Authors

  • Shreekant Malviya
    Tata Consultancy Services, Plano, Texas, USA

DOI:

https://doi.org/10.37547/tajiir/Volume07Issue07-04

Keywords:

Snowflake Cost Optimization Property & Casualty Insurance Data Workloads Metadata-Driven Cost Control Query Performance Tuning

Abstract

The use of Snowflake as a cloud-native data warehouse has dramatically changed the management of analytics workload for Property and Casualty (P&C) insurers, while simultaneously presenting serious cost governance challenges. The heavy volume of searches, big data retention, and decentralized business intelligence operations are industry-standard procedures that tend to lead to uncontrolled credit usage and overspending on storage. This research introduces a modular five-layer optimization framework focused on property and casualty insurance data, combining workload segmentation, and compute sizing with Snowflake's account usage metadata. The framework is tested and validated using Kaggle’s Insurance Agency Data, representing real-world P&C operations across 17 states. Benchmark queries simulating core insurance workloads were designed using modified TPC-H logic, a standard decision support benchmark that enables realistic performance evaluation under analytical query conditions, achieving up to 82% cost reduction and a 64% reduction in execution time without compromising the results. These results highlight the efficiency of the framework to facilitate proactive and elastic cost control. Future studies can investigate AI-driven query forecasting, scalable warehouse dynamics, and real-time anomaly detection to further advance cloud-native data ecosystem governance.


background image

The American Journal of Interdisciplinary Innovations
and Research

08

https://www.theamericanjournals.com/index.php/tajiir

Investi

Type

Original Research

PAGE NO.

28-43

DOI

10.37547/tajiir/Volume07Issue07-04


















OPEN ACCESS

SUBMITED

28 May 2025

ACCEPTED

16 June 2025

PUBLISHED

07 July 2025

VOLUME

Vol.07 Issue07 2025

CITATION

Shreekant Malviya. (2025). A Five-Layer Framework for Cost Optimization in
Snowflake: Applied to P&C Insurance Workloads. The American Journal of
Interdisciplinary Innovations and Research, 7(07), 28

43.

https://doi.org/10.37547/tajiir/Volume07Issue07-04

COPYRIGHT

© 2025 Original content from this work may be used under the terms
of the creative commons attributes 4.0 License.

A Five-Layer Framework
for Cost Optimization in
Snowflake: Applied to P&C
Insurance Workloads

Shreekant Malviya

Tata Consultancy Services, Plano, Texas, USA


Abstract:

The use of Snowflake as a cloud-native data warehouse
has dramatically changed the management of analytics
workload for Property and Casualty (P&C) insurers,
while

simultaneously

presenting

serious

cost

governance challenges. The heavy volume of searches,
big data retention, and decentralized business
intelligence

operations

are

industry-standard

procedures that tend to lead to uncontrolled credit
usage and overspending on storage. This research
introduces a modular five-layer optimization framework
focused on property and casualty insurance data,
combining workload segmentation, and compute sizing
with Snowflake's account usage metadata. The

framework is tested and validated using Kaggle’s

Insurance Agency Data, representing real-world P&C
operations across 17 states. Benchmark queries
simulating core insurance workloads were designed
using modified TPC-H logic, a standard decision support
benchmark

that

enables

realistic

performance

evaluation under analytical query conditions, achieving
up to 82% cost reduction and a 64% reduction in
execution time without compromising the results. These
results highlight the efficiency of the framework to
facilitate proactive and elastic cost control. Future
studies can investigate AI-driven query forecasting,
scalable warehouse dynamics, and real-time anomaly
detection to further advance cloud-native data
ecosystem governance.

Keywords:

Snowflake Cost Optimization, Property &

Casualty Insurance Data Workloads, Metadata-Driven
Cost Control, Query Performance Tuning


background image

The American Journal of Interdisciplinary Innovations and Research

The American Journal of Interdisciplinary Innovations
and Research

9

https://www.theamericanjournals.com/index.php/tajiir

1.

INTRODUCTION

The transition to cloud data warehousing has
revolutionized big data analysis and management in the
property and casualty insurance industry. Innovative
cloud-native platforms like Snowflake provide elastic
scaling and pay-as-you-go pricing, thereby enhancing
the ease of analytical responsiveness [1]. This transition
introduces

a

new

operational

risk

of

cost

unpredictability to Property & Casualty (P&C) Insurance.
Ad hoc queries, complete data scans, and the sheer
amount of historical claims data can result in increased
compute and storage consumption, especially when big
data cost governance models are not aligned with
workload behavior [2].

Despite Snowflake's wide array of features such as auto-
suspend, materialized views, result caching, and query
telemetry, these features are often applied ineffectively
in insurance scenarios. Past research and industry
practices reveal some structural and technical sources of
insurance data platform cost inefficiency, some of these
include over-provisioned warehouses, underutilized
storage features, redundant business intelligence
queries, and poor resource allocation [3] [4].

Further, FinOps adoption remains limited among most
insurers [5], [6]. Traditional practices like monthly usage
reviews or manual query monitoring are too slow and
rigid to keep up with the fast-changing demands of
insurance data. For example, during events like open
enrollment or catastrophe modeling which requires real
time responsiveness, these outdated governance
methods can either drive up unnecessary costs or
restrict essential workloads at critical moments.

This work presents a new, workload-aware, modular
cost optimization method particularly tailored for
Snowflake deployments within the property and
casualty insurance business. The solution brings
together query telemetry, warehouse isolation, dynamic
scaling, and observability under an integrated method to
manage cost more effectively while maintaining
performance. This work used an insurance data set from
Kaggle [5] and TPC-H-inspired query benchmarks [6] to
simulate baseline and optimized workloads, quantifying
computing efficiency and byte savings. The outcome
provides a predictable approach to domain-aligned cost
management in cloud-native insurance analytics
environments. However, there is limited research on

how Snowflake’s cost control features can be

systematically

adapted

to

insurance-specific

workloads. This study addresses that gap by exploring
how a workload-aware strategy can improve cost
efficiency and performance in Snowflake for P&C
insurance use cases.

2.

LITERATURE REVIEW

2.1

Overview of Snowflake Architecture and Pricing

Snowflake is a new-generation data warehousing cloud
platform that has become popular due to its cloud-
native architecture as well as simplicity. In contrast to
legacy systems, Snowflake is designed for the cloud
and keep storage and computation separate, resulting
in higher flexibility and scalability

Figure 1.

It enables

companies to manage resources autonomously, thus
making it simpler to manage performance as well as
costs [7].

For example, Snowflake architecture accommodates
the ability to auto-suspend and resume so that
compute resources can remain suspended when
inactive and resumed as necessary. This prevents
unnecessary expenses and enhances effectiveness
compared to legacy data systems. It is also designed
with a multi- cluster architecture, which allows
customers not to encounter bottlenecks during peak
periods, hence enhancing performance in teams [8],
[9].

An important aspect is caching at varying levels

query

results, metadata, and in-memory. Such caches help
the system prevent re-computation of the same
queries again and again, saving resources, and time.
Such automation draws Snowflake to enterprises with
lots of data but who desire to reduce infrastructure
maintenance.

The Snowflake pricing is based on usage. Storage is
priced per terabyte per month, and compute is priced
per second based on the warehouse size. The
companies have a number of warehouse sizes, ranging
from X-Small to 6X-Large, to be able to accommodate
their workload. The pricing is transparent with distinct
prices for storage and compute that enable the teams
to know what they are spending. Overall, the price
offered by Snowflake is transparent, where companies
can pick based on their needs [10], [11].


background image

The American Journal of Interdisciplinary Innovations and Research

The American Journal of Interdisciplinary Innovations
and Research

10

https://www.theamericanjournals.com/index.php/tajiir

In general terms, Snowflake's architecture and pricing
model provide flexibility, performance, and simplicity,

and are extremely well-placed in sectors such as banking
and insurance, where performance and price are
paramount.

Figure 1. Snowflake Architecture

2.2

Cost Optimization Techniques in Cloud Data

Warehouses

With more frequent utilization of cloud data platforms,
cost management has become a fundamental
requirement for most businesses. Snowflake, as much as
other platforms such as Redshift and BigQuery, has
mechanisms and processes in place to enable the
mitigation of unwanted expenditures. Getting to know
the proper use of the technologies is essential to
achieving the maximum potential of the platform.

One of the initial measures is computation usage
optimization. Snowflake's auto resume and suspend
capabilities switch off compute resources while they are

not in use. Several companies’ de

-isolate environments,

such as different warehouses for production and
development, to ensure that there's no overlap and
resources are still allocated where they're needed [12].
This measure avoids the error of performing costly
operations in unnecessary contexts and maintaining
computer expenditure in check. Another category is
storage. Snowflakes have columnar formats, such as
Parquet, and have zero-copy cloning. These capabilities
eliminate redundancy in storage as well as decrease
costs in settings where teams need to test or create
production-like data [13], [14].

Companies shift cold data into less expensive forms of
storage, such as S3 Glacier, to lower the cost of long-
term storage. This is particularly useful in industries
such as insurance, where storage of data for several
years is necessary to meet regulations. Query
construction has a significant cost effect.

Badly constructed searches can traverse vast data sets
and consume a lot of processing power. Snowflake, as
well as other platforms, promotes the use of filters early
on, the selection of columns required, and the utilization
of materialized views or result caching. These actions
minimize the level of data that is read and the time it
takes for queries to execute [15], [16]. Monitoring
wasteful or expensive queries regularly can result in
significant savings overtime. This optimization on a
technical level is also complemented by cost control.
Resource monitors in Snowflake enable monitoring
credit usage by a warehouse and can notify usage when
usage rates exceed thresholds that are established in
advance [12].

Teams utilize job schedulers, such as Airflow or dbt
Cloud, to execute large jobs during off-peak periods
when system loads are low. These measures introduce a
non-technical level of control, which is very beneficial in


background image

The American Journal of Interdisciplinary Innovations and Research

The American Journal of Interdisciplinary Innovations
and Research

11

https://www.theamericanjournals.com/index.php/tajiir

maintaining reasonable cloud expenses [17]. Lastly,
developing cost-consciousness in teams is a long-term
benefit. Increasingly, businesses are looking into FinOps
practices that encourage shared responsibility between
finance and engineering. Ascribing workloads or building
dashboards that reflect usage by team or project raises
visibility, of the effect of individual action on cloud
expenditures [18]. Such a cultural shift makes a
significant contribution to sustaining sustainable
operations. Combined, these techniques offer a robust
foundation for optimizing cloud expenditure while
maintaining high performance. Uniformity and
observability are critical in both tools and Teams.

2.3

Cost Challenges in Cloud-Based Data Warehouses

for P&C Insurance

Although cloud solutions like Snowflake have numerous
advantages, insurance businesses have specific
challenges when it comes to keeping costs low. This is
mostly because of the usage and storage of data in

business.

Table 1

summarized the most significant cost

drivers in the P&C workloads, Analysts and departments
within insurance organizations, in most cases repeat
similar

queries

repeatedly,

in

some

cases

uncoordinated. Redundant or inefficient queries can
take up 30

40% of compute load [3]. This not only

increases the cost but also makes the system slower,
particularly when a large number of users are accessing
data simultaneously.

The second problem pertains to storage. Insurance
companies have to retain historical data for many years,
sometimes

decades.

This

encompasses

policy

documents, claims, and histories of transactions.
Snowflakes incur a cost of approximately $23 per
terabyte per month, which quickly adds up when
retaining data for multiple years. Most businesses lack
adequate procedures for archiving or purging data, so
they end up with wasteful storage costs that may be
sitting idle forever [19].

Table 1.

Summary of Key Cost Drivers and Their Operational Impact in P&C Insurance Cloud Data Warehouses

Cost Issue

Quantitative Impact

Operational Effect

Redundant BI

Queries

30

40% of compute spends

Increased costs, slower reporting, elevated

risk

Large Historical

Storage

$23/TB/month; $100k

$500k+

annually

Storage inflation, maintenance overhead

Data Redundancy

Up to 30% of IT budget

Complexity, inefficiency, and decision-

making delays

Another challenge is that most firms don't segment or
classify work by department. The calculations utilized by
the underwriting employees might not be distinct from
those which are used by the

claim’s

employees. Without

this demarcation, it is hard to recognize which
department should manage higher costs [18]. The
absence of visibility hinders proactive action and tends
to result in a reaction too late, due to factors such as
analyzing consumption by itself when the monthly
invoice is too large. Real-time cost monitoring is not
common. Most insurers continue to use quarterly check-
ups or ad hoc audits. The insurance sector is highly
reactive; unexpected events like natural disasters can
trigger a massive peak in data activity. Without real-time
notification and automation, these spikes can result in
shocking overages. The intricacy of insurance data

pipelines make it worse. Data tends to get pulled from
various

systems

and

undergoes

numerous

transformations. Without proper monitoring, minor
inefficiencies can translate into major money issues.
Despite increased awareness of FinOps, most insurance
companies continue to act in silos, and it becomes
challenging to have cost ownership that's collaborative.

Although there are recent studies that have examined
methods to reduce cloud-based query expenses, the
majority do not fulfill the specific needs of property and
liability (P&C) insurance. One study proposed an
approach to optimize cost and speed for cloud native
queries; however, it is not customized to address the
particular regulations and challenges within the
insurance sector [20]. Another study proposed a


background image

The American Journal of Interdisciplinary Innovations and Research

The American Journal of Interdisciplinary Innovations
and Research

12

https://www.theamericanjournals.com/index.php/tajiir

technique to minimize cloud query expenses, primarily
concentrating on infrastructure configurations and
neglecting

platform-specific

functionalities

like

Snowflake’s caching and warehouse isolation [21].

Additional work examined the role of FinOps in cloud
cost management [22]; however, a practical guide for
its application in actual insurance data systems is
lacking. This indicates the necessity for a comprehensive
framework that aligns with the data and cost challenges
in property and casualty insurance.

3.

PROPOSED SOLUTION FRAMEWORK

This paper proposes a layered cost optimization
framework divided into 5 layers tailored to the
operational dynamics of Snowflake deployments within
Property and Casualty (P&C) insurance environments.
The aim is to provide a solution that is scalable and
workload-aware, that aligns data platform efficiency
with business imperatives such as claims processing,
underwriting,

fraud

analytics,

and

regulatory

compliance.

Figure 2.

5-layer high level architecture overview diagram

An architecture of a framework in

Figure 2

illustrates

intended for Snowflake cost optimization for property
and casualty insurance workloads. The framework
begins with segmentation of P&C workloads, isolating
important functionalities like claims, underwriting &
pricing into dedicated compute layers. Subsequent
layers ensure query efficiency, storage and movement,
and system observability, each leveraging Snowflake
metadata to drive automation and cost control. The
layered design ensures scalability, governance, and
alignment with insurance-specific analytics demands.

3.1

Five-layered proposed solutions

Layer 1: Workload Segmentation

This layer segments the workloads, and the loads are
segmented by functional goals and time-based
implementation patterns. Allocation of a virtual
warehouse on a dedicated basis for a specific domain
guarantees segregation and scalability. Segmentation is
utilized with workload-specific controls such as auto-
suspend/resume settings and resource monitors and
subsequently lowers idle computing costs.


background image

The American Journal of Interdisciplinary Innovations and Research

The American Journal of Interdisciplinary Innovations
and Research

13

https://www.theamericanjournals.com/index.php/tajiir

Layer 2: Compute Optimization

The compute optimization layer focuses on historical
performance metrics from Snowflake metadata, like
QUERY_LOAD_PERCENT, EXECUTION_TIME, and
CREDITS_USED_COMPUTE, to help with decisions like
expanding the warehouse and ensure that the
computing resources are of the right size. After
analyzing the metrics, multi-clustering is only turned on
for loads that need a lot of parallel processing, like fraud
detection, and single-threaded loads can be allocated to
small compute.

Layer 3: Query Optimization and Caching

This layer solely emphasizes the performance tuning of
queries using techniques that reduce BYTES_SCANNED
and increase PERCENTAGE_SCANNED_FROM_CACHE. In
this layer, recursive queries are tuned using materialized
views and result caching. Query telemetry is monitored
cautiously to detect performance bottlenecks, paying
close attention to inefficient scanning and poor data

trimming.

Layer 4: Storage and Data Movement Optimization

This layer focuses on optimization of storage and data
migration. In this layer, long tail and historical data are
relocated to low-cost storage tiers. Data egress and
stage access fees are alleviated through the
management of OUTBOUND_DATA_TRANSFER_BYTES
and optimizing external file interactions. Memory-
intensive operations like spill events and unloads are
monitored and reorganized as needed.

Layer 5: Observability and Governance

An integrated observability layer consolidates telemetry
for real-time performance and cost measurement.
Departmental cost allocation is established with
metadata tagging, and credit consumption anomalies
are identified based on thresholds. External function
calls and high-impact queries are constantly monitored
to ensure ongoing optimization.

Figure 3. Detailed 5-layer Snowflake Cost Optimization Framework for P&C Insurance


background image

The American Journal of Interdisciplinary Innovations and Research

The American Journal of Interdisciplinary Innovations
and Research

14

https://www.theamericanjournals.com/index.php/tajiir

The layered framework in

Figure 3

. visually maps out

how compute, query, and storage efficiencies align
with governance in Snowflake for insurance workloads.
Feedback loops enable continuous cost control through
telemetry-driven insights.

Snowflake offers access to the key metadata fields for
monitoring and performance evaluation.

Table 2.

Maps

key Snowflake metadata fields to their respective roles
in the proposed five-layer optimization framework. Each

field provides granular insight into query behavior,
compute usage, storage patterns, or network activity.
These metrics inform targeted actions such as
warehouse

right-sizing,

query

tuning,

cache

optimization, and data lifecycle management. By
aligning field-level telemetry with each optimization
layer, the framework enables evidence-based cost
control. This structured mapping also supports
automation and ongoing performance monitoring
across P&C insurance workloads.

Table 2.

Mapping of Snowflake Metadata Fields to Optimization Layers

Snowflake Field

Used In
Layer

Purpose in Optimization

CREDITS_USED

Layer 2 ,
Layer 5

Overall credit consumption per query/workload,
basis for cost attribution and monitoring.

CREDITS_USED_COMPUTE

Layer 2

Measures compute usage for right-sizing and scaling
warehouses.

CREDITS_USED_CLOUD_SERVICES

Layer 5

Tracks cloud service charges; useful in optimizing
metadata operations and automation.

EXECUTION_TIME

Layer 2 ,
Layer 5

Identifies slow queries; helps size warehouses
appropriately.

QUERY_LOAD_PERCENT

Layer 2

Assesses resource utilization for concurrency tuning
and load balancing.

QUEUED_PROVISIONING_TIME

Layer 2

Indicates warehouse provisioning delays; informs
warehouse scaling decisions.

QUEUED_REPAIR_TIME

Layer 2

Flags infrastructure recovery delays may indicate
system-level inefficiencies.

QUEUED_OVERLOAD_TIME

Layer 2

Captures overload bottlenecks; suggests need for
multi-cluster or query optimization.

BYTES_SCANNED

Layer 3

Core metric for evaluating query scan efficiency;
high values trigger pruning strategies.

PERCENTAGE_SCANNED_FROM_CA
CHE

Layer 3

Measures cache effectiveness; used to assess result
reuse opportunities.

BYTES_WRITTEN

Layer 3

Identifies heavy write operations; relevant for
assessing data movement cost.

BYTES_WRITTEN_TO_RESULT

Layer 3

Indicates result set size; can inform result set
optimization or compression.

BYTES_READ_FROM_RESULT

Layer 3

Reflects cache re-use behavior; high values =
optimized performance.


background image

The American Journal of Interdisciplinary Innovations and Research

The American Journal of Interdisciplinary Innovations
and Research

15

https://www.theamericanjournals.com/index.php/tajiir

ROWS_PRODUCED

Layer 3

Helps evaluate result yield vs. scan size; inefficient
queries can be restructured.

BYTES_SPILLED_TO_LOCAL_STORA
GE

Layer 4

Indicates local disk spill; signals under-provisioned
compute or inefficient query.

BYTES_SPILLED_TO_REMOTE_STOR
AGE

Layer 4

Highlights of remote spill events can slow queries
and increase costs.

BYTES_SENT_OVER_THE_NETWOR
K

Layer 4

Captures inter-node communication cost; reduced
by optimizing joins/distributions.

ROWS_INSERTED,
ROWS_UPDATED, ROWS_DELETED

Layer 4

Flags frequent DML operations; supports lifecycle
management and data tiering.

BYTES_UNLOADED

Layer 4

Used to monitor unload activity; excessive
unloading may need redesign.

OUTBOUND_DATA_TRANSFER_BYT
ES

Layer 4

Tracks cross-region/cloud data transfer; high values
prompt architecture review.

INBOUND_DATA_TRANSFER_BYTES

Layer 4

Similar to the above, informs ingress costs and data
loading efficiency.

LIST_EXTERNAL_FILES_TIME

Layer 4

Measures time listing files in external stages;
optimized by partitioned/exact paths.

IS_CLIENT_GENERATED_STATEMEN
T

Layer 5

Identifies BI tool-generated queries; used to
analyze/report tool inefficiencies.

TRANSACTION_BLOCKED_TYPE

Layer 5

Reveals

lock/contention

issues;

helps

tune

concurrency and transaction design.

EXTERNAL_FUNCTION_TOTAL_* (all
fields)

Layer 5

Used to monitor usage, latency, and data volume of
external UDFs; cost driver if misused.

TOTAL_ELAPSED_TIME

Layer 2 ,
Layer 3

Combined metric to flag long-running queries and
analyze performance trends.

PARTITIONS_SCANNED,
PARTITIONS_TOTAL

Layer 3

Indicates pruning effectiveness; informs clustering
and filtering strategies.

BYTES_DELETED

Layer 4

Tracks data removal efficiency; supports data
lifecycle and tiered storage planning.

RELEASE_VERSION

Layer 5

Used for compatibility and audit tracking; less
directly linked to cost but relevant.

4.

IMPLEMENTATION CONSIDERATIONS

The envisioned five-layer Snowflake cost optimization
architecture has to be implemented by way of strategic
planning, metadata instrumentation, automation, and
P&C insurance workload-specific performance

monitoring. Each of the layers of optimization relies on
the level to which Snowflake capabilities are mapped to
insurance industry-specific business activities, including
underwriting, claims handling, and regulatory reporting.

Tools and Instrumentation.

The first stage in


background image

The American Journal of Interdisciplinary Innovations and Research

The American Journal of Interdisciplinary Innovations
and Research

16

https://www.theamericanjournals.com/index.php/tajiir

implementation begins with workload segmentation,
enabled through analysis of metadata fields such as
CREDITS_USED, IS_CLIENT_GENERATED_STATEMENT,
and QUERY_LOAD_PERCENT, which are available in the
Snowflake Account Usage schema, facilitating isolation
and classification of workloads. Dedicated Virtual
Warehouses (VWHs) are provisioned for core functions,
each configured with auto-suspend and auto-resume
policies to eliminate idle compute costs. Resource
monitoring is essential for setting credit thresholds and
alerting stakeholders to anomalous spend behavior.

Compute and Query Optimization.

The second phase in

implementation is right-sizing and scaling of virtual
warehouses

driven

by

metrics

such

as

EXECUTION_TIME, QUEUED_PROVISIONING_TIME, and
CREDITS_USED_COMPUTE. Visualization tools such as
Looker or Tableau can provide surface-level compute
trends and usage patterns. For query optimization, dbt
(data build tool) can be utilized for complex
transformation logics to test for high-cost queries and
enforce linting standards. SQL checks should detect
excessive BYTES_SCANNED, missing filters, or ineffective
joins, while automated alerts can flag regression in

cache

utilization

(e.g.,

drops

in

PERCENTAGE_SCANNED_FROM_CACHE).

Storage and Data Movement.

The third stage handles

the Cold data, such as historical claims or policy records,
which can be archived to cost-efficient tiers. Monitoring
fields such as OUTBOUND_DATA_TRANSFER_BYTES,
BYTES_UNLOADED, and LIST_EXTERNAL_FILES_TIME is
critical for reducing inter-region egress charges and
optimizing external stage interactions. Compression
formats like Parquet can further reduce I/O costs.

Automation and Governance.

The final phase involves

an observability layer that should incorporate tagging
logic for departmental attribution, scheduled cost
reviews, and automated job audits. Integration with
orchestration tools like Airflow or dbt Cloud enables the
scheduling of metadata scans and optimization tests.
Additionally,

external

function

activity

(e.g.,

EXTERNAL_FUNCTION_TOTAL_INVOCATIONS) must be
monitored for cost and performance viability.

To evaluate effectiveness, organizations should track
KPIs shown in

Figure 4

.

Figure 4

. Key Performance Indicators (KPIs) for effective optimization evaluation

The KPIs were selected to directly represent
Snowflake's cost determinants and were picked for
their significance to essential P&C operations. They
assess efficiency in computation, caching, storage, and
concurrency, facilitating precise optimization. Each KPI

is associated with a particular metadata field and
corresponds to the priorities of real-world insurance
analytic.


background image

The American Journal of Interdisciplinary Innovations and Research

The American Journal of Interdisciplinary Innovations
and Research

17

https://www.theamericanjournals.com/index.php/tajiir

A dedicated FinOps or cloud analytics team should own
this continuous improvement cycle, ensuring the
framework evolves alongside the organization's data
strategy. Establishing such a governance loop ensures
both immediate cost savings and long-term platform
sustainability.

5.

EVALUATION

To assess the effectiveness of the proposed five-layer

Snowflake cost optimization framework, we executed a
series of benchmark queries

Table 3

modeled after TPC-

H logic and tailored for Property and Casualty (P&C)
insurance analytics. These queries were run on the

“Insurance

Agency

Dataset”

(Kaggle) [5], which contains

over 213,000 records and 49 columns detailing agency-
level insurance indicators across U.S. states from 2005
to 2015.

Table 3:

Description of Benchmark Queries and Optimization Objectives in P&C Insurance Context

Query

ID

Description

Key Fields Used

Layer(s)

Optimized

Business

Objective

Q1

Total Premium by

State & Product Line

STATE_ABBR, PROD_LINE,

WRTN_PREM_AMT

Layers 1, 2,

3

Market

segmentation

Q2

Top 10 Agencies by

Loss Ratio (2015)

AGENCY_ID, LOSS_RATIO,

STAT_PROFILE_DATE_YEAR

Layers 2, 3,

4

Risk

assessment

Q3

3-Year Average Loss

Ratio by State

LOSS_RATIO_3YR, STATE_ABBR

Layers 1, 2,

3

Performance

benchmarking

Q4

Policy Growth by

Vendor

VENDOR, POLY_INFORCE_QTY

Layers 2, 3

Sales analysis

Q5

Retention Ratio by

Agency and Year

AGENCY_ID, RETENTION_RATIO

Layers 1, 2,

5

Customer

loyalty

Each query was executed under two conditions: a
baseline configuration using a SMALL warehouse
without optimizations, and an optimized configuration
applying framework-specific techniques such as

compute right-sizing and query tuning. The queries
simulated realistic insurance workloads, including
premium aggregations, loss ratio evaluations, and policy
performance tracking.

Table 4:

Snowflake query performance metadata across 5 P&C Workload queries

Query

Metric

Baseline

Optimized

% Change

Q1

BYTES_SCANNED

1720400

1720400

0.00%

Q1

EXECUTION_TIME

504

384

-23.81%

Q1

ROWS_PRODUCED

12

12

0.00%

Q2

BYTES_SCANNED

1343920

1343920

0.00%

Q2

EXECUTION_TIME

314

113

-64.01%

Q2

ROWS_PRODUCED

10

10

0.00%

Q3

BYTES_SCANNED

606408

606408

0.00%


background image

The American Journal of Interdisciplinary Innovations and Research

The American Journal of Interdisciplinary Innovations
and Research

18

https://www.theamericanjournals.com/index.php/tajiir

Q3

EXECUTION_TIME

205

229

11.71%

Q3

ROWS_PRODUCED

6

6

0.00%

Q4

BYTES_SCANNED

770312

770312

0.00%

Q4

EXECUTION_TIME

273

123

-54.95%

Q4

ROWS_PRODUCED

10

10

0.00%

Q5

BYTES_SCANNED

1032424

1032424

0.00%

Q5

EXECUTION_TIME

125

100

-20.00%

Q5

ROWS_PRODUCED

1623

1623

0.00%

Tables 4 and 5

show Performance metrics

CREDITS_USED, EXECUTION_TIME, BYTES_SCANNED,
and

ROWS_PRODUCED

were

collected

using

Snowflake’s

ACCOUNT_USAGE.QUERY_HISTORY view.

QUERY_TAG was applied to systematically distinguish
between baseline and optimized executions. Execution
environments were isolated using a dedicated
warehouse (CLAIMS_WH) with auto suspend enabled to
avoid idle credit consumption.

Table 5:

Warehouse allocation and estimated credit usage (baseline vs optimized)

Query

Warehouse

(Baseline)

Warehouse

(Optimized)

Estimated

Credits

Baseline

Estimated

Credits

Optimized

% Credit Change

Q1

SMALL

XSMALL

0.00028

0.000107

-61.90%

Q2

SMALL

XSMALL

0.000174

0.000031

-82.01%

Q3

SMALL

XSMALL

0.000114

0.000064

-44.15%

Q4

SMALL

XSMALL

0.000152

0.000034

-77.47%

Q5

SMALL

XSMALL

0.000069

0.000028

-60.00%

This evaluation was conducted using the Free trial
version of

Snowflake’s Enterprise Edition

, which

provides access to a limited range of monitoring
capabilities. Several critical metadata fields

such as

detailed credit attribution, query caching, and
warehouse-level usage tracking

are not available in

the Free Trial version. Therefore, for accurate
performance evaluation and cost governance in real-
world scenarios, a properly configured Enterprise
environment with access to organizational production
data and metadata is essential.

5.1

RESULTS

The execution performance improved significantly
across most benchmark queries after applying the

optimization framework. Queries Q1, Q2, Q4, and Q5
experienced reductions in execution time, with Q2
showing the highest drop of 64.01%. Only Q3 showed a
slight increase, likely due to fluctuations in execution,
related to background warehouse congestion or result
cache behavior. Due to varying filter pushdowns or cold
cache conditions, the query may have had fewer caching
benefits because it uses a multi-year average. In order
to smooth such anomalies, future iterations should use
repeated trials or account for warehouse loads during
benchmarking.


background image

The American Journal of Interdisciplinary Innovations and Research

The American Journal of Interdisciplinary Innovations
and Research

19

https://www.theamericanjournals.com/index.php/tajiir

Figure 5. Query Wise execution time (Baseline vs optimized)

Addition

to

the

speed

improvements,

the

BYTES_SCANNED remained constant across both
baseline and optimized runs. This consistency suggests
that performance gains were primarily driven by
compute and cache enhancements rather than changes
in data volume or pruning strategies.

Credit usage saw the most substantial impact. Moving
from a SMALL to XSMALL warehouse resulted in credit
savings between 44.15% and 82.01% across queries.
These results in

Figure 5

directly align with Layer 2 of the

framework Compute Optimization and demonstrate the
benefits of right-sizing warehouses.

Figure 6. Query Wise credit usage: Baseline vs optimized


background image

The American Journal of Interdisciplinary Innovations and Research

The American Journal of Interdisciplinary Innovations
and Research

20

https://www.theamericanjournals.com/index.php/tajiir

To enhance traceability and clarify how each layer of the
proposed framework contributed to measurable

improvements,

Table 6

summarizes the observed

impact of each layer on query performance and cost
efficiency.

Table 6. Layer wise impact on Performance Metrics

Layer

Optimization Focus

Key Metrics

Impacted

Observed Result

Framework

Contribution

Layer 1:

Workload

Segmentation

Warehouse isolation

based on business

domains (e.g.,

claims, underwriting)

N/A

(foundational)

Enabled dedicated

execution and

monitoring

Supports cost

attribution and

resource control

Layer 2:

Compute

Optimization

Right-sizing

warehouses (SMALL

XSMALL)

Credits Used,

Execution Time

44%

82% credit

savings, significant

time reduction (Q1,

Q2, Q4, Q5)

Direct reduction in

compute cost

Layer 3: Query

Optimization and

Caching

Query tuning,

materialized views,

result cache

Execution Time,

Cache Utilization

(indirect)

Improved runtime for

Q2, Q4; consistent

BYTES_SCANNED

Boosted compute
efficiency without

changing data

volume

Layer 4: Storage

& Data

Movement

Reducing spill

events, monitoring

external file usage

Not directly

visible in small

dataset

No observed storage-

related bottlenecks

Structural

readiness for

larger production

datasets

Layer 5:

Observability &

Governance

Tagging, telemetry

review, anomaly

detection

Credit

Attribution,
Query Tags

Enabled tracking of

baseline vs optimized

runs

Facilitated

visibility and

structured cost

governance

Finally, the ROWS_PRODUCED metric remained
unchanged for all queries, affirming that output integrity
was preserved. This confirms that the optimization
strategy maintains data quality while achieving
improved performance and cost-efficiency, validating its
application in real-world insurance workloads. The
improvements in execution time and credit usage
highlighted in

Figures 5 and 6

show the effectiveness of

the proposed optimization framework. Specifically,
warehouse right-sizing (Layer 2) and runtime tuning
(Layer 3) delivered measurable gains in performance
and cost savings without compromising data accuracy.
These results support the adoption of a multi-layer
Snowflake optimization strategy for P&C insurance
workloads.

6.

DISCUSSION

These results confirm that focused Snowflake
optimization can greatly improve the cost efficiency of
analytics in P&C insurance settings. In areas like
insurance, where data architectures are complex, big,
and experience a broad variety of analytical workloads
across underwriting, claims, pricing, and fraud
detection, the simulation demonstrates that a one-size-
fits-all approach to data warehousing grows increasingly
ineffective.

Workload

isolation,

warehouse

optimization, query optimization, storage management,
and observability are all components of our tiered
approach, which demonstrates how modular and elastic
architecture can decouple cost centers and enhance
productivity without compromising on analytical agility


background image

The American Journal of Interdisciplinary Innovations and Research

The American Journal of Interdisciplinary Innovations
and Research

21

https://www.theamericanjournals.com/index.php/tajiir

or data availability.

This framework can bring FinOps principles together
with Snowflake-native capabilities in a way that
translates directly into business complexity, making it
genuinely "next-gen" for cloud-native insurance IT. It
enables insurers to enforce fine-grained cost
governance dynamically

at the query, warehouse, or

product line level

versus static cost controls (such as

pre-defined budgets, manual query policing, or user-
level timeouts). Organizations can move from reactive
cost containment to proactive, intelligent cost shaping
with features like QUERY_TAG, result caching,
clustering, and auto-suspend policies

without the

need

for

continuous

human

intervention

or

performance degradation. The solution provides
operational precision and business alignment over
conventional cost management approaches, which tend
to rely on user discipline or blanket spend limits.

It provides secure experimentation within boundaries
rather than limiting users or stifling innovation. Since
backend technologies guarantee that searches are
bounded, credits are assigned, and optimization
opportunities are surfaced automatically, analysts can
model risk continuously or tinker with novel pricing
models. With multi-cluster warehouses, it is also capable
of handling scale spikes without incurring long-tail idle
compute costs. There are, however, trade-offs with this
strategy. Under-scaling for high workload scenarios,
such as catastrophe modeling, where compute demand
can spike unexpectedly, is a risk with this model.

Time-sensitive simulations can time out or queue
excessively when warehouses are set for cost
aggressively. Further, although materialized views and
result in caching lower costs, they also have a
maintenance cost and the potential for stale results if
not properly managed. In spite of these edge cases, the
simulation provides the foundation for the future by
finding equilibrium between context and cost, agility
and control, and thereby positions it as a sustainable and
feasible model for cloud-native insurance data systems.

7.

CONCLUSION

This study emphasizes the need for proactive cost
optimization and management approach for cloud-
based Insurance data warehousing. As the property and
casualty insurance industry moves more towards
cloud-native applications like Snowflake, balancing

analytic agility and cost management has become a
matter of business necessity. The proposed layered
framework addresses this by aligning compute
resources, query behavior, storage policies, and
observability practices with a variety of workload
characteristics. Simulated benchmarks using insurance
data patterns demonstrate measurable optimization in
both execution time and credit usage, supporting the

framework’s practical value.

The layered approach also opens the door for AI-driven
enhancements in the future. Predictive modeling of
query patterns, intelligent warehouse scaling, and
adaptive caching policies can help automate cost control
while maintaining performance. For example, telemetry
data can inform real-time decisions on warehouse sizing
or suggest optimizations before resource bottlenecks
occur. These capabilities can complement traditional
FinOps practices with a more adaptive and data-
informed model of governance.

That said, the proposed solution has limitations

.

The

evaluation was conducted using a free trial version of

Snowflake’s

Enterprise Edition, which restricts access to

critical metadata fields such as query cache hit ratios,
clustering metrics, and warehouse-level performance
statistics. Additionally, the use of a publicly available
dataset, while helpful for demonstrating structural
concepts, does not capture the complexity and
variability of real-world insurance operations. These
constraints may limit the direct applicability of results to
production environments, highlighting the need for
further validation.

Further research could explore how AI and automation
tools can be operationalized within this structure and
evaluated for long-term scalability and business impact.
Overall, the framework provides a practical foundation
for the insurance industry seeking to improve cost
transparency and efficiency in their data platforms.

REFERENCES

1.

Cyber Security Senior Data Analyst, Department of
Cyber Security, Truist Financial, CA, USA and D. Kodi,

“Performance and Cost Efficiency of Snowflake on
AWS Cloud for Big Data Workloads,”

Int. J. Innov.

Res. Comput. Commun. Eng.

, vol. 12, no. 06, Jun.

2024, doi: 10.15680/IJIRCCE.2023.1206002.


background image

The American Journal of Interdisciplinary Innovations and Research

The American Journal of Interdisciplinary Innovations
and Research

22

https://www.theamericanjournals.com/index.php/tajiir

2.

D. Mazumdar, J. Hughes, and J. Onofre, “The Data
Lakehouse: Data Warehousing and More,” 2023,

arXiv

. doi: 10.48550/ARXIV.2310.08697.

3.

“The

Cost of

Redundancy.”

Accessed: Jun. 15, 2025.

[Online].

Available:

https://www.highwing.io/insights/the-cost-of-
redundancy

4.

A. Pimpley

et al.

, “Optimal Resource Allocation for

Serverless Queries,” Jul. 19, 2021,

arXiv

:

arXiv:2107.08594. doi: 10.48550/arXiv.2107.08594.

5.

“Insurance

Data.”

Accessed: Jun. 15, 2025. [Online].

Available:
https://www.kaggle.com/datasets/moneystore/ag
encyperformance

6.

“TPC

-H

Homepage.”

Accessed: Jun. 15, 2025.

[Online]. Available: https://www.tpc.org/tpch/

7.

K. Allam,

“Cloud

Data Warehousing: How Snowflake

Is Transforming Big Data

Management”.

8.

“Multi

-cluster

warehouses

|

Snowflake

Documentation.” Accessed: Jun. 15, 2025. [Online].

Available:

https://docs.snowflake.com/en/user-

guide/warehouses-
multicluster?utm_source=chatgpt.com

9.

“(5) Snowflake’s Multi

-Cluster Shared Data

Architecture: Scalability, Performance & Cost

Optimization | LinkedIn.” Accessed: Jun. 15, 2025.

[Online].

Available:

https://www.linkedin.com/pulse/snowflakes-multi-
cluster-shared-data-architecture-scalability-anuj-r--
nbi9f/

10.

D. A. S. George, “Deciphering the Path to Cost

Efficiency and Sustainability in the Snowflake

Environment,”

Partn. Univers. Int. Innov. J. PUIIJ

, vol.

01, no. 04, pp. 231

250, Aug. 2023, doi:

10.5281/zenodo.8282654.

11.

D. Seenivasan, “OPTIMIZING CLOUD DATA
WAREHOUSING: A DEEP DIVE INTO SNOWFLAKE’S

ARCHITECTURE AND

PERFORMANCE,”

Mar. 31,

2021,

Social Science Research Network, Rochester,

NY

: 5148190. doi: 10.2139/ssrn.5148190.

12.

“Snowflake

Documentation.”

Accessed: Jun. 15,

2025.

[Online].

Available:

https://docs.snowflake.com/

13.

X. Zeng, Y. Hui, J. Shen, A. Pavlo, W. McKinney, and

H. Zhang, “An Empirical Evaluation of Columnar
Storage

Formats,”

Nov.

07,

2023,

arXiv

:

arXiv:2304.05028. doi: 10.48550/arXiv.2304.05028.

14.

T. Koreeda, H. Honda, and J. Onami, “Snowflake

Data Warehouse for Large-Scale and Diverse

Biological Data Management and Analysis,”

Genes

,

vol. 16, no. 1, Art. no. 1, Jan. 2025, doi:
10.3390/genes16010034.

15.

D.

M.

Compagnoni,

“Optimize

Snowflake

performance and reduce credit

usage,”

Nimbus

Intelligence. Accessed: Jun. 15, 2025. [Online].
Available:
https://nimbusintelligence.com/2024/10/5-ways-
to-optimize-snowflake-performance-and-reduce-
credit-usage/

16.

“Fundamentals of Snowflake Query Design &
Optimization | Keebo.” Accessed: Jun. 15, 2025.

[Online].

Available:

https://keebo.ai/2024/10/29/fundamentals-of-
snowflake-query-design-optimization/

17.

JayaAnanth, “Part 2

- Orchestrating Snowflake Data

Transformations with DBT on Amazon ECS through
Apache

Airflow,”

JayaAnanth. Accessed: Jun. 15,

2025.

[Online].

Available:

https://jayaananthdevops.github.io/posts/snowflak
e_dbt_ecs_part2/

18.

“FinOps

Principles.”

Accessed: Jun. 15, 2025.

[Online].

Available:

https://www.finops.org/framework/principles/

19.

“Understanding

Data Warehouse Cost & Pricing

Models | Rivery.” Accessed: Jun. 15,

2025. [Online].

Available:

https://rivery.io/data-learning-

center/data-warehouse-costs/

20.

C. Wang, Z. Arani, L. Gruenwald, and L. d’Orazio,
“Adaptive Time, Monetary Cost Aware Query
Optimization on Cloud Database Systems,” in

2018

IEEE International Conference on Big Data (Big
Data)

, Seattle, WA, USA: IEEE, Dec. 2018, pp. 3374

3382. doi: 10.1109/BigData.2018.8622401.

21.

V. Leis and M. Kuschewski, “Towards cost

-optimal

query processing in the cloud,”

Proc. VLDB Endow.

,

vol. 14, no. 9, pp. 1606

1612, May 2021, doi:

10.14778/3461535.3461549.


background image

The American Journal of Interdisciplinary Innovations and Research

The American Journal of Interdisciplinary Innovations
and Research

23

https://www.theamericanjournals.com/index.php/tajiir

22.

P. Bhardwaj, “The Role of FinOps in Large

-Scale

Cloud Cost Optimization,”

INTERANTIONAL J. Sci.

Res. Eng. Manag.

, vol. 09, no. 01, pp. 1

5, Jan. 2025,

doi: 10.55041/IJSREM28086.

References

Cyber Security Senior Data Analyst, Department of Cyber Security, Truist Financial, CA, USA and D. Kodi, “Performance and Cost Efficiency of Snowflake on AWS Cloud for Big Data Workloads,” Int. J. Innov. Res. Comput. Commun. Eng., vol. 12, no. 06, Jun. 2024, doi: 10.15680/IJIRCCE.2023.1206002.

D. Mazumdar, J. Hughes, and J. Onofre, “The Data Lakehouse: Data Warehousing and More,” 2023, arXiv. doi: 10.48550/ARXIV.2310.08697.

“The Cost of Redundancy.” Accessed: Jun. 15, 2025. [Online]. Available: https://www.highwing.io/insights/the-cost-of-redundancy

A. Pimpley et al., “Optimal Resource Allocation for Serverless Queries,” Jul. 19, 2021, arXiv: arXiv:2107.08594. doi: 10.48550/arXiv.2107.08594.

“Insurance Data.” Accessed: Jun. 15, 2025. [Online]. Available: https://www.kaggle.com/datasets/moneystore/agencyperformance

“TPC-H Homepage.” Accessed: Jun. 15, 2025. [Online]. Available: https://www.tpc.org/tpch/

K. Allam, “Cloud Data Warehousing: How Snowflake Is Transforming Big Data Management”.

“Multi-cluster warehouses | Snowflake Documentation.” Accessed: Jun. 15, 2025. [Online]. Available: https://docs.snowflake.com/en/user-guide/warehouses-multicluster?utm_source=chatgpt.com

“(5) Snowflake’s Multi-Cluster Shared Data Architecture: Scalability, Performance & Cost Optimization | LinkedIn.” Accessed: Jun. 15, 2025. [Online]. Available: https://www.linkedin.com/pulse/snowflakes-multi-cluster-shared-data-architecture-scalability-anuj-r--nbi9f/

D. A. S. George, “Deciphering the Path to Cost Efficiency and Sustainability in the Snowflake Environment,” Partn. Univers. Int. Innov. J. PUIIJ, vol. 01, no. 04, pp. 231–250, Aug. 2023, doi: 10.5281/zenodo.8282654.

D. Seenivasan, “OPTIMIZING CLOUD DATA WAREHOUSING: A DEEP DIVE INTO SNOWFLAKE’S ARCHITECTURE AND PERFORMANCE,” Mar. 31, 2021, Social Science Research Network, Rochester, NY: 5148190. doi: 10.2139/ssrn.5148190.

“Snowflake Documentation.” Accessed: Jun. 15, 2025. [Online]. Available: https://docs.snowflake.com/

X. Zeng, Y. Hui, J. Shen, A. Pavlo, W. McKinney, and H. Zhang, “An Empirical Evaluation of Columnar Storage Formats,” Nov. 07, 2023, arXiv: arXiv:2304.05028. doi: 10.48550/arXiv.2304.05028.

T. Koreeda, H. Honda, and J. Onami, “Snowflake Data Warehouse for Large-Scale and Diverse Biological Data Management and Analysis,” Genes, vol. 16, no. 1, Art. no. 1, Jan. 2025, doi: 10.3390/genes16010034.

D. M. Compagnoni, “Optimize Snowflake performance and reduce credit usage,” Nimbus Intelligence. Accessed: Jun. 15, 2025. [Online]. Available: https://nimbusintelligence.com/2024/10/5-ways-to-optimize-snowflake-performance-and-reduce-credit-usage/

“Fundamentals of Snowflake Query Design & Optimization | Keebo.” Accessed: Jun. 15, 2025. [Online]. Available: https://keebo.ai/2024/10/29/fundamentals-of-snowflake-query-design-optimization/

JayaAnanth, “Part 2 - Orchestrating Snowflake Data Transformations with DBT on Amazon ECS through Apache Airflow,” JayaAnanth. Accessed: Jun. 15, 2025. [Online]. Available: https://jayaananthdevops.github.io/posts/snowflake_dbt_ecs_part2/

“FinOps Principles.” Accessed: Jun. 15, 2025. [Online]. Available: https://www.finops.org/framework/principles/

“Understanding Data Warehouse Cost & Pricing Models | Rivery.” Accessed: Jun. 15, 2025. [Online]. Available: https://rivery.io/data-learning-center/data-warehouse-costs/

C. Wang, Z. Arani, L. Gruenwald, and L. d’Orazio, “Adaptive Time, Monetary Cost Aware Query Optimization on Cloud Database Systems,” in 2018 IEEE International Conference on Big Data (Big Data), Seattle, WA, USA: IEEE, Dec. 2018, pp. 3374–3382. doi: 10.1109/BigData.2018.8622401.

V. Leis and M. Kuschewski, “Towards cost-optimal query processing in the cloud,” Proc. VLDB Endow., vol. 14, no. 9, pp. 1606–1612, May 2021, doi: 10.14778/3461535.3461549.

P. Bhardwaj, “The Role of FinOps in Large-Scale Cloud Cost Optimization,” INTERANTIONAL J. Sci. Res. Eng. Manag., vol. 09, no. 01, pp. 1–5, Jan. 2025, doi: 10.55041/IJSREM28086.