Parallax Intel — Capital allocation diligence for AI in banking

01 The read

AI use cases look very different once scored on common terms

⟵Operational burden

Value evidenced⟶

Six sample banking functions, scored on the value peers have evidenced against the burden of running them — the same read we build for the functions our clients are weighing.

Design choices made by peers help answer different strategic questions — and show where value has been evidenced

85/100

Value Evidenced Score

In the cohort of six sample use cases, Treasury Forecasting is the most strongly evidenced value case.

Product Capability

Tactical liquidity or strategic capital planning?

BofAUp to 1 year of forecasts
JPMUp to 91 days of forecasts
PNC31 / 60 / 90 days, rolling forecasts
CitiShort / Medium / Long-term

Monetization Strategy

Monetized product or bundled feature?

BofAPricing not disclosedBundled in bank platform
JPMFree — charging under considerationBundled in bank platform
PNCPricing not disclosedBundled in bank platform
CitiSaaS pricingVia TIS — not bundled

Sourcing Strategy

Build, partner, or distribute?

BofAInternal buildNo third party disclosed
JPMInternal buildNo third party disclosed
PNCInternal build with partnerFISPAN
CitiDistribute third-party SaaSTIS

Data Scope

Which non-bank systems do we ingest from?

BofAMulti-bank
JPMMulti-bank + ERP
PNCMulti-bank + ERP + TMS
CitiMulti-bank + ERP + TMS + Working Cap.

Integration Architecture

Bank-side vs client-side integration?

BofABank platform
JPMBank platform with SAP Connector
PNCBank platform with ERP integration
CitiSaaS platform as integration layer

Burgundy marks where a bank diverges from the cohort norm — most often Citi's third-party SaaS model.

BofACashPro Forecasting JPMCash Flow Intelligence PNCPINACLE Cash Forecasting CitiCashforce / TIS

Ongoing regulatory costs create hidden friction that is unique for each case

50/100

Regulatory Intensity Score

Treasury Forecasting draws fewer high-enforcement regimes than customer-facing use cases.

U.S. regulatory relevance · Treasury ForecastingRelevance

Third-party & outsourcing riskInteragency TPRM guidance · Primary driver

High

Model riskSR 26-2

Medium

Tech governanceFFIEC IT Examination Handbook · OCC Heightened Standards

Low

Financial crime & payment systemsBSA/AML

Not relevant

AI-specific state lawsSB 26-189 (Colorado ADMT Act) · Texas Responsible AI Governance Act (TRAIGA) · Utah AIPA

Not relevant

Consumer protection & fair lendingFCRA/Reg V · ECOA/Reg B · UDAAP

Not relevant

Data privacy & securityGLBA · CCPA/CPRA + CPPA ADMT · Illinois BIPA

Not relevant

Analysis reflects regulatory applicability to typical workflows of AI in treasury forecasting. Actual applicability may differ based on the design and deployment specifics of any particular implementation. Coverage is bounded to 31 regulatory instruments across the U.S., Canada, UK, and EU (14 U.S. instruments shown here). Banks with operations in additional geographies, or exposure to instruments outside this coverage universe, may face different regulatory profiles. This analysis is for decision-support. Counsel is required for definitive applicability and compliance determinations.

AI creates new operating requirements that are cross-functional — high-intensity use cases can significantly diminish realized value

56/100

Operating Model Intensity

Operating choices adopted in typical treasury use cases are of low intensity in most categories — regulatory and tech requirements often shape these choices.

Scroll horizontally to compare all six sample use cases →

	Algorithmic Design			Data & Feature Lifecycle		Operational Behavior				Governance & Trust
	Model Architecture	Paradigm Type	Feature Engineering	Training Data	Model Refresh	Inference Mode	Human-in-the-Loop	Exception Handling	Output Integration	Explainability	Fairness & Discrimination	Model Validation	Data Controls	Data Privacy
Algorithmic Trading
AML
Credit Underwriting
Fraud Detection
Pricing Optimization
Treasury Forecasting

High intensity Moderate intensity Limited intensity Low intensity / Not Applicable

Assessing tech complexity across use cases separates light builds from multi-year initiatives

40/100

Tech Complexity Score

Applying disclosures and inferences to our view of tech complexity shows Treasury Forecasting has roughly half the build complexity of AML or fraud detection.

AI Components — Treasury Forecasting

Disclosed

Ensemble methods Time-series forecasting Scenario simulation AutoML Drift monitoring

Inferred

Neural networks domain-typical Anomaly detection technical implication

Notably absent

Large language models Generative AI Graph neural networks Federated learning Computer vision NLP

Data Modalities

Disclosed

Structured / Tabular Time-Series / Temporal

Absent

Graph / Network / Relationships Text Image Video Voice Code Geospatial

Integration Components

Disclosed

Multi-bank Ingestion ERP (SAP) TMS

02 The coverage

Coverage spans the whole bank — retail and commercial banking, wealth and asset management, capital markets, and operations

28

Banking functions

31

Legal instruments

14

Operating model categories

164

AI technologies

03 Where the read is useful

AI investment is hard to put through the discipline banks apply to every other capital decision

01 Where banks are

A growing portfolio of AI use cases

Banks have moved past whether to invest in AI. Most now manage a portfolio of initiatives across many functions, at different stages, all competing for the same capital and governance attention.

02 What breaks

The portfolio can't be assessed on common terms

Each use case carries a different payoff and a different cost to run safely — from back-office efficiency tools to customer-facing decision systems. There is rarely a consistent basis for weighing one against another.

03 Where it leads

Capital follows visibility, not value

In the absence of common criteria, decisions lean on inconsistent inputs — vendor claims, internal conviction, heuristics. Capital can flow toward the most visible use cases rather than the most valuable.

04 The diligence

Four questions this framework answers before any financial analysis is conducted

01 Value evidenced Where has value been disclosed in similar use cases, and how does it compare to alternatives? Read → 02 Regulatory intensity Which regulatory regimes apply, and how heavily? Read → 03 Operating model intensity What is the ongoing burden to run and govern, and where does it fall? Read → 04 Technical complexity Which AI components are involved, and how complex are they? Read →

05 Engagements

Two ways the work is applied

Bounded, senior-led, and delivered as board-ready analysis. For banks weighing where AI capital should go.

Engagement 01

Peer-evidence read

A structured view of where AI has delivered value across comparable banks, and where the regulatory and operating traps sit — focused on the functions you're weighing. The fastest way to decide where to start, and where not to.

Engagement 02

Portfolio assessment

The framework applied to your own AI investments, existing and planned — separating what peer disclosure can evidence from what is specific to your environment, and producing a four-dimensional read your risk and governance functions can stand behind.

See how engagements work →

06 About

Mahendra Wadhwa

Founder · Toronto

Parallax Intel was founded after eighteen years in capital allocation and corporate development across J.P. Morgan, RBC, Sun Life, and banks in Asia and Europe — evaluating competing investments, structuring transactions, and helping institutions decide where capital should go.

Assessing AI use cases is the same discipline applied to a newer question. An AI portfolio is a set of competing claims on a bank's capital and risk appetite; the framework brings a capital-allocation lens to deciding which of those claims hold up.

More about the practice →

Know what has actually worked, before you commit capital to AI.

AI use cases look very different once scored on common terms

Design choices made by peers help answer different strategic questions — and show where value has been evidenced

Product Capability

Monetization Strategy

Sourcing Strategy

Data Scope

Integration Architecture

Ongoing regulatory costs create hidden friction that is unique for each case

AI creates new operating requirements that are cross-functional — high-intensity use cases can significantly diminish realized value

Assessing tech complexity across use cases separates light builds from multi-year initiatives

Coverage spans the whole bank — retail and commercial banking, wealth and asset management, capital markets, and operations

AI investment is hard to put through the discipline banks apply to every other capital decision

A growing portfolio of AI use cases

The portfolio can't be assessed on common terms

Capital follows visibility, not value

Four questions this framework answers before any financial analysis is conducted

Two ways the work is applied

Peer-evidence read

Portfolio assessment

Mahendra Wadhwa

Deciding where to start with AI — or where to take it next?