Strategic White Paper The Three Spheres of Data Value From LLM to
Enterprise Language Model Why your Enterprise Language Model--trained on
proprietary data that no competitor can access--is the true source of
AIdriven competitive advantage -- Three layers of language model value:
LLM Market LM Enterprise LM -- Why the Enterprise Language Model is the
only layer that truly differentiates -- Industry applications for
insurance and asset management -- An action plan for CTOs, CIOs, and
technology leaders 1

THE THREE SPHERES OF DATA VALUE -- STRATEGIC TECHNOLOGY WHITE PAPER
Contents 01 Executive Summary 02 The Convergence Problem: When AI Models
Commoditise 03 The Three Spheres Framework 04 The Large Language Model
Layer -- Low Value Baseline 05 The Market Language Model Layer -- Higher
Value, Shared Edge 06 The Enterprise Language Model -- The True
Differentiator 07 Industry Application: Insurance 08 Industry
Application: Asset Management 09 The Strategic Imperative for Technology
Leaders 10 Building Your Enterprise Language Model: An AI-Accelerated
Action Plan 11 Sources 2

THE THREE SPHERES OF DATA VALUE -- STRATEGIC TECHNOLOGY WHITE PAPER 01
Executive Summary The artificial intelligence landscape is undergoing a
fundamental shift. As foundation models from Anthropic, Google, Meta,
Mistral, and OpenAI converge on broadly similar capabilities--all
trained on the same public internet data and built on comparable
architectures--the competitive advantage once associated with choosing a
particular AI model is rapidly eroding This convergence creates a
paradox: AI has never been more capable, yet the technology itself is
becoming commoditised. McKinsey's 2025 State of AI survey found that
while 79% of organisations report using generative AI, only 5.5% are
driving significant enterprise-level EBIT impact.1 The differentiator is
not which model you deploy. It is what data you feed it. " For these
foundation models to reach their peak value, you need to train them not
just on publicly available data, but you need to make privately owned
data available to those models. Larry Ellison, Oracle AI World 2025 This
white paper introduces the Three Spheres of Data Value--a strategic
framework that maps three distinct layers of AI capability: the Large
Language Model (LLM) trained on public data, the Market Language Model
enriched with subscription industry data, and the Enterprise Language
Model trained on your organisation's proprietary data. It is this
innermost layer--the Enterprise Language Model--that delivers the true,
defensible competitive advantage. Training AI on the intimate knowledge
of your clients, decisions, and operations adds incredible value that
cannot be gained anywhere else. 3

THE THREE SPHERES OF DATA VALUE -- STRATEGIC TECHNOLOGY WHITE PAPER 02
The Convergence ProblemBack When AI Models Commoditise Today's leading
large language models--Claude, GPT, Gemini, Grok, Mistral, Llama--are
all trained on broadly available public internet data and built on
comparable transformer architectures. Performance gaps between models
are shrinking with every release cycle.2 The question is no longer
"Which AI model are you using?" but rather "What data have you trained
it on that makes it uniquely yours?" This convergence has profound
implications for enterprise AI strategy. If every competitor has access
to the same foundation models and the same public training data, then
deploying AI on public data alone delivers parity, not advantage. The
organisation's unique data--its claims histories, underwriting
decisions, client interactions, portfolio performance records-- becomes
the scarce input that transforms a generic LLM into something far more
powerful: an Enterprise Language Model that understands your business at
a level no competitor can replicate. The Investment Reality
Organisations are investing heavily. McKinsey reports that 92% of
companies plan to increase AI spending over the next three years, with
high performers committing more than 20% of their digital budgets to AI
technologies--making them five times more likely to see transformative
impact.1 Yet only 1% of leaders describe their organisations as "mature"
in AI deployment. The gap between investment and impact is,
overwhelmingly, a data problem. Not a model problem, not a compute
problem, and not a talent problem alone. It is the failure to recognise
that language models exist in layers of strategic value--and that the
greatest returns come from building an Enterprise Language Model trained
on data that no competitor can access. 4

Increasing value

THE THREE SPHERES OF DATA VALUE -- STRATEGIC TECHNOLOGY WHITE PAPER 03
The Three Spheres Framework AI capability exists in three concentric
layers of increasing strategic value. Each layer represents a
fundamentally different language model paradigm--defined not by
architecture, but by the data on which the model learns. Understanding
this hierarchy is essential for directing AI investment where it creates
defensible, long-term differentiation. Large Language Model Public
internet data / LLM training corpus Low value. Available to everyone
Market Language Model Subscription / Industry data feeds Higher value
but shared with competitors Enterprise Language Model Proprietary client
and organisational data Highest value. Differentiating. Inimitable.
Figure 1: The Three Spheres of Data Value. The outer layer (LLM) is
trained on public internet data and is low value. The middle layer
(Market Language Model) adds subscription industry data for higher
value. The inner core (Enterprise Language Model) is trained on
proprietary organisational data and delivers the highest, most
differentiating value. The critical insight is this: as you move from
the outer layer inward, the language model becomes more valuable and
more differentiating. The LLM layer is the foundation--necessary but
ubiquitous and low value. The Market Language Model adds domain-specific
intelligence but is shared across competitors. The Enterprise Language
Model, at the core, is what makes your AI genuinely different from
everyone else's--trained on the intimate knowledge contained in your
client data, expert decisions, and operational history. 5

THE THREE SPHERES OF DATA VALUE -- STRATEGIC TECHNOLOGY WHITE PAPER

04 The Large Language Model Layer Low Value Baseline The outer layer
represents the standard Large Language Model (LLM)--trained on publicly
available internet data. This includes Wikipedia, academic papers, news
articles, public filings, open-source code repositories, social media,
forums, and government datasets. This is the corpus on which foundation
models like GPT, Claude, Gemini, and Llama are built. Public data gives
these LLMs their broad reasoning capabilities, language fluency, and
general world knowledge. Without it, there would be no foundation AI.
But precisely because this data is available to everyone and every model
is trained on essentially the same corpus, the LLM layer is low value
from a competitive standpoint. Any organisation deploying a standard LLM
on public data alone is, by definition, working with the same
intelligence as its competitors. The strategic role of the LLM layer in
an enterprise context is foundational but insufficient. It provides the
general intelligence layer--your AI can understand language, reason
about concepts, and process unstructured text. But it cannot tell your
AI how your specific organisation prices risk, serves clients, or makes
investment decisions. For that, you must move deeper into the sphere.

LLM (Public Data) Low value - available to all -- Wikipedia, news, web
pages -- Public regulatory filings -- Open-source code and docs --
Academic papers, patents -- Social media and forums -- Government open
data Market Language Model Higher value - subscription -- Bloomberg,
Reuters terminals -- Verisk, LexisNexis, Moody's -- S&P Global, FactSet
feeds -- Satellite and geospatial data -- Credit bureau data (Experian)
-- Industry benchmarking data -- Operational process data Enterprise
Language Model Highest value - proprietary -- Historical claims and
outcomes -- Client interaction records -- Internal pricing models --
Expert decision audit trails -- Portfolio performance history --
Operational process data Figure 2: Examples of data at each layer. The
shift from left to right represents increasing value and
exclusivity--from the low-value LLM layer to the high-value Enterprise
Language Model. 6

THE THREE SPHERES OF DATA VALUE -- STRATEGIC TECHNOLOGY WHITE PAPER 05
The Market Language Model Layer Higher Value, but a Shared Edge The
middle layer is the Market Language Model--an AI system enriched with
data available through commercial subscription services: market data
feeds, industry benchmarks, credit bureau records, alternative data
providers, and specialist analytics platforms. This layer has
significantly higher value than the base LLM because it incorporates
domain-specific, curated data that the general internet does not
contain. The alternative data market alone was valued at approximately
USD 4.6­9.6 billion in 2025, with investment managers spending an
estimated \$2.8 billion on alternative data sources.3,4 This market is
projected to grow at 16­40% CAGR, reaching \$22­49 billion by the early
2030s. Investment and trading firms account for roughly 41% of
alternative data revenues,3 underscoring how central third-party data
has become to financial services AI. The Market Language Model Paradox:
The Market Language Model is significantly more valuable than a standard
LLM--it speaks your industry's language and understands domain-specific
nuances. But because your competitors subscribe to the same data and
build the same enrichments, it creates parity rather than
differentiation. Bloomberg terminals are powerful--but every major firm
has one. Verisk data is invaluable to insurers--but every major insurer
accesses it. Market data raises the floor for the entire industry.
Organisations that fail to invest in a Market Language Model fall
behind. But organisations that rely on it exclusively are, at best,
keeping pace. The true value--the genuine competitive edge--comes from
what sits at the core of the sphere: the Enterprise Language Model,
trained on proprietary data that no competitor can access. Consider two
insurance companies both subscribing to the same catastrophe modelling
data. They receive identical flood zone maps, identical climate
projections. Their Market Language Models are functionally equivalent.
The company that then trains its Enterprise Language Model on 30 years
of its own claims history, adjuster notes, and granular loss-ratio data
will build a fundamentally superior risk model. The market data is the
same; the proprietary layer is the differentiator. 7

THE THREE SPHERES OF DATA VALUE -- STRATEGIC TECHNOLOGY WHITE PAPER 06
The Enterprise Language Model The True Differentiator The innermost
layer is where lasting competitive advantage is built. The Enterprise
Language Model is an AI system trained on the data your organisation
uniquely possesses: decades of transaction records, expert decision
trails, client interaction histories, internal process data, and the
codified institutional knowledge that resides in your people, systems,
and workflows. This is the data that competitors cannot access, cannot
purchase, and cannot replicate. When you fine-tune or augment AI models
with this proprietary data, the resulting Enterprise Language Model
understands your business--your clients, your risk appetite, your
operational nuances--in ways that no generic LLM or Market Language
Model ever could. The intimate knowledge contained in this layer is what
makes your AI services genuinely more valuable than anyone else's. As
Cloudera recently articulated: "At enterprise scale, competitive
advantage increasingly comes from the ability to adapt models to
proprietary data and run models where that data resides."2 Why the
Enterprise Language Model Creates an Unassailable Moat -- Inimitability:
LLMs can be copied; architectures can be replicated. But 30 years of
claims data, or a decade of portfolio construction decisions, cannot be
reverse-engineered or purchased. Your Enterprise Language Model is built
on data that is uniquely, irreplaceably yours. -- Compounding returns:
Every new data point your organisation generates improves your
Enterprise Language Model further, widening the gap with competitors who
lack that historical depth. The model gets smarter with every client
interaction, every expert decision, every operational cycle. -- Embedded
expertise: The Enterprise Language Model captures the implicit
judgements of your best people--the underwriter who intuitively adjusts
for a risk factor, the portfolio manager who spots a pattern in earnings
calls. Training AI on this data codifies and scales that expertise
across the entire organisation. -- Regulatory defensibility: In
financial services, where explainability and auditability are paramount,
AI decisions grounded in your own well-governed proprietary data are far
easier to justify to regulators than those based on opaque third-party
signals or generic LLM reasoning. The Value Equation: Using an
Enterprise Language Model for training adds incredible value not gained
anywhere else. It is the true value to add to AI models--the difference
between AI that matches your competitors and AI that fundamentally
outperforms them. BCG's 2025 analysis of the insurance sector reinforces
this point: the industry has "deep data reserves, including longitudinal
data on customer practices and interests" and has "long relied on
data-driven decision making." Yet only 7% of insurers have successfully
scaled their AI systems.5 The barrier is not technology--it is the
failure to properly harness proprietary data into an Enterprise Language
Model. 8

THE THREE SPHERES OF DATA VALUE -- STRATEGIC TECHNOLOGY WHITE PAPER

07 Industry Application: Insurance Insurance is perhaps the industry
where the three-layer language model framework is most viscerally
evident. Insurers are data businesses at their core--underwriting,
claims, and pricing are all fundamentally exercises in data-driven risk
assessment. AI adoption in insurance surged in 2025, with full adoption
jumping from 8% to 34% year-over-year and two-thirds of the \$5.08
billion in insurtech funding flowing to AI-focused companies.6,7

Layer LLM (Public)

Data Examples

AI Application

Competitive Value

Regulatory filings, news, weather Basic risk screening, general NLP for

data, public financial statements, document processing, public sentiment

census data

analysis

Low value. Every insurer has access to the same LLM capabilities.

Market LM

Verisk loss data, catastrophe

Industry benchmarking, standardised

models, credit scores, LexisNexis risk scoring, regulatory compliance

records, IoT/telematics feeds checks, fraud pattern databases

Higher value but shared. Creates parity, not edge.

Enterprise LM

30+ years of claims history, underwriter decision logs, adjuster notes,
loss-ratio data, client interaction records

Bespoke underwriting models, personalised pricing, predictive claims
triaging, fraud detection tuned to your book

Highest value. Differentiating. Competitors cannot replicate your
history.

Case in Point: Underwriting Transformation McKinsey's 2025 analysis
describes a leading North American insurer using agentic AI throughout
its underwriting workflow. The insurer uncovered implicit judgements
that underwriters had traditionally relied on and codified them into new
rules and protocols--enhancing efficiency and consistency. This
capability to embed unique expertise into an Enterprise Language Model
is becoming "core to insurers' intellectual property."8 The results at
industry level are striking: underwriting timelines collapsing from 3
days to 3 minutes, straight-through processing rates jumping from 10­15%
to 70­90%, claims resolution time reduced by 75%, and fraud detection
improved by over 30%.9 But these headline figures represent the ceiling
achievable only by organisations that have invested in building a
genuine Enterprise Language Model from their proprietary data.
Organisations working solely with a standard LLM or Market Language
Model will not reach these levels of performance. As Ray Ash of
Westfield Specialty articulated: "Can we teach an AI tool the metrics
that we use to evaluate risk so that it can provide informed suggestions
on a new submission? Can we teach acceptable ranges to an AI platform so
it can help classify risks?"10 The answer is yes--but only when the
Enterprise Language Model is trained on your organisation's proprietary
underwriting data, not just the public or market data that every
competitor also has.

9

THE THREE SPHERES OF DATA VALUE -- STRATEGIC TECHNOLOGY WHITE PAPER

08 Industry Application: Asset Management

In asset management, the three-layer language model framework maps
directly to the hierarchy of data that drives investment decisions. The
industry is at an inflection point: MSCI survey data shows 68% of wealth
managers view AI as moderately to very important, yet only 27% believe
their segment leads in AI adoption.11 The gap, once again, is
fundamentally about which layer of language model capability firms are
investing in.

Layer LLM (Public)

Data Examples

AI Application

Competitive Value

SEC/FCA filings, earnings transcripts, macroeconomic data, central bank
reports, news sentiment

Automated earnings analysis,

Low value. Same data

macro regime detection, basic NLP available to every fund.

sentiment scoring

Market LM

Bloomberg/Reuters feeds, FactSet, S&P; Global data, satellite imagery,
credit card transaction data, ESG scores

Quantitative screening, factor modelling, alternative data signal
generation, ESG compliance analytics

Higher value but subscribed by competitors. Creates parity.

Enterprise LM

Decade+ of portfolio decisions,

Bespoke alpha generation,

proprietary factor models, analyst notes, AI-driven portfolio
construction,

client suitability profiles, trade execution personalised client
proposals,

data, CRM interaction records

institutional knowledge scaling

Highest value. Your alpha source. No competitor can access your decision
history.

The Alpha Factory: Enterprise Language Model in Action A compelling
framework from the CFA Institute describes a four-stage "Human+AI
Funnel" for investment management: regime-aware allocation, AI-driven
idea generation (screening 5,000 stocks down to 500), deep analysis
leveraging a firm's proprietary prompt library, and portfolio
construction where human judgement meets machine precision.12 Crucially,
the stages where the Enterprise Language Model creates the most value
are stages 3 and 4. At the deep analysis stage, generative AI reads and
analyses corporate filings, management tone, and competitive
positioning--but it does so using the firm's proprietary prompt library
and historical analyst notes that are embedded in its Enterprise
Language Model. At the portfolio construction stage, AI optimises
position sizing using the firm's own historical performance data and
risk exposures. A competitor using the same base LLM but lacking this
proprietary Enterprise Language Model overlay would produce
fundamentally different--and likely inferior--results. MSCI's research
underscores that asset managers "looking to integrate AI for alpha
generation typically implement their AI tools in-house, in addition to
sourcing and preparing their proprietary datasets."11 These firms are,
in effect, building Enterprise Language Models--and the compounding
advantages they create widen over time.

10

THE THREE SPHERES OF DATA VALUE -- STRATEGIC TECHNOLOGY WHITE PAPER 09
The Strategic Imperative For Technology Leaders at Every Level The Three
Spheres framework carries different implications depending on your role
within the technology organisation. Below we outline the strategic
priorities for each level of leadership. For the CTO and CIO: Setting
the Vision -- Reframe the AI conversation around your Enterprise
Language Model, not generic AI. Board-level discussions should focus on
"What proprietary data assets do we have, and how are we building our
Enterprise Language Model?" rather than "Which LLM should we buy?" LLMs
are interchangeable and low value; your Enterprise Language Model is
not. -- Establish an Enterprise Language Model strategy. Commission a
comprehensive audit of all three layers within your organisation.
Identify where proprietary data exists but is not being captured,
structured, or governed. Every underwriter note not digitised, every
client interaction not recorded, is Enterprise Language Model training
data being discarded. -- Invest in data infrastructure with a 10-year
horizon. The value of your Enterprise Language Model compounds. Every
month of structured proprietary data you capture now strengthens your AI
capabilities in the future. High-performing organisations commit 20%+ of
their digital budgets to AI technologies.1 A significant portion should
flow to the data platforms that power your Enterprise Language Model. --
Govern for trust and regulatory confidence. In regulated industries, an
Enterprise Language Model grounded in wellgoverned proprietary data is
far easier to explain, audit, and defend. Build governance frameworks
that treat data quality and lineage as strategic priorities, not
compliance burdens. For VPs and Directors of Engineering / Data:
Executing the Strategy -- Build the data platform that powers your
Enterprise Language Model. Implement a modern data architecture
(medallion architecture with bronze/silver/gold layers on a cloud data
lake) that ingests, cleanses, and structures proprietary data for AI
training. Prioritise unstructured data: adjuster notes, call
transcripts, email correspondence, expert decision logs. -- Create
feedback loops between AI outputs and training data. Every AI-assisted
decision should feed back into the Enterprise Language Model's training
corpus, creating a virtuous cycle. When an underwriter overrides an AI
recommendation, that correction becomes training data. When a portfolio
manager adjusts an AI-suggested allocation, that adjustment encodes
expertise. -- Instrument proprietary data pipelines for quality. Your
Enterprise Language Model is only as good as its training data.
Implement automated data quality monitoring, anomaly detection, and
lineage tracking. A single data quality issue can undermine the entire
Enterprise Language Model. -- Design for fine-tuning and RAG, not just
prompting. Basic RAG (retrieval-augmented generation) on proprietary
documents is a starting point, but the real value of the Enterprise
Language Model comes from fine-tuning models on your organisation's data
so that domain knowledge is embedded in how the model behaves, not just
retrieved at query time.2 11

THE THREE SPHERES OF DATA VALUE -- STRATEGIC TECHNOLOGY WHITE PAPER

10 Building Your Enterprise Language Model

An AI-Accelerated Action Plan Traditional enterprise transformation
programmes measured progress in quarters and years. AI itself has
collapsed those timelines. Agentic AI can catalogue and classify data
assets in days rather than weeks, automated pipeline generation replaces
months of manual platform engineering, and modern fine-tuning and RAG
frameworks enable production-ready Enterprise Language Models in weeks
rather than months. The early phases of this plan should now be measured
in weeks, not months-- with only the final production-scaling phase
retaining a longer horizon, reflecting the organisational change
management, regulatory approval, and continuous learning cycles that
cannot be fully automated.

Phase 1 Audit

Timeline Weeks 1­4

2 Foundation

Weeks 4­10

3 Activate

Weeks 8­16

4 Scale

6­18 months

Key Actions

Outcome

Deploy AI-assisted data cataloguing agents to map all data assets across
three layers (LLM, Market LM, Enterprise LM). Automated discovery of
proprietary data that is unstructured, ungoverned, or inaccessible. AI
classifies data by value layer and readiness for Enterprise LM training.

Complete data asset inventory classified by language model layer, with
AIgenerated readiness scores

Stand up cloud-native data platform using AI-accelerated
infrastructureas-code and automated pipeline generation. Implement
governance, quality monitoring, and lineage tracking. Begin AI-powered
ingestion and structuring of unstructured proprietary data for
Enterprise LM training.

AI-ready data platform with proprietary data flowing to Enterprise LM

Deploy initial Enterprise Language Model use cases: RAG pipelines on
proprietary documents within days, fine-tuned domain models within
weeks. Establish automated feedback loops. Measure uplift vs. base LLM
rigorously. Iterate rapidly using AI assisted evaluation and prompt
engineering.

First Enterprise LM capabilities in production with measurable ROI
vs. generic LLM

Expand Enterprise LM across the organisation. Implement agentic AI
systems powered by Enterprise LM. Embed into production workflows with
full change management, regulatory sign-off, and continuous learning
cycles. Build compounding advantages through continuous proprietary data
capture.

Organisation-wide Enterprise Language Model driving AI capabilities
competitors cannot match

Measuring Success As you execute this plan, measure progress not just by
AI deployment metrics (models in production, use cases launched) but by
Enterprise Language Model metrics: Time to first Enterprise LM value:
With AI-accelerated delivery, initial RAG and fine-tuned models should
be in testing within weeks, not quarters. Track this as a leading
indicator of execution pace. Proprietary data volume and coverage: What
percentage of your expert decisions, clientinteractions, and operational
processes are being captured as Enterprise LM training data? Enterprise
LM uplift vs. base LLM: Track the measurable performance improvement
your Enterprise Language Model achieves compared to a standard LLM on
the same tasks. This is your competitive delta--the value that
proprietary training data adds. Feedback loop velocity: How quickly do
AI outputs and human corrections cycle back into your Enterprise LM
training data? Faster loops mean faster compounding of advantage.

12

THE THREE SPHERES OF DATA VALUE -- STRATEGIC TECHNOLOGY WHITE PAPER The
Bottom Line: In a world where LLMs converge and public data is
universally accessible, the organisations that win will be those that
build a genuine Enterprise Language Model--trained on the intimate
knowledge of their clients, their decisions, and their operations. Using
this proprietary data for training adds incredible value not gained
anywhere else. It is the true value to add to AI models. Your Enterprise
Language Model is your moat. Build it, deepen it, and defend it. Author:
Oliver Parker, CTO DataArt Finance Practice 13

THE THREE SPHERES OF DATA VALUE -- STRATEGIC TECHNOLOGY WHITE PAPER 11
Sources 1 McKinsey & Company, "The State of AI: Global Survey 2025,"
November 2025.
https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai
2 Cloudera, "When AI Models Converge, Proprietary Data Becomes the
Advantage," March 2026.
https://www.cloudera.com/blog/business/when-ai-models-converge-proprietary-data-becomes-the-advantage.html
3 Future Market Insights, "Alternative Data Market Size, Share &
Forecast to 2036," February 2026.
https://www.futuremarketinsights.com/reports/alternative-data-market 4
Neudata Intelligence, "The State of the Alternative Data Market in
2026," February 2026.
https://www.neudata.co/intelligence/the-state-of-the-alternative-data-market-in-2026
5 BCG, "Insurance Leads AI Adoption. It's Time to Scale," September
2025.
https://www.bcg.com/publications/2025/insurance-leads-ai-adoption-now-time-to-scale
6 Datagrid, "42 AI Agent Statistics for Insurance (Adoption + Impact),"
November 2025.
https://www.datagrid.com/blog/ai-agent-for-insurance-statistics 7 Risk &
Insurance, "AI Investment Surges in Insurance, But ROI Questions
Persist," February 2026.
https://riskandinsurance.com/ai-investment-surges-in-insurance-but-roi-questions-persist/
8 McKinsey & Company, "The Future of AI in the Insurance Industry," July
2025.
https://www.mckinsey.com/industries/financial-services/our-insights/the-future-of-ai-in-the-insurance-industry
9 Vantage Point, "Insurtech Trends 2026: How AI Is Transforming Claims
and Underwriting," March 2026.
https://vantagepoint.io/blog/sf/insights/insurtech-trends-2026-ai-claims-underwriting
10 Risk & Insurance, "How Underwriting and Claims Are Reshaped by AI in
Insurance," September 2025.
https://riskandinsurance.com/how-underwriting-and-claims-are-reshaped-by-ai-in-insurance-and-how-they-stay-the-same/
11 MSCI, "Why Wealth Managers Should Think Differently About AI,"
January 2026.
https://www.msci.com/research-and-insights/blog-post/why-wealth-managers-should-think-differently-about-ai
12 CFA Institute Enterprising Investor, "Reducing the Cost of Alpha: A
CIO's Framework for Human+AI Integration," December 2025.
https://blogs.cfainstitute.org/investor/2025/12/04/reducing-the-cost-of-alpha-a-cios-framework-for-humanai-integration/
14