Insights
AI data market
290 guides · page 3 of 7
The future of deidentified work data: four scenarios, eight dated signposts
Licensing past work without naming workers: four scenarios for deidentified work data, eight dated signposts, the risks and the diligence questions to ask.
Human correction data future: why rarer fixes may be worth more
The human correction data future, tested: four scenarios for who holds correction records as agents take longer tasks, scored on seven dated signposts.
Computer use agent data economics: screens are cheap, checks are not
Computer use agent data economics from two public benchmarks: 1,181 tasks counted by starting state, success check and refusal case, and what to ask a seller.
Cross-application agent data: the gap is real; is it a data gap?
Cross-application agent data, tested: agents fail far more often on multi-app tasks, but is the gap about data? A gap ledger and six diligence questions.
Energy maintenance planning data: three rules make deferrals checkable
Energy maintenance planning data is worth backing where a rule makes a deferral checkable. A five-segment scorecard, its missing proofs and the risks.
Engineering change data opportunity: the value is in linked chains
The engineering change data opportunity, tested: 0.9% of 2025 Linux kernel commits linked report, test and review. A five-source scorecard, risks and a pilot.
An AI data investment committee market map that runs on two numbers
An AI data investment committee market map that places each revenue line by labor per sale and rights controlled, with metrics, filing evidence and gaps.
Human-agent handoff data: the opportunity sits in the outcome gap
Is human agent handoff data an opportunity? Four agent stacks now log the pause and the reply, not the outcome. An input map, value chain, risks and questions.
Real-world AI learning inputs market: a five-input map with falsifiers
The real-world AI learning inputs market for investors: five input types, each tested on observed demand, rights-cleared supply and a credible substitute.
Synthetic data business model comparison: who cheap generation helps
A synthetic data business model comparison for investors: tooling, services, licensed seeds, validators and finished datasets, and the test that prices each.
Fresh vs historical dataset market sizing: year one is not a run rate
Fresh vs historical dataset market sizing as a stock and a flow: why archive licensing runs down, what refresh recurs, and a six-year model with sensitivity.
Historical business archive data TAM: five filters, a stock not a flow
Historical business archive data TAM from an official firm count: five filters, the full arithmetic, a sensitivity chart and why exclusive licenses deplete it.
Enterprise data licensing investment thesis: the 7-job missing middle
An enterprise data licensing investment thesis built on the seven jobs between company records and an accepted AI dataset, with a falsifiable evidence map.
Source system dependency and the data moat: a five-shock stress test
Source system dependency can turn a data moat into a lease. Test five shocks, from API terms to retirements and buyer concentration, and score what survives.
Permission, not compute? A six-constraint test for the AI data rights bottleneck
A six-constraint test for the AI data rights bottleneck: when permission, not compute, limits AI models, and the funnel metrics that show it either way.
Next big thing in AI data? A call, watch or pass evidence checklist
Next big thing in AI data? Grade any market on buyer breadth, renewals, utility against substitutes, rights-cleared supply and unit economics before calling it.
Logistics exception AI data market: the event feed is not the asset
The logistics exception AI data market rests on recovery decisions and costs, not tracking events: a tally of 116 standard codes and a six-type scorecard.
Failed decisions as training data: the value is in the documented fix
Failed decisions training data value sits in the documented fix, not the failure. A six-claim thesis map with evidence columns, falsifiers and risks.
Serviceable enterprise data supply: five filters on official counts
Serviceable enterprise data supply, modeled: official company counts, five filters with labeled ranges, the arithmetic, and the input to measure first.
SourceX licensed business data report: five ledgers, no rankings
How the SourceX licensed business data report will work: five separate ledgers, minimum cell sizes, approval gates and no rankings. No results are published.
Proprietary dataset replacement analysis: three routes a buyer tries first
Proprietary dataset replacement analysis for investors: test public data, synthetic substitutes and direct sourcing, and find the price ceiling each sets.
Agent escalation data value: one stop skill is cheap, one is scarce
Agent escalation data value splits in two: rule boundaries anyone can write, and judgment calls only outcome records teach. A threshold model shows the gap.
Thomson Reuters AI data business analysis: licensing vs subscriptions
Thomson Reuters AI data business analysis from its own filings: transactional AI content licensing, AI-enabled subscriptions, and the revenue fields left blank.
Turing AI data business model: where talent access ends, data begins
Turing AI data business model, traced through Turing's own terms: where client-owned work ends, where data assets could begin, and which facts it leaves out.
Qualify once, sell once: the data procurement platform business model
A data procurement platform business model can copy G-Cloud's supplier vetting and standard terms, but not its economics: 56 checks once, 139 per listing.
Workflow versus document dataset value: a path test on 14,521 real cases
Workflow versus document dataset value, measured on two public event logs: when a workflow dataset shows paths, timing and choices that no document holds.
AI data investment demand thesis: who pays, for which task, judged how
Test an AI data investment demand thesis before calling it a market: name the budget owner, the task and the acceptance process, then check the dated evidence.
Benchmark saturation data opportunity: 2 of 7 tests still rank models
Is benchmark saturation a data opportunity? In OpenAI's own table, two of seven tests still separate its models. A worksheet tells used-up tests from real gaps.
Enterprise workflows as an AI training thesis: 3 things manuals omit
The enterprise workflows AI training thesis, tested: what operating logs teach that manuals omit, a falsifiable evidence map, and signs the thesis is wrong.
Four AI data business downside risks that time will not repair
AI data business downside risks sorted by what time cannot fix: lost rights or supply, legal bars, a buyer substitute and concentration, with evidence to check.
Agent environment or dataset: 5 questions that set the business model
Agent environments and datasets carry different duties after delivery. Five questions classify any offering and show who funds resets, verifiers and upkeep.
Portable data competitive advantage: which of seven moat layers survive
Portability rules let customers take raw records elsewhere. Test seven layers of portable data competitive advantage, each with a result that would disprove it.
What the food supply chain data market can offer AI evaluation
An investor scorecard for the food supply chain data market: six record families rated on outcomes, rights, preparation and public exposure, with gaps named.
Handshake AI business model analysis: a talent network buying archives
Handshake AI business model analysis from Handshake's own pages: what its expert network, data partnerships and contractor terms disclose, and what is unknown.
Dataset value after a model breakthrough: what decays, what compounds
Model breakthrough dataset value decay is uneven: solved-skill examples fade while tests and outcomes hold. A five-part split, four scenarios, five signposts.
AI data buyer demand volatility: three shocks and what each strands
AI data buyer demand volatility, stress-tested: how category shifts, model upgrades and budget freezes reach a data supplier, how fast, and what survives.
Saturated AI benchmarks and future data: who pays for the next test
When public AI benchmarks saturate, evaluation demand moves to hidden, refreshed or task-specific tests. Four scenarios and a dated scorecard for investors.
Healthcare administration data investment: six workflows scored
Healthcare administration data investment, tested on six nonclinical workflows: which have outside graders, standard formats and a workable permission path.
The AI labs in-house data sourcing threat: a make-or-buy test by data type
How real is the AI labs in-house data sourcing threat? A make-or-buy test by data type, with coordination costs, vendor neutrality and dated signposts.
AI data demand downside scenario: 4 futures, 8 dated signposts
An AI data demand downside scenario, tested: four futures for licensed data budgets, eight dated signposts and a scorecard of where the evidence points now.
Business archive data preservation economics: three clocks, four futures
Business archive data preservation economics: the three decay clocks that destroy AI value, four futures for preserved records and six dated calls to grade.
Standardized data rights market future: cheaper deals, thinner moats
Standardized data rights market future: a six-layer friction ledger and four scenarios showing which deal costs fall, whose margins thin and what stays bespoke.
Software data access restrictions: four scored AI market scenarios
Software data access restrictions now bar outside AI apps from training on vendor API data. Four scored scenarios show what that means for the AI data market.
Evaluation versus training data: a flywheel test of the market's future
Evaluation versus training market future: a six-trigger flywheel test of when evaluation could outgrow training data licensing, and what would disprove it.
The AI data commoditization scenario: what the tenth seller still owns
An AI data commoditization scenario for investors: when many firms sell the same records, which layers lose value, which keep it, and 7 dated signposts.
Unavailable data and AI market constraints: six routes when records never sell
Unavailable data is an AI market constraint, not a dead end: six permitted routes, what each recovers, who captures the value and five dated forecasts.
Legal operations AI data opportunity: what 162 benchmark tasks leave out
The legal operations AI data opportunity is the process layer benchmarks skip: a six-workflow scorecard, privilege limits, risks and diligence questions.
Vertical or horizontal AI data business? Price the next category first
Compare a vertical and a horizontal AI data business on reuse, category concentration and the cost of entering the next industry, on one per-category model.