PLEN

altkom software · data platforms

Why does the same loss ratio come out three different ways?

The board is asking for the fleet loss ratio for the last quarter. The teams responsible for the portfolio, claims and controlling each reach for their own data — and every result can be justified.

64.2%policy admin system
58.9%claims system
61.5%finance spreadsheet
The export will go through. The question is whether it survives an audit. See where the errors — and the discrepancy — come from.

What happens when someone asks: “where does that number come from?”

Reports are built from exports, spreadsheets and manual reconciliations.
A data warehouse, for example, but the numbers still need explaining.

Where does the data start to change the report?

One month, one product. Data from the policy admin system, broker files and documents read via OCR. At first glance the differences look minor; in a report they can shift the result or make it hard to explain.

This sample hides 21 inconsistencies. Click the suspicious cells and see what a typo in a client's name, a different date format, a missing currency or a status saved as an abbreviation can change.

policy_noclient_eidclient_nameinception_dateexpiry_datepremiumcurrencystatustrade_lic
FL/2023/AE001784-1990-1234567-8AL-RASHIDI MOHAMMED2023-01-152024-01-143600.00AEDActive
FL/2023/AE001784-1990-1234567-8Al-Rashidi Mohammed15/01/202314/01/20243,600.00AEDACT
FL2024AE0017841990123Al Rashidi Mohammed15-01-20232024.01.143600,001
FL/2023/AE002784-1975-9876543-2Gulf Logistics LLC2023-02-012024-01-3128500.00AEDactiveDED-12345
FL/2023/AE003784-1985-5678901-3Al-Mansoori Fleet Co.2023-03-102024-03-0945000.00SARActive
FL/2023/AE004784-1980-3456789-4Mohamed Al-Hamdan01/15/202301/14/20245100.00AEDActive
FL/2023/AE005784-1988-6789012-5Muhammad Al-Hamdan2023-04-012024-03-31NULLAEDN/A
FL/2023/AE005784-1988-6789012-5Mohammed Al-Hamdan2023-04-012024-03-315100.00AEDActive
FL/2023/AE006784-1992-7890123-6Emirates Transport2023-05-202024-05-191.950,00AEDActiveADNOC-789
FL/2023/AE007784-1970-8901234-7Al-Maktoum Holdings2023-06-012024-05-31120000.00ActiveDED-99001
FL/2023/AE007784-1970-8901234-7Al-Maktoum Holdings2023-06-012024-05-31120000.00USDActiveDED-99001
FL/2023/AE008-Anonymous Client2023-07-101200.00AEDActive

The good news: data can be cleaned up before it reaches analysis

A shared data platform joins information from policies, claims, broker files and OCR, then checks it against agreed rules. It detects inconsistencies, labels their type and triggers a fix before the differences reach the report.

"AL-RASHIDI MOHAMMED"
"Al-Rashidi Mohammed"
case normalisationone client record
"31/12/2023 / 2023-12-31 / 1445-06-19"
"2023-12-31"
date normalisationa normalised date format
[record A] + [record B]
[golden record, one full client profile]
deduplicationdata for the same client, merged
"Loss ratio: 64.2% / 58.9% / 61.5%"
"Loss ratio: 61.5% (def. v2.1)"
definitions dictionaryagreed rules for the loss ratio
raw datacleaned databusiness datareports

What does your data platform need to carry?

You are starting from zero and that is an advantage. You can pick the architecture that fits your needs from the start. Set the sliders to your situation.

Data warehouse

Clear fit
Fit50%

When reports, audit and numbers you can explain matter most.

It fits less well when your main challenge is raw data, many formats, documents, OCR or fast-moving AI models.

It works well for reporting loss ratios, premium, reserves, portfolio, product profitability and sales results. It helps align definitions, control data quality and show where a result came from — in a board report, an audit or a conversation with the regulator.

Data lake

Fit30%

When you want to use data that does not fit standard reporting today.

Note: without a data catalogue, owners and quality rules it can become yet another place where data has to be explained by hand.

It takes in data from many systems, files, documents, processes and external sources — even when they have different formats and quality levels. It gives room to analyse claims history, segment clients, scoring, anomaly detection, work on documents and prepare AI projects.

Lakehouse

Fit30%

When reporting, audit and AI need to run on the same data foundation.

Note: it requires clear data layers, shared definitions, quality rules, data owners and consistent change management.

It combines the order of a data warehouse with the flexibility of a data lake. It lets you work on raw, trusted and reporting data in one architecture. It fits modernising reporting, data quality control, portfolio analytics, scoring, claims prediction and anomaly detection.

This is not a finished architecture choice, only a first pointer. The result depends on your current systems, data quality, number of sources, regulatory requirements, cost, team skills and analytics roadmap. To pick the right solution you have to translate this result into real assumptions: what you have today, what to clean up first and what decisions the data platform should support in the coming months.

Your business questions choose the architecture. Not fashion.

AI needs to know where its answer comes from, too

If you are thinking about bringing AI into reporting, analytics or work with portfolio data, you first need a solid foundation: tidy sources, shared definitions and quality rules. Without it, AI will not solve the problem of divergent data. It can only reproduce the same chaos faster — in a more convincing form.

definitions dictionary · ai chat

|

The same question, an entirely different route to the answer

What is the fleet loss ratio for the last 4 quarters?

before the rollout5 days3 spreadsheets · 4 people · 3 results
after the data platform10 s1 dashboard · 1 result with a full trail

Your data doesn't have to look like this

45 minutes with a data architect. No sales pitch — we go straight to your situation.

altkom software · data platforms