01 Causal AI
Discovery from data, or extraction from experts. The field splits here.
“Causal AI”, software that explains why outcomes happen, not just what correlates with what, is now a real market, worth tens of billions and growing fast. But it splits along one line. Most of the funded field tries to discover cause and effect automatically, from a company’s own data. Rung3 takes the opposite view: the model of how a business works lives with the people who run it, and has to be drawn out of them, not mined from a data warehouse. The tiers below run from most directly competitive to merely adjacent.
Expert-built model tools & their consultants direct
These vendors build models the same way Rung3 does, from expert knowledge, not raw data, and most also sell consulting and training. This is the real head-to-head.
Bayesia / BayesiaLabFrance · USThe closest competitor of anyone here: draws knowledge out of experts, runs cause-and-effect scenarios, publishes models as simple web tools. Now goes further, a dedicated web-based elicitation tool (BEKEE) captures each expert's view individually and compiles conflicting perspectives into one model, and the platform explicitly treats domain expertise, data, and LLM-derived knowledge as three separate, named input sources. The only rival already pairing structured elicitation with an AI layer this deliberately. One difference still matters, though: BayesiaLab works by bucketing continuous quantities, price, dose, blood pressure, into ranges, and leans toward discovering structure in data. Rung3’s tooling keeps those quantities continuous, modeling the actual mechanism rather than an approximation of it.
AgenaRisk / AgenaUKA specialist shop building expert-driven models for risk, safety, and legal reasoning, strong exactly where data is thin, which is Rung3’s territory too.
BayesFusion / GeNIeUSModeling software plus training, consulting, and custom development.
Hugin ExpertDenmarkLong-established modeling engine with an applications and consulting practice.
Norsys / NeticaCanadaEstablished modeling software, widely used in teaching and applied work.
Lumina / AnalyticaUSDecision-modeling software, a neighboring tradition with the same “model the decision” posture.
Bayes ServerUK · my tool of choiceNot a rival, this is the engine Rung3 builds in. Its makers do offer support and consulting, so worth tracking as a service provider. Rung3 delivers through its R and Python APIs.
Rung3’s edgeApart from Bayesia, none put a plain-language AI front end on the model, and none hand the client a connected library of models they own and keep. The tool vendors sell software; Rung3 sells the modeling and the finished, usable result.
Causal-AI platforms, automated discovery different problem
Well-funded software that hunts for cause and effect automatically inside a company’s data, then simulates decisions at scale. Broad and heavily capitalized, and the tier keeps thickening: one 2026 industry estimate puts more than 60 companies now active in productized causal decision intelligence. But in Rung3’s view, most are still aimed at a different problem: spotting patterns in data, not capturing the business logic the data never recorded.
causaLensUKOne of the category’s flagships; “decision intelligence” built on finding cause and effect in data automatically, then testing decisions on it.
AlembicUSCause-and-effect AI for marketing and revenue; real-time scenario simulation, heavily funded (raised $145M; NVIDIA-backed). Strong on marketing-mix.
CausifyUSEnterprise causal AI for energy, industrial operations, and capital markets; automated causal discovery across 50+ data connectors, positioned explicitly against boutique consulting, their own language is “not a consultancy engagement, production software.” Its CEO is also a leading voice for pairing causal reasoning with LLM agents, the same thesis EARA is built on, made here by a funded platform instead of a named practitioner.
AitiaUSAutomated cause-and-effect discovery at scale, rooted in biomedicine, extending to enterprise decisions.
Xplain DataGermanyFinds cause and effect in event and customer data; pairs a discovery tool with an analytics store.
Causality LinkUSBuilds cause-and-effect maps from text, aimed at finance and markets.
HowsoUSTransparent machine learning with “what-if” and attribution features.
VedraiItalyWhAI, a no-code platform for simulating business decisions and their effects.
OthersvariousCausa (UK), CausaAI (Netherlands), Parabole.AI (US), Actable AI (UK), Data Poem (US), Geminos (US), a long tail of cause-and-effect analytics vendors.
Rung3’s edgeThe argument against this whole tier in one line: you can’t learn how your business works from your business’s data alone. Their automation shines where the answer is already in the data, and misleads where it isn’t.
04 The free toolkit
Open-source frameworks & the firms using them free toolkit
The free libraries any analytics team can download. They lower the barrier to entry, which means the toolkit is never the advantage, the expertise on top of it is.
Microsoft PyWhyUS · OSSThe de-facto free toolkit for measuring cause and effect; the most common thing a generalist competitor reaches for.
CausalNexQuantumBlack / McKinseyA free model-building library blending expert knowledge with data, and a sign that McKinsey’s analytics arm is in this space.
CausalMLUber · OSS“What works on whom” methods, widely used in marketing and product.
R & Python BN libsOSSbnlearn, pgmpy, pyAgrum and peers, the open toolchain underneath much of this work.
Rung3’s edgeAnyone can install these; almost no one can supply the expert-built models, the validation, and the AI-explained library that go around them.
05 Scale & brand
Big-firm causal practices generalist
The large consultancies and platform vendors with cause-and-effect teams folded into bigger engagements. They win on brand, scale, and existing relationships, rarely on depth, and never cheaply.
QuantumBlackMcKinseyMcKinsey’s AI arm; cause-and-effect work folded into enterprise engagements.
BCG XBCGBCG’s build and data-science unit, with cause-and-effect and decision offerings.
IBMUSCause-and-effect frameworks and managed services; ongoing research (incl. with MIT).
DynatraceUSAutomated root-cause for IT operations, a different vertical, but “causal” in market language.
Cognizant · DataRobotUSCause-and-effect and decision features inside large analytics and services portfolios.
Rung3’s edgeA focused specialist and a named expert, delivering a model the client’s own team can run and keep, not a slide deck and an open-ended retainer.
06 One domain, crowded
Marketing-mix modeling vendors overlapping vertical
Measuring what marketing spend actually drives is where “causal” has gone mainstream and competition is thickest, a field Rung3 works in too. Worth watching both as rivals and as the benchmark Rung3’s marketing work is judged against.
Alembic · IncrmntalUS · IsraelMarketing-mix and attribution platforms with real-time and incrementality angles.
Recast · Data PoemUSMarketing-mix specialists positioning against legacy approaches.
Meridian · RobynGoogle · Meta · OSSFree marketing-mix libraries from Google and Meta, the baselines any engagement now has to beat or build on.
Rung3’s edgeMarketing-mix as one model in a larger connected library, built to answer “what if we shift spend” properly, and explainable to a CMO through the AI layer.
07 The real day-to-day
Independent causal-inference boutiques direct
The hardest group to pin down, and the one Rung3 meets when competing for work: solo experts and small shops offering bespoke cause-and-effect modeling, often academic spin-outs or single-industry specialists in areas like health economics and actuarial work.
Independent practitionersfragmentedIndividuals and small consultancies with deep cause-and-effect credentials, usually anchored in one industry. No directory; found by reputation and referral.
Academic-adjacent groupsvariousUniversity labs and spin-outs offering applied modeling, credible, but rarely packaged as something the client can keep and maintain.
Rung3’s edgeThe same depth, turned into a product: a tested, reusable model library with a plain-language AI front door and a method documented well enough to hand over, not a one-off study that walks out with the consultant.
08 Where Rung3 sits
The field divides cleanly. The funded mainstream automates discovery from data, a different bet. The big firms sell scale. The tool vendors sell software. The independents sell depth, but not something the client keeps. Rung3 holds the corner almost no one else does:
Built from expert knowledge: not mined from data, on the conviction that the answer isn’t in the data.
Delivered by a focused specialist: not a subscription or a big-firm team.
A tested model the client owns and keeps: every assumption traceable.
Connected into a library and explained by AI in plain language: the newest advantage; only Bayesia is visibly reaching for the same thing.
Across industries: healthcare, insurance, finance, marketing, policy, one method, many models.
The open ground: no competitor combines expert-built models, an AI-explained model library, focused boutique delivery, and a finished tool the client keeps. The closest threat is Bayesia: the one name to watch, though even it discretizes the continuous mechanisms Rung3 keeps intact.
A different axis: set against the agentic-AI mainstream rather than the causal field, the contrast is sharper still. Agentic systems put the reasoning inside the language model, planning and deciding in a vector no one can inspect, tools called around it. Rung3 inverts that: the language model is a bounded linguistic interface, and the reasoning is delegated to explicit models that are inspectable, testable, and maintainable. When a recommendation is wrong, you open the model and see why, you do not re-roll a prompt.
Compiled from current market research (market analyses and vendor sources), 2026. Vendor categories blur, several firms could sit in two rows, and the independent-boutique tier is fragmented and largely undocumented, so it is necessarily described rather than listed. Treat as a living map, not a closed census.