Category Creation Playbook · First-Mover Window Open The Lean AI
Category Playbook
Economic unsustainability is the enemy. Building the future’s AI and data foundations requires a pivotal shift: philosophically, economically, architecturally, and operationally. This is the playbook for the team that names that shift, owns the category, and brings the practice to market.
Macro Category
AI Economics
The boardroom frame. Managing cost, efficiency, and ROI of AI at enterprise scale.
ai-economics.com ✓The Philosophy
Lean AI
High-performance AI built to eliminate waste. Right task, right model, less token burn, cleaner workflows, better outcomes.
lean-ai.com ✓Operational Discipline
TokenOps
Managing token spend, caching, and context routing with the rigor of cloud cost management.
token-ops.io ✓Performance Standard
Precision AI
Right intelligence, right resources, right cost. Not cheap. Precise. The measurable proof that the stack works.
precision-built.ai ✓ Section 01 · The Stack
01
Overview: The AI Economic Stack
Four layers. Four domains secured. One overarching philosophy. This is the architecture for the category we are creating and the vocabulary that will define how the enterprise market talks about AI efficiency for the next decade.
The Problem
The Enemy: Unsustainable Spend
In this new tech frontier we are quickly learning that our default behaviors will not stand the test of time: max-model by default for all tasks, excessive token spend, and no ROI accountability is not sophistication. It is waste. Enterprises are plowing through AI operating budgets at breakneck speed. The bill and the wake-up call have arrived.
The Philosophy
Lean AI is the discipline that follows
A new way of building and running AI is not an option. It is non-negotiable. Every transformative technology follows the same arc. Expensive and unoptimized first, then efficient and commoditized. Cloud did it. AI is next. Lean AI is the name for the efficiency discipline that follows every technology life cycle.
The Opportunity
A Universal Problem Exists with No Unified Name
Five names. One category. We are naming all of them: AI Economics, Lean AI, TokenOps, Precision AI, Frugal AI. No analyst has published a market guide. No vendor owns any of them. The practitioner who publishes first owns the reference point permanently.
The Tagline
Better. Faster. Cheaper.
"Precision performance systems run on lean economics." High-performance AI is not built by spending more. It is built by wasting less.
Section 02 · The Origin
02
Three Threads, No Commercial Owner
Frugal AI is not one lineage. It is three independent threads, academic, French national policy, and an applied research institute, that converged on the same idea without ever producing a commercial category. The vocabulary exists. The institutional owner does not.
Jul 2019
Green AI names the efficiency problem first
Roy Schwartz, Jesse Dodge, Noah A. Smith, and Oren Etzioni publish "Green AI" at the Allen Institute for AI, arguing that AI research had normalized a 300,000x increase in training compute from 2012 to 2018 and proposing efficiency as an evaluation criterion alongside accuracy. This is the first credible academic naming of the problem Frugal AI later organizes.
Source: arXiv 1907.10597
2020
"Frugal AI" enters the academic literature directly
Researchers begin publishing under the term "frugal AI" specifically, distinct from Green AI's framing, drawing on the older discipline of frugal innovation (doing more with less, long established in development economics and Indian and emerging-market engineering practice) and applying it explicitly to machine learning systems.
2022–2025
France adopts Frugal AI as national policy, not just research
France's national AI strategy (SNIA) names Frugal AI one of four research priorities for its 2022–2025 phase, backed by €2.2B, alongside Embedded AI, Trusted AI, and Generative AI. The French Ministry of Ecological Transition publishes a "Référentiel général pour l'IA frugale," a formal methodology for measuring and reducing AI's environmental impact, in 2024. CNRS defines Frugal AI around minimizing energy use and ecological impact, what the French call "sobriété." This is environmental and regulatory framing, distinct from the cost-optimization angle the enterprise market is about to need.
Sources: Société Numérique (SNIA), GreenTech Innovation (Ecolab référentiel)
Nov 2024
The Frugal AI Hub launches at Cambridge Judge Business School
Serish Venkata Gandikota and Elizabeth Osta co-found the Frugal AI Hub at Cambridge Judge, hosting the inaugural Frugal AI Workshop on November 22, 2024. The Hub builds the first applied, practitioner-facing institution around the term, publishing four principles (Resource Efficiency, Sustainability, Accessibility & Inclusion, Impact & Scalability) and white papers through 2025 and into 2026, with partnerships including UNICC. This is the most credible existing institution using the term, and it is academic and nonprofit, not a commercial category.
Source: frugalai.org
2024
The bill arrives in the enterprise
AI chip spend hits $22B. CFOs start asking hard ROI questions. MIT publishes that 95% of AI pilots deliver zero measurable P&L impact. The efficiency conversation, academic and policy-driven in Europe for five years, becomes an enterprise budget conversation in America with no warning and no framework.
2026
Three threads, zero commercial category king
Academic Green AI (2019), French environmental policy (2022–2025), and Cambridge's applied Frugal AI Hub (2024–present) have each built credibility in their own lane. None has built the commercial enterprise category: the boardroom frame a CFO and CTO use to manage AI cost and ROI at scale. Uber exhausts its annual AI budget four months into the year. The COO cannot draw a line between spending and outcomes. This is the window.
The Strategic Gap
The academic origin is real and well documented: Green AI in 2019, Frugal AI in the literature by 2020, French national policy by 2022. What does not exist is the commercial enterprise category. Cambridge Judge's Frugal AI Hub is the closest thing to an institutional owner, and it operates as an academic and nonprofit catalyst, not a market category with a vendor ecosystem. AI Economics is the name for what comes next: the boardroom frame that organizes Frugal AI's academic credibility into something a CFO can act on.
Section 03 · Naming Ecosystem
03
The Category and Its Four Names
Frugal AI has credible academic and institutional roots. The broader commercial category has not been named anywhere. We are creating the convention that organizes everything: four terms, one architecture, one category king.
The Category We Are Creating
AI Economics is the macro category: the boardroom frame for managing the cost, efficiency, and ROI of AI at enterprise scale. It is the umbrella that organizes every term below it. No analyst has published a market guide for it. No vendor owns it. This is the name we are planting.
ai-economics.com secured
The Philosophy
Lean AI
High-performance AI built to eliminate waste. Right task, right model, less token burn, cleaner workflows, better outcomes. The operating discipline of AI Economics.
lean-ai.com secured
Operational Discipline
TokenOps
Managing token spend, caching, and context routing with the rigor of cloud cost management. The FinOps layer for the model stack.
token-ops.io secured
Performance Standard
Precision AI
Right intelligence, right resources, right cost. Not cheap. Precise. The measurable proof that the stack works: cost per correct output as the primary metric.
precision-built.ai secured
Frugal AI: The Named Precursor
Frugal AI has an established definition rooted in the Allen Institute's 2019 Green AI paper, the academic literature since 2020, and the work of the Cambridge Judge Business School Frugal AI Hub: the discipline of designing AI systems to achieve high impact with minimal resources across compute, energy, data, and capital, guided by four principles, resource efficiency, sustainability, accessibility and inclusion, and impact and scalability. Sources: frugalai.org, arXiv 1907.10597, Société Numérique
Frugal AI named the concept academically and institutionally. AI Economics names the commercial category that organizes it for the enterprise buyer. Our four-term ecosystem is the convention that translates established academic credibility into a framework a CFO and CTO can act on together.
What We Are Not Claiming
We did not invent Frugal AI. We are not the first to name resource-efficient AI. We are building the commercial category and naming ecosystem on top of existing academic and policy momentum, by giving the problem a commercial name and frame, and by laying out the strategy, principles, and execution layers required to actually practice AI Economics and Lean AI, not just theorize about them. Section 05 is that practice layer. Sections 10 and 12 are how that practice becomes citable and discoverable, by Wikipedia editors and by AI answer engines alike.
Section 04 · The Problem
04
The Real Enemy Is Unsustainable AI Spend
Maximum-by-Default is a symptom. The root problem is that the default architecture makes runaway cost structurally inevitable. If AI is going to be widely adopted and used in perpetuity, it has to be cost-managed. The organizations that solve this first own the next decade.
The Root Problem
Unsustainable AI spending based on inefficient data systems, bloated architectures, and default user behaviors.
Problem 01
Environmental
Data center energy and water draw is becoming a visible public issue. AI is the fastest-growing slice. The political and grid pushback has already begun.
Problem 02
Economic
Frontier model costs for training and inference are rising faster than clear ROI for most enterprise deployments. CFOs are asking hard questions. The honeymoon spend phase is over.
Problem 03
Access & Equity
If only five labs and the Fortune 500 can afford to play, AI benefits concentrate. Startups, the Global South, public sector, education: all locked out of the default big-model path.
Problem 04
The Overkill Reflex
A huge share of production AI calls are doing work a distilled 3B-parameter model could handle. That is pure waste built into the default architecture decision.
Problem 05
Diminishing Returns
The scaling curve is flattening for most tasks. Doubling parameters no longer doubles usefulness. The cost-per-marginal-capability ratio is getting worse, not better.
Problem 06
No ROI Accountability Framework
There is no standard methodology for measuring AI return on investment at the workflow level. MIT found 95% of AI pilots deliver zero measurable P&L impact. The spend is real. The accountability infrastructure does not exist yet.
Structural Force 01: Physical
Hardware costs are inverting
AI chip spend tripled from $22B (2024) to $52B (2025). Supply chain controlled by three companies operating fabs that take 5–7 years to build. The next chip generation makes it worse, not better.
Structural Force 02: Economic
ROI hit rate is unsustainable
95% of AI pilots deliver zero measurable P&L impact (MIT NANDA). A 21–25% ROI hit rate against $675B projected annual spend is not a stable configuration. The pressure that follows demands efficiency.
Structural Force 03: Regulatory
Sovereignty is now a legal requirement
EU AI Act, India data localization, China AI regulations: all require inference to stay in-country. Edge and on-premise is no longer a preference. It is law in an increasing number of markets. Both pressures point the same direction: smaller, local, efficient.
IEA, 2026
"From 2024 to 2030, data centre electricity consumption grows by around 15% per year: more than four times faster than the growth of total electricity consumption from all other sectors." International Energy Agency, Energy and AI Report 2026
Section 05 · The Practice Layer
05
Practicing AI Economics, Not Just Naming It
A category with no practice behind it is a marketing position waiting to be exposed. This section is the operating discipline: five principles a team commits to, and under each, the concrete mechanics that turn the principle from a slide into a Tuesday-morning decision.
Principle 01
Right-size before you scale
No model gets deployed to production without a documented reason it could not be smaller.
Phase 1
Prompt and context audit. Most LLM calls waste 40–60% of input tokens on stale history, boilerplate system prompts, or full-file context the model never uses. Trim before touching the model.
Phase 1
INT8 quantization as the default, not the exception. Roughly 50% memory reduction with under 1% measured quality loss on most tasks. This is a configuration change, not a research project.
Phase 2
Task-fit review before every model selection. Classification, routing, extraction, and summarization are solved problems at the 3–8B parameter scale. A frontier model on these tasks is a procurement failure, not a sophistication signal.
Phase 3
The compression pipeline, applied in order: prune, distill, quantize (P-KD-Q). Each step compounds on the last. Distillation alone can cut cost 90% on stable, high-volume tasks once the teacher model's behavior is well understood.
Principle 02
Cache and route before you compute
If the question has been answered before, the system should know that before it spends a token answering it again.
Phase 1
Semantic caching for repeated or near-duplicate queries. Even a distilled model wastes compute re-answering semantically similar questions. Caching is infrastructure, not a model change, and it is the fastest win available.
Phase 1
Prefix caching and PagedAttention enabled by default. Standard in vLLM and TGI; the most common failure is having it available and not configured correctly. 2–4x throughput increase for no architectural change.
Phase 2
A routing layer that sends each request to the smallest sufficient model. This is TokenOps in practice: every inference request is classified by required capability before it is dispatched, the same way a cloud router sends traffic to the cheapest adequate compute tier.
Phase 3
Speculative decoding for latency-sensitive paths. A small draft model proposes tokens a larger model verifies. 2–3x latency reduction with no quality loss, worth the added complexity once a workflow is latency-critical and stable.
Principle 03
Cost per correct output is the only metric that matters
Benchmark leaderboard position is not a business metric. It never appears in a board deck next to revenue.
Phase 2
Every workflow gets a cost-per-correct-output baseline before optimization starts. You cannot prove a 75% cost reduction if you never measured the starting number. This is the single most skipped step.
Phase 2
ROI is owned jointly by the team that builds the workflow and the team that pays for it. Engineering reports cost per output; finance reports business impact per dollar. Neither number means anything without the other.
Phase 4
A quarterly inference cost review, modeled on cloud FinOps reviews. Tag spend by workflow, not by team. Flag any workflow whose cost-per-output has not improved in two consecutive quarters; stagnant cost on a mature workflow signals a missed optimization, not stability.
Principle 04
Evals gate every deploy, not just launch
A cheaper model that fails silently is more expensive than an expensive model that works. Quality and cost are reviewed in the same meeting.
Phase 1
A task-specific eval suite exists before a smaller model is allowed to replace a larger one. Without it, "we saved 70% and quality dropped" is a story you only discover from customers.
Phase 2
Quality thresholds are written down as a number, not a feeling. "95% of the original performance" only means something if the original performance was measured the same way.
Phase 3
Continuous eval monitoring on production traffic, not just at deploy time. A distilled model's quality can drift as the input distribution shifts. Catch it before a customer does.
Principle 05
Sovereignty and sustainability are architecture decisions, not afterthoughts
Where inference runs and how much energy it costs are decided at design time, not discovered at audit time.
Phase 2
Data residency requirements are mapped before model selection, not after. EU AI Act, India data localization, and China's AI regulations require in-country inference. A right-sized model is also more often a model that can run on-premise or at the edge, which solves this and the cost problem together.
Phase 4
Carbon and energy cost per inference is tracked alongside dollar cost. The two move together almost always; a workflow that gets cheaper almost always gets greener, which makes this principle nearly free to practice once Principle 03 is in place.
The Implementation Sequence, Compressed
In practice, the five principles are not implemented in parallel: they are implemented in four phases, ordered by effort and payoff. Phase 1: prompt and context trimming, INT8 quantization, semantic and prefix caching, eval suite stood up. No model retraining, immediate cost reduction. Phase 2: task-fit review, routing layer, cost-per-output baselines, quality thresholds, data residency mapping. Phase 3: the P-KD-Q compression pipeline, speculative decoding, continuous eval monitoring. Phase 4: quarterly FinOps-style cost reviews and carbon tracking. A team that has cleared Phase 2 is not theorizing about Lean AI. They are practicing it, and they have the cost curve to prove it.
Section 06 · Six Frameworks
06
Six Frameworks. One Conclusion.
Every major strategy framework that addresses market leadership converges on the same insight. The conclusion is always the same: define the category, name the enemy, and competition becomes the best marketing you never paid for.
1
Conclusion they all reach
76%
Market value captured by category king
0
Competitors who have named this yet
Framework 01
Category Pirates
Lochhead, Cole, Yamada
Category kings capture 76% of total market value: not by beating competitors, but by designing the category before competitors understand it exists. The Magic Triangle: Company + Product + Category must all be designed together.
"The second entry never wins the category: they grow it for the king. Miller Lite didn't beat Bud Light. It created light beer and handed the king a bigger market."
Lean AI Application
Company (AI Economics practice) + Product (Signal Scout) + Category (Lean AI / AI Economics). When McKinsey enters "AI cost optimization," they grow the category we designed.
Framework 02
Play Bigger
Ramadan, Peterson, Lochhead, Maney
Category creation is the highest-leverage business activity. The POV document: a lightning strike moment that declares the category: is worth more than any product launch. Category kings don't respond to competitors; competitors respond to them.
"When your competitor announces a product in your category, they've just run an ad for the problem you invented. The market hears the problem; it already knows who invented the solution."
Lean AI Application
This playbook IS the POV document. Signal Scout IS the lightning strike. Every competitor announcement from here is a free ad: with us as the reference point.
Framework 03
Blue Ocean Strategy
Kim & Mauborgne
Create uncontested market space by making competition irrelevant. Value innovation = simultaneous differentiation AND cost reduction. Eliminate what the red ocean competes on; create what it never offered.
"When imitators enter your blue ocean, they create a red ocean: which they fight in while you've already moved to the next blue ocean. The original remains yours."
Lean AI Application
Red ocean: competing on headcount, certs, logos. Blue ocean: proprietary signal intelligence + AI Economics framework. No one is playing this game yet. The canvas is blank.
Framework 04
Zero to One
Peter Thiel
"Competition is for losers." Build monopolies, not competitive businesses. The last mover advantage: define the category so definitively that you set the standard everyone else must reference. Start small, monopolize, expand.
"Imitators don't threaten the monopoly: they validate that you saw something real before anyone else did."
Lean AI Application
Start: monopolize "Lean AI." Expand to "AI Economics." Moats: Signal Scout intelligence, owned vocabulary (5 terms), framework adoption, category king position.
Framework 05
Crossing the Chasm
Geoffrey Moore
The gorilla captures 70%+ of market profits once it crosses from early adopter to mainstream. The Whole Product: not just core offering but everything the customer needs to succeed: is what crosses the chasm. The gorilla doesn't react to chimps.
"Market growth from competition benefits the gorilla disproportionately: they have the most installed base, most ecosystem, most to gain from a larger market."
Lean AI Application
Whole Product: Signal Scout + AI Economics framework + assessment + certified practitioners + case studies. Bowling alley: data engineering → ML platform → enterprise architecture → CFO suite.
Framework 06
22 Immutable Laws
Ries & Trout
Own a word in the mind of the prospect. The Law of Leadership: better to be first in the mind than first in the market. Positioning is not about the product: it's about what you own in the mind. Once owned, it cannot be taken by a "better" competitor.
"Once you own a position, you cannot be dislodged by a competitor who is 'better.' Only by one who creates a new category: which you can prevent by creating it first."
Lean AI Application
The word to own: "Frugal AI." Once we own "Lean AI" in the practitioner's mind, competitors are permanently on the lower rungs. They can be better. They can never be first.
The Principle Every Framework Agrees On
"The category king isn't the best option among many. They're the only option in a category of their own creation. When competitors arrive, they pay to educate the market about a problem you named first." Six frameworks. Same conclusion.
Section 07 · Paradigm Shift
07
Old World vs. New World
Six decision dimensions. The old world defaults to maximum. The new world defaults to precision. Every dimension compounds: organizations that shift all six own a structural cost advantage permanently.
Maximum-by-Default: The Old Game
- Model Selection: Biggest model by default: capability as the primary signal, prestige over precision
- Spend: Maximum spend: inference costs treated as rounding errors, finance arrives after the bill
- Success Metric: Benchmark performance: MMLU scores and leaderboard position, ROI assumed never measured
- Decision Driver: Benchmark-driven: "which model performs best?" never "which model is sufficient?"
- Ownership: CTO owns it alone: CFO enters the conversation after the spend is committed
- Competitive Position: Vendor lock-in: whoever has the biggest model defines the category
VS
Frugal AI / The AI Economic Stack: The New Game
- Model Selection: Right-sized model: smallest that achieves deterministic-enough output, 1/100th the cost
- Spend: Cost efficiency: AI spend managed with the rigor of cloud FinOps, tagged and optimized per workload
- Success Metric: Inference ROI: cost per correct output, every deployment tied to measurable P&L impact from day one
- Decision Driver: Outcome-driven: right intelligence, right resources, right cost: not maximum of all three
- Ownership: CFO + CTO co-own: finance is in the room before the architecture is drawn
- Competitive Position: Category king: defines the framework before vendors understand the game
The Shift Is Structural
This is not a preference change. It is a mandate arriving from the CFO floor. Uber exhausted its annual AI budget in four months. The COO could not draw a line between spend and outcomes. The organizations that build the efficient stack now own the margin advantage permanently: before the mandate is issued.
Category Durability
01
Resource Efficiency
Minimize compute, memory, and energy per unit of AI output. Right-sized model = right-sized cost.
02
Sustainability
Reduce environmental footprint across the full AI lifecycle. Carbon per inference is becoming auditable.
03
Accessibility
Make capable AI available beyond hyperscale infrastructure. Not just Fortune 500: every organization that needs it.
04
Inclusion
Enable organizations of all sizes to deploy AI economically. Startups, public sector, Global South: all on the same efficient stack.
05
Impact
Optimize for measurable business outcomes, not raw capability. The metric is cost per correct output: measurable ROI, not benchmark performance.
06
Scalability
Build architectures that grow efficiently without cost spirals. The efficient stack compounds. The wasteful stack doesn't.
Section 08 · The Window
08
Why Now Is the Only Window That Matters
The enterprise AI bill has arrived. The CFO is in the room. The category name is still unclaimed. These three conditions will not exist simultaneously for long.
The Signal
The bill has arrived.
AI compute costs tripled year-over-year. FinOps is already becoming a job title for AI. The CFO/CIO alignment conversation has started and no clean framework exists to answer it.
The Gap
No one owns the discipline
Five terms competing without a named commercial owner: Frugal AI, Efficient AI, Sustainable AI, Responsible AI, Green AI. All describing the same imperative. No analyst has named it. No vendor owns it.
The Window
First-mover closes permanently
The practitioner who publishes the first credible, dated, cited definition owns the reference point forever. The timestamp is the moat. The citation is the lock.
Data Point
$52B
Total AI chip spend in 2025, up from $22B the year prior. Nearly tripled in 12 months. The spend is real. The ROI accountability framework does not exist yet.
Data Point
80% over-resourced
Enterprise AI tasks running on frontier models that are 100x too powerful for the job. Classification, routing, extraction: all on GPT-4. Waste is the default architecture.
Data Point
95% zero P&L impact
MIT study: 95% of AI pilots deliver zero measurable P&L impact. The efficient stack is not a feature request. It is the fix for the most expensive experiment in enterprise history.
The Structural Insight
This is not an execution problem. This is a naming problem. Every transformative technology category follows the same arc: expensive and unoptimized first, then efficient and commoditized. DevOps named silos. FinOps named cloud waste. AI Economics is next: and the category does not have a king yet.
The Timing Truth
Uber exhausts its annual AI budget four months into the year. The COO cannot draw a line between spending and outcomes. The enterprise market is primed. The category name is unclaimed. These two conditions exist simultaneously right now. They will not for long.
Section 09 · The Playbook
09
How to Build the Category
Category kings are not discovered. They are built through a deliberate sequence of moves.
P1
Publish the category POV: with a timestamp
The manifesto: the problem (Maximum-by-Default), why existing solutions fail, the new paradigm (Lean AI / AI Economics), and why this is the inevitable direction. Date it publicly. The timestamp is the moat every future claim of "we invented this" must reference.
P2
Define the category criteria: set the bar others are measured against
Publish the standard: right-sized model selection methodology, inference ROI measurement framework, workflow-level economic accountability. Every competitor who enters will be measured against criteria you wrote.
P3
Design the Magic Triangle as a unified system
Company (AI Economics practice) + Product (Signal Scout intelligence + framework) + Category (Lean AI / AI Economics): all three must reinforce the same central idea in every communication. If they diverge, the category fractures.
P4
Plant the flags: everywhere that matters
Wikipedia (Lean AI, Frugal AI, AI Economics, TokenOps, Precision AI: all five blank or thin). Wikidata. Google Scholar via arXiv preprint. GitHub org. Substack newsletter. LinkedIn long-form series. Analyst pre-briefings (Gartner, Forrester, IDC). CNCF / Linux Foundation SIG proposal. Each flag compounds the others: the four-lock moat is Wikipedia + Wikidata + arXiv + analyst citation simultaneously.
P5
Publish "The AI Economics Stack" as the definitive practitioner framework
Model selection methodology, inference cost modeling, ROI measurement, right-sizing architecture guide. Free PDF. When buyers search "AI Economics framework," this is the result. When analysts write first coverage, this is what they cite.
The Wikipedia Opportunity
Search Wikipedia for "Lean AI," "AI Economics," "TokenOps," or "Precision AI." You will find a stub, a redirect, or nothing. "Frugal AI" has academic grounding through frugalai.org and the arXiv literature, but no dedicated, well-cited Wikipedia article anchoring it as a commercial category. The commercial category terms remain unclaimed. The practitioner who publishes the first credible, dated, cited definition that connects the academic record to the enterprise category owns the reference point permanently.
Section 10 · Publication Strategy
10
Standing Up the Category Language
Saying you invented a category is not the same as the record proving it. Category language infrastructure is the work of making ownership permanent, verifiable, and defensible. It is not marketing. It is evidence architecture.
Why This Exists
Wikipedia is where analysts start. It is where journalists verify. It is where enterprise buyers confirm a category is real before putting it in a board deck. A category without a Wikipedia article is a category that does not yet exist in the institutional record. The team that writes the article controls the definition. The team that gets there second argues with someone else's framing forever. The goal of this seven-phase sequence is to build a citation infrastructure so credible and so timestamped that no future entrant can contest the origin point.
The Naming Strategy
Wikipedia and academic journals do not accept promotional content. The path to a durable Wikipedia article on "AI Economics" runs through peer-reviewed or preprint citations that establish notability on Wikipedia's terms: independent, verifiable, and already referenced elsewhere. Frugal AI already has this foundation, the Green AI paper (2019), the academic literature since 2020, the Cambridge Judge Frugal AI Hub. The preprint paper is not a marketing asset. It is the citation that connects this existing academic credibility to the commercial category. Published on arXiv or SSRN with a datestamp, it becomes the source every future reference traces back to.
Write and publish the preprint paper
A practitioner-authored paper on AI Economics published to arXiv or SSRN. The paper establishes: the problem (Maximum-by-Default), the terminology (AI Economics, Lean AI, building on the established Frugal AI literature), the framework (right-sizing methodology, inference cost modeling), and the evidence base (Signal Scout signal data, enterprise cost benchmarks). It explicitly cites the prior art (Green AI 2019, Frugal AI Hub) to position AI Economics as the commercial extension of established academic work, not a competing claim. This is the founding document. Everything else cites it.
No preprint, no citation. No citation, no Wikipedia article. Citing prior art correctly is what makes the new claim credible rather than disputable.Phase 2
Parallel to Phase 1
Engage a Wikipedia editor
Wikipedia prohibits article creation by parties with a conflict of interest in the subject. Engaging an independent, experienced Wikipedia editor to draft and submit the articles is the compliant path. The editor is briefed on the notability argument: the preprint paper, the existing Frugal AI academic record, press coverage, and the Signal Scout intelligence record. The editor operates independently. This team provides the sourcing, not the article text.
COI violations get articles deleted and flagged. An independent editor protects the article's permanence.Phase 3
Parallel to Phases 1–2
Build the citation infrastructure
A single preprint is a thin foundation. The citation infrastructure layers additional independent references: the LinkedIn manifesto series (public, timestamped, archived by Wayback Machine), a Signal Scout briefing published publicly with a date, a company blog post defining the AI Economics framework, and outreach to independent writers or analysts willing to cover the category emergence. Each piece becomes an additional citation. Thickness of the citation record determines article survival during Wikipedia's deletion review process.
Wikipedia articles with thin citation records get nominated for deletion. Multiple independent sources create durability.Generate independent press and analyst coverage
Press coverage in recognized publications is the highest-value Wikipedia citation because it satisfies notability through editorial independence. Target outlets covering AI infrastructure and enterprise technology: The Information, VentureBeat, The Register, MIT Technology Review, and relevant Substack newsletters with documented readership. The pitch is a category story, not a product announcement: "Frugal AI has academic roots back to 2019. Here is the enterprise category it has been missing, and the data proving the market needs it now." Analyst briefings with Gartner, Forrester, and IDC run in parallel.
Press in recognized publications is Wikipedia's gold standard for notability. Citing the real academic history makes the pitch harder to dismiss as opportunistic.Submit the five articles in sequence, strongest case first
Order follows notability strength, not narrative preference: (1) Frugal AI first, already clearing the threshold on existing academic and policy coverage; (2) TokenOps second, backed by the Linux Foundation Tokenomics Foundation; (3) AI Economics third, framed alongside DevOps and FinOps as precedent for practitioner-created categories; (4) Lean AI fourth; (5) Precision AI last, the most complex submission, requiring explicit disambiguation from medical and defense uses of the term. Each submission waits until the prior article has been live and stable before the next is filed, so the ecosystem builds citation support sequentially rather than all at once.
A single article is fragile. Five interlocking articles, submitted in order of notability strength and anchored in real prior art, create a record that is structurally hard to dismantle.Post-publication defense, maintenance, and peer-reviewed follow-through
Wikipedia articles on emerging categories attract edits, neutrality tags, and occasional deletion nominations. Watchlist monitoring and a deletion-defense plan exist for each article before it's submitted, not improvised afterward. In parallel, the preprint pursues peer-reviewed placement, IEEE Access first, then JIT or MIS Quarterly Executive, once it has had time to circulate; acceptance turns the founding citation into the strongest possible reference. A quarterly citation log (new press, analyst mentions, conference references) feeds the editor on a regular cadence.
Publishing the article is the beginning, not the end. Category ownership on Wikipedia, and the strength of its citation record, is actively maintained, not passively held. The Bottom Line
"We invented Frugal AI" would be false and easily disproven, the term has real academic roots back to 2019 and an active institutional home at Cambridge Judge. The accurate and defensible claim is narrower and stronger: we are naming and owning the commercial enterprise category, AI Economics, that organizes the existing Frugal AI body of work into something a CFO can act on. A dated arXiv preprint that correctly cites prior art, Wikipedia articles, press coverage, and an analyst note from Gartner is a record. A false origin claim is a liability the moment anyone checks frugalai.org. Get the history right and the claim becomes uncontestable instead of disprovable.
Section 11 · Roadmap
11
The Roadmap
This is the execution sequence, not a wishlist. A companion document, the AI Economics Category Ecosystem Wikipedia Submission Execution Playbook, carries the checkbox-level detail (exact URLs, named contacts, submission templates) for every step below. This roadmap shows the order and the dependencies; that document shows the how.
Six Phases, Not Five
The real plan runs six phases, two of them in parallel from the start. Phase 1 (preprint) and Phase 2 (hiring the Wikipedia editor) begin together, because the editor needs lead time to assess the draft articles before any submission happens. Phase 3 (citation infrastructure) also runs parallel to both. Wikipedia submission does not begin until Phase 4 has produced enough independent coverage to clear a notability threshold, article by article. Touching Wikipedia before that threshold is cleared gets articles deleted, and that deletion record makes every future attempt harder.
Done
Intelligence infrastructure + vocabulary record locked
Signal Scout operational. 84-signal sweep. Maximum-by-Default named. AI Economics positioned as the commercial extension of the established Frugal AI research lineage. Four domains secured: ai-economics.com, lean-ai.com, token-ops.io, precision-built.ai.
Phase 1
Write and publish the preprint
The single most important deliverable in the plan: the academic-register practitioner paper that becomes the citation anchor for every article that lacks independent academic coverage. Submitted to SSRN and Zenodo for an immediate DOI each; arXiv pursued in parallel pending an academic endorser. Every Wikipedia editor, AfC reviewer, and journalist in later phases gets sent this document first.
Phase 2
Hire the Wikipedia editor (parallel to Phase 1)
Non-negotiable and started early, not after the preprint is done. Every team member has a disqualifying conflict of interest with these articles: the categories were named in-house and the domains are owned in-house. An independent, disclosed editor using Articles for Creation is the only path that doesn't put every article under immediate deletion scrutiny.
Phase 3
Build the citation infrastructure (parallel)
GitHub org with a dated README, Wikidata entries for all five terms, Google Scholar indexing verification, and a coordinated LinkedIn long-form series, all timestamped before any Wikipedia submission. None of this requires Wikipedia-level notability on its own; it's the scaffolding that makes the articles defensible once they're live.
Phase 4
Generate independent press and analyst coverage
Gartner and Forrester vendor briefings, VentureBeat pitch, an HBR or MIT Sloan submission, and engagement with the Linux Foundation Tokenomics Foundation community. This is the phase that actually clears each article's citation threshold: Frugal AI already clears it; AI Economics, Lean AI, and Precision AI each need two to three more independent sources before they're submission-ready.
Phase 5
Submit the five articles in sequence
Order matters and follows notability strength, not narrative preference: Frugal AI first, the strongest existing case; TokenOps second, backed by the Linux Foundation Tokenomics Foundation; AI Economics third, framed alongside DevOps and FinOps as practitioner-created categories; Lean AI fourth; Precision AI last, the most complex submission, requiring explicit disambiguation from medical and defense uses of the term. Each submission waits until the prior article has been live and stable before the next is filed. The editor executes; the team answers questions and does not touch the articles directly.
Phase 6
Post-publication defense and maintenance (ongoing)
Watchlist monitoring on all five articles, a quarterly citation log fed back to the editor, and a pursued peer-reviewed placement (IEEE Access, JIT, or MIS Quarterly Executive) once the preprint has had time to circulate. A deletion-defense plan exists for each article before it's submitted, not improvised after a deletion nomination arrives.
Ongoing
AEO structure, assessment tool, and certification pathway
In parallel with Phase 6: apply the Section 12 AEO checklist (heading structure, schema, comparison pages, monthly citation-share tracking) to every owned definitional page. Once the Wikipedia ecosystem and the citation record are stable, publish the free AI Economics Assessment built on the five Section 05 practice-layer principles, and open the practitioner community and certification pathway.
Section 12 · AEO Positioning
12
Winning the Answer Engine, Not Just the Search Result
Wikipedia is where humans verify a category. Answer engines, ChatGPT, Perplexity, Google AI Overview, Gemini, are increasingly where humans first encounter it. The same citation infrastructure built in Section 10 is the raw material; AEO is the discipline of shaping it so AI systems extract, trust, and cite it as the answer.
2.8x
Citation lift for pages with sequential H2>H3>H4 heading structure versus unstructured equivalents
83%
Of AI citations for commercial and evaluation-stage queries come from pages updated within the past 12 months
3x
More likely a page loses citations if it is not refreshed quarterly, versus a recently updated competitor page
4.4x
Higher conversion rate for visitors who arrive via an AI citation versus standard organic search traffic
Why This Is Not a Separate Strategy
AEO and the Section 10 publication strategy are the same underlying work pointed at two audiences. A Wikipedia editor and an AI answer engine are both looking for the same signal: independent, citable, well-structured, frequently updated authority. The preprint, the press coverage, and the analyst notes that build the Wikipedia case are the exact corpus a large language model retrieves from when a user asks "what is AI Economics" or "what is the difference between Frugal AI and Lean AI." Build the citation record once; it pays out in both channels.
Answer-first structure on every owned page
Lead with a direct, 40–60 word answer to the implied question before any supporting detail. "AI Economics is the discipline of managing AI cost, efficiency, and ROI at enterprise scale" belongs in the first sentence, not the third paragraph.
Sequential heading hierarchy, no skipped levels
H2 for each major concept (What is AI Economics, What is Maximum-by-Default), H3 for sub-distinctions, H4 for examples. This is the single highest-measured citation-lift signal across the 2026 AEO research.
FAQPage, HowTo, and Article schema on every definitional page
Structured data is how an answer engine confirms what kind of content it is looking at before deciding whether to extract and cite it. This is a one-time technical implementation, not ongoing work.
A visible "last updated" date and a real quarterly refresh cadence
Freshness is not cosmetic, it is a ranking input answer engines weight directly. The same quarterly review cadence that tracks inference cost (Principle 03, Section 05) doubles as the trigger to refresh the public-facing definitional content.
Entity consistency across every surface
"AI Economics," "Lean AI," "TokenOps," and "Precision AI" are defined identically on the website, the Wikipedia articles, the arXiv preprint, and the GitHub README. Answer engines cross-reference entity definitions across sources; inconsistency reads as lower confidence and suppresses citation.
A comparison page for every adjacent term
"AI Economics vs Frugal AI," "Lean AI vs Efficient AI," "TokenOps vs FinOps." Comparison and definition queries are exactly the conversational, qualifier-heavy phrasing ("what is X for enterprise," "how is X different from Y") that answer engines are built to resolve, and exactly where an unclaimed category can win the citation outright.
Monitor citation share across platforms monthly
Track whether ChatGPT, Perplexity, Gemini, and Google AI Overview cite the owned definitions when asked about AI Economics, Lean AI, or Maximum-by-Default. This is the AEO equivalent of the Wikipedia deletion-review monitoring in Section 10: the category is not secured once, it is defended continuously.
The Compounding Effect
SEO and AEO are converging, not competing: the same structural clarity, entity authority, and freshness that win a Wikipedia citation also win an AI Overview citation. A team executing Section 10's six-phase publication strategy and this section's structural checklist in parallel is not running two campaigns. It is building one citation graph that every surface, human encyclopedia, search engine, and AI answer engine, draws from to decide who gets to define the category.
Section 13 · Category Kings
13
Category Kings. Digital and Physical.
The pattern works across every industry, every era, every medium. Name the category. Name the enemy. Build the ecosystem before competitors understand what you are building. When they arrive, you are the reference point the market uses to evaluate everyone else.
Digital
Salesforce
No Software
Enemy: On-premise CRM (Siebel). When Oracle and SAP entered SaaS, both validated Salesforce's premise with every campaign dollar they spent.
$240B+
Oracle and SAP's CRM products are afterthoughts.
HubSpot
Inbound Marketing
Enemy: Outbound / interruption marketing. Every agency that rebranded as "inbound" was selling HubSpot-licensed vocabulary. The king set the curriculum.
$30B+
Every inbound agency in the world is a distribution channel.
Snowflake
Cloud Data Platform
Enemy: Legacy data warehouses. Databricks, BigQuery, and Redshift each ran massive campaigns explaining cloud-native data. Snowflake captured the premium tier.
$60B+
Largest software IPO in history at the time.
Slack
Workplace Collaboration
Enemy: Email. "Email is where knowledge goes to die." When Microsoft Teams launched free inside Office 365, Slack's brand grew. Teams entered Slack's world, not the other way around.
$27.7B
Salesforce acquisition. The category king captured it.
Zoom
Video-First Meetings
Enemy: Conference rooms and phone calls. Named the category before enterprise knew they needed it. When every competitor launched video products in 2020, the verb was already "Zoom."
$100B+
Peak valuation. "Zoom" became a verb. No competitor ever became a verb.
Stripe
Developer Payments
Enemy: Legacy payment processors. "Payments infrastructure for the internet." Made payments a developer-first product. Every fintech that followed built on or against Stripe's standard.
$65B+
Still private. Every payments startup is measured against the category they defined.
Physical
Liquid Death
Canned Water
Enemy: Plastic water bottles. "Murder Your Thirst." Sold water in a beer can. Made hydration a cultural statement. Competitors selling "canned water" after them just proved the category existed.
$1.4B
Valued at $1.4B in 2024. Selling water.
Red Bull
Energy Drinks
Enemy: Soda and coffee. "Red Bull Gives You Wings." Created the energy drink category before anyone knew they needed one. Monster and Rockstar entered and grew the market Red Bull owned.
$10B+
Still the category king 40 years later.
Yeti
Premium Coolers
Enemy: Cheap Coleman coolers. "Wildly Stronger. Keeps Ice Longer." Made a cooler a status object. Every premium cooler brand that followed validated the category Yeti invented.
$3.5B+
Turned a commodity into a $300 aspirational purchase.
Patagonia
Outdoor Activism
Enemy: Fast fashion and disposable gear. "Don't Buy This Jacket." Made environmental conviction a product attribute. Competitors who copied "sustainable outdoor gear" validated the category Patagonia invented.
$3B+
Still private. No competitor owns the activism positioning they created.
Beats by Dre
Premium Consumer Headphones
Enemy: Stock earbuds that ship with devices. Made headphones a status object and a fashion statement. Every premium headphone brand that followed competed in the category Beats invented.
$3B
Apple acquisition. Turned audio hardware into a cultural signal.
Dollar Shave Club
Subscription Razors
Enemy: Overpriced Gillette. "Our blades are great." One YouTube video, $4,500 production budget. Gillette and Schick launched subscription products in response: more free marketing for the king.
$1B
Unilever acquisition. $4,500 video created a billion-dollar category.
The AI Economics Play
When McKinsey, AWS, or Gartner enters "AI cost optimization": they will pay to educate the market about a problem you named first. Every competitor announcement is a free ad. Every analyst report validates the category. Every Big 4 white paper points the spotlight at the people who invented the space. The window is open. Zero competitors have named this category. This is the moment every framework points to.
Section 14 · The Proof
14
The Evidence the Category Is Real
Category creation is not a bet. It is pattern recognition. The signals that a $100B market is forming without a named king are already in the data.
Phase 1: Signal
Talk Track A: Fresh signal, first-mover take
Every new quantization paper, every efficiency breakthrough: publish the practitioner interpretation first. The category creator who responds to emerging signals builds the reference point. Speed is a durable moat.
Phase 2: Manifesto
Talk Track B: Name Maximum-by-Default publicly
The long-form manifesto: "We're not competing in the AI market. We're creating the enterprise category that Frugal AI's academic and policy work made possible." Explicitly frames the author as commercial category originator, building on, not replacing, the existing research.
Phase 3: Keynote Frame
Talk Track C: The $100B category nobody named yet
"Every $100B enterprise software category was created by someone who named a problem the market did not have language for." DevOps → silos. FinOps → cloud waste. AI Economics → next.
The Beachhead: ML & Data Engineering Teams First
Launch into ML architects, data engineers, and MLOps practitioners first. They feel the problem most acutely and convert each other fastest. Saturate that community before expanding to the CTO suite. Word-of-mouth velocity in a tight technical community is 20x faster than broad market advertising.
The Pattern
DevOps named silos. FinOps named cloud waste. AI Economics is next: and the category does not have a king yet. Every $100B enterprise software category was created by someone who named a problem the market did not have language for. The window is open. The pattern is proven. The opportunity is now.