| L3 | Factory | City B | 30,000 | Construction "unknown" |
| L4 | Factory | City B | 30,000 | Duplicate of L3 |
| L5 | Retail | City C | 4,000 | Values in EUR 000s, city-level geocode only |
How does the total change step by step?
The total insured value changes sharply as each fix is applied, which is why order and documentation matter.
| Step | Action | Total value (USD m) |
|---|---|---|
| 0 | Sum as received, all rows read as USD 000s | 8,076.5 |
| 1 | Correct L2 to USD 8.0m | 84.5 |
| 2 | Remove duplicate L4 (minus 30.0) | 54.5 |
| 3 | Convert L5 at an illustrative 1.10 USD per EUR (4.0 becomes 4.4) | 54.9 |
Step 0 is 12.5 + 8,000 + 30 + 30 + 4 = 8,076.5. The clean total is 12.5 + 8.0 + 30.0 + 4.4 = 54.9.
A magnitude check catches step 1 immediately. Steps 2 and 3 are quieter and far easier to miss.
What does the cleaning change for accumulation?
The cleaning changes accumulation most in City B, where the duplicate doubled the apparent value.
Before cleaning, City B showed USD 60.0m. After removing L4 it shows USD 30.0m, so the raw file overstated that zone by 100%.
The model would have shown a concentration that does not exist. The opposite error, a missed location, would hide one that does.
What Method Turns a Cedent Schedule Into Model-Ready Input?
The method that turns a cedent schedule into model-ready input generalises the worked example into eight repeatable steps.
- Profile the file. Count rows, read header notes, and list units, currencies and value columns.
- Normalise units and currencies. Convert to one unit and currency, recording the rates used.
- Remove duplicates. Match on address, coordinates and values, and confirm before deleting.
- Improve geocodes. Record the resolution of each location, from rooftop to city level.
- Map construction and occupancy. Translate cedent descriptions into your model's codes.
- Split values by coverage. Separate buildings, contents and business interruption where possible.
- Reconcile to the submission. Compare totals with the slip, the risk profile and the cedent's summary.
- Document assumptions. List every default, code choice and unresolved gap for the modeller.
Why should you reconcile to the submission before modelling?
Reconciliation to the submission before modelling proves that nothing was lost or doubled during cleaning.
If the cleaned total differs from the cedent's stated total, someone must explain the gap before the model runs.
How should unknown construction and occupancy be handled?
Unknown construction and occupancy should be coded as unknown in the model, not guessed, and the share of unknowns should be reported.
A sensitivity run with a cautious assumption then shows the underwriter how much the unknowns could matter.
Which Controls Catch Exposure Errors Before the Model Runs?
The controls that catch exposure errors before the model runs are simple checks applied to every schedule, every time.
| Control | What it catches | Example rule (illustrative) |
|---|---|---|
| Magnitude check | Unit errors | Flag any location above 10x the schedule median |
| Duplicate check | Repeated rows or tabs | Same address and value appears twice |
| Reconciliation | Lost or doubled values | Clean total within 1% of submission total |
| Geocode resolution report | Weak location data | Share of value geocoded at city level or worse |
| Unknown attribute report | Missing building data | Share of value with unknown construction |
| Top locations review | Concentration surprises | Top 10 locations' share of total value |
Set your own thresholds by class and territory. The point is that every check runs before modelling, not after a surprising result.
Which Edge Cases Need a Modeller's Judgement?
Several edge cases in cat model input preparation cannot be settled by rules and need a modeller's judgement.
What if the cedent only provides aggregate exposure?
If the cedent only provides aggregate exposure by zone, the modeller must choose how to distribute values within each zone.
That choice should be recorded, because it can move modelled losses for peril-sensitive areas such as coastlines.
How should business interruption values be treated?
Business interruption values should be kept separate from building and contents values, because models treat them differently.
If the cedent only gives one combined value, record the split assumed and test the sensitivity.
What about values at an old valuation date?
Values at an old valuation date should be flagged, because rebuilding costs may have risen since the schedule was compiled.
Any indexation applied should be stated, with the source and the factor used.
How should blanket or loss limits on multi-location policies be treated?
Blanket or loss limits on multi-location policies should be captured as policy terms, not spread across locations as values.
If the limits are ignored, the model may overstate the insurer's loss at policy level, and so the loss to the treaty.
Does an Open Exposure Standard Help?
An open exposure standard helps by giving cedents, brokers and reinsurers one agreed format for passing exposure into models.
The Open Exposure Data (OED) standard is curated by Oasis LMF and describes itself as model-agnostic.
Artemis reported in January 2023 that Aon was implementing OED for property reinsurance placements.
Aon said it was working with RenaissanceRe, SCOR, Hannover Re and Swiss Re on that adoption.
What does OED not solve?
OED does not solve poor source data. A duplicate row in OED format is still a duplicate row.
The standard reduces conversion work between formats, but the eight-step method above is still needed.
For why consistent inputs matter across vendor models, see our guide to catastrophe model blending.
Where Does Insurnest Fit in Cat Model Input Preparation?
Insurnest fits at the very start of cat model input preparation, where exposure arrives inside broker emails and attachments.
Reinsurance Email Intelligence for submission intake reads the email with its PDFs and spreadsheets.
It extracts risk and exposure details, including segment, policy count, exposure, exclusions and conditions, alongside structure, limits and territory.
These details land in one structured submission view, with missing information surfaced early so it can be chased.
Insurnest does not geocode, code construction or run catastrophe models. Exposure analysts and modellers validate and finish the input.
What Should the Head of Cat Modelling Set Up First?
The head of cat modelling should set up a short, written standard before the next renewal peak.
- Publish the eight-step method and the control table as the team's minimum standard.
- Require a one-page assumptions note with every model-ready file.
- Report the share of unknown attributes and weak geocodes to underwriting with every result.
- Send cedents a specific gap list, not a general request for "better data".
The chief risk officer should see the same quality indicators, because they shape how far modelled accumulations can be trusted.
For how modelled results feed buying decisions, see our guide to how much catastrophe reinsurance to buy.
For the systems behind the models, see our guide to catastrophe modelling infrastructure.
Key Terms
- Total insured value (TIV): The full insured value at a location across buildings, contents and business interruption.
- Geocode resolution: How precisely a location is placed, from rooftop to postcode or city level.
- Construction and occupancy codes: Model codes describing how a building is built and used.
- Accumulation: The total exposed value in one zone that a single event could affect.
- Open Exposure Data (OED): An open, model-agnostic exposure data format curated by Oasis LMF.
- Sensitivity run: A model run with changed assumptions to show how much a result depends on them.
Sources
- Oasis LMF (GitHub): Open Data Standards – Open Exposure Data (OED)
- Artemis: Aon embraces Open Exposure Data Standard (23 January 2023)
- Insurnest: Reinsurance Email Intelligence
Frequently Asked Questions
What is cat model input preparation?
Cat model input preparation is the work of turning a cedent's exposure schedule into clean, model-ready data. It covers unit and currency normalisation, duplicate removal, geocoding, construction and occupancy coding, value splits and reconciliation to the submission, before any catastrophe model is run.
Why does cat model input preparation matter to reinsurers?
Cat model input preparation matters because a catastrophe model can only be as reliable as the exposure it receives. Duplicate locations, unit errors and poor geocodes can overstate or understate accumulations, distorting modelled losses, pricing and capacity decisions for the whole programme.
How do you prepare cedent exposure data for a cat model?
To prepare cedent exposure data for a cat model, profile the schedule, normalise units and currencies, remove duplicates, improve geocodes, map construction and occupancy to model codes, split values by coverage, reconcile totals to the submission, and record every assumption for the modeller and underwriter.
What are the most common cat model input errors?
The most common cat model input errors are values in thousands mixed with whole currency, duplicate locations, geocodes at city or postcode level only, unknown construction or occupancy, missing coverage splits, and values at an old valuation date. Each can shift modelled accumulations materially.
Who owns cat model input preparation?
The head of cat modelling usually owns cat model input preparation standards, while exposure analysts do the cleaning and underwriters chase cedents for missing data. The chief risk officer relies on the result for accumulation and capital views, so ownership and assumptions should be documented.
Can AI automate cat model input preparation?
AI can automate parts of cat model input preparation, such as reading exposure schedules and submission documents, extracting values and flagging gaps. Coding choices, assumptions for unknown fields and the final model-ready file still need review by an exposure analyst or cat modeller.
How does OED compare with a proprietary exposure format?
Open Exposure Data, curated by Oasis LMF, is an open, model-agnostic format for exposure data, while proprietary formats are tied to one vendor's model. An open format can reduce conversions between models, but the underlying cedent data still has to be cleaned first.
Should reinsurers reject poor-quality exposure schedules?
Reinsurers should rarely reject poor-quality exposure schedules outright, but they should price the uncertainty and ask for fixes. A clear list of specific gaps, plus documented assumptions and sensitivity tests, is usually more useful to the cedent and underwriter than a flat refusal.

Hitul Mistry
CEO, Insurnest
An InsurTech leader with more than a decade of experience across insurance and technology, focused on solving business problems with the help of technology. Has worked with brokers, insurance carriers, and reinsurance firms across the India, UAE, and US markets.
View LinkedIn profile →