What are the main business models in AI governance?

What do AI governance companies actually sell today?

AI governance companies today make money through six main models: governance software, governance bundled into larger enterprise platforms, AI security and runtime controls, managed governance services, independent assurance and certification, and vertical governance products.

That is broader than the usual idea of “AI compliance software.” The IAPP’s latest AI Governance Vendor Report already separates the market into policy and compliance, technical assessments, assurance and auditing, and consulting. Gartner has gone one step further by treating AI Governance Platforms as their own enterprise-software category and evaluating 13 vendors in its first dedicated Magic Quadrant, including IBM, ServiceNow, OneTrust, Credo AI, ModelOp and Holistic AI.

Commercially, though, we need a slightly different split because companies with similar features can make money in very different ways. Credo AI can sell an enterprise governance platform. ServiceNow can attach governance to an existing workflow estate. A cybersecurity company can sell runtime protection to the CISO. Schellman can charge for an independent certification audit. A consultancy can run the whole governance program for a client.

A customer can buy several of these at once. A bank might use a central governance platform to inventory its AI, a cybersecurity product to protect agents in production, outside consultants to design the process, and an independent certification body to verify the resulting management system.

So there is already a real AI governance market, but there is no single AI governance business model. The useful question is which part of the governance process each vendor owns and how often the customer has to pay for it.

AI governance business model	What customers pay for	Typical revenue model	Examples
Governance platform	AI inventory, approvals, risk controls, evidence and lifecycle workflows	Annual enterprise software contract	Credo AI, ModelOp, Holistic AI
Governance inside a larger platform	AI controls added to existing GRC, IT or privacy workflows	Module, bundle or account expansion	ServiceNow, OneTrust, IBM, AuditBoard
AI security and runtime controls	Red-teaming, attack prevention, monitoring and enforcement	Subscription, platform contract and potentially usage	Protect AI, Lakera, Virtue AI
Managed AI governance	Experts who implement or operate the governance program	Project fees and recurring retainers	Angularis.ai, large consultancies
Assurance and certification	Independent audits and proof that controls meet a standard	Audit, surveillance and recertification fees	Schellman, TÜV Rheinland, BABL AI
Vertical AI governance	Governance built around one regulated workflow or industry	Vertical SaaS plus specialist services	Monitaur, FairNow

Why are companies paying for AI governance now?

Companies are paying for AI governance now because their AI portfolios are growing much faster than the teams reviewing them.

Credo AI’s current State of AI Governance study surveyed 371 senior leaders and found that 60% of organizations were already deploying AI across multiple departments or company-wide, while only 4% said their governance operated at comparable scale. In the same research, 45% said manual intake alone consumed 11 to 20 hours a week.

That gap is becoming harder to absorb manually. A company once had a handful of predictive models built by a central data-science team. These days the same company can have internal models, ChatGPT-style tools, AI inside purchased SaaS products, coding assistants, custom agents, third-party APIs and employees experimenting with tools that procurement has never reviewed.

Each addition creates practical questions. Who owns the system? What data can it access? Can it make decisions? Which jurisdictions does it affect? Has security tested it? Does a human need to approve its actions? What happens when the underlying model changes?

Agents increase the workload again because the object being governed can now take actions. Governance has to follow permissions, tools, identity and behavior after deployment instead of ending when somebody approves a model card.

That operational problem is where the market is forming. Companies need a way to process far more AI without growing legal, risk and compliance teams at the same rate. That gives AI governance software a much stronger reason to exist than the vague goal of making AI “responsible.”

Is regulation really driving the AI governance market?

Regulation currently creates some of the fastest AI governance purchases, but a company built around compliance deadlines alone has a fragile business.

The EU AI Act now provides a useful real-world test. Enforcement powers are active for several parts of the law, including obligations around general-purpose AI and new transparency requirements. Certain violations can carry fines of up to €15 million or 3% of worldwide annual turnover. Companies selling into Europe therefore have concrete reasons to map systems, document responsibilities and collect evidence.

At the same time, Europe has shown how quickly the timetable can move. The AI Omnibus extended the application of major high-risk-system requirements, with Annex III rules now scheduled for late 2027 and rules for AI embedded in regulated products later still. A vendor that built its sales forecast around one imminent high-risk compliance deadline suddenly has more time to fill.

The United States gives another example. Colorado replaced its original high-risk AI framework with a revised automated-decision law that will take effect in 2027, and regulators are currently writing the detailed rules. The compliance work remains, while the exact shape and timing have changed.

Regulation is therefore a powerful sales trigger. It gets the legal team involved, creates deadlines and makes budget approval easier. The recurring business appears after that first compliance project, when the customer keeps using the product to review new systems, monitor changes, manage evidence and update controls.

A regulatory database can become outdated. A governance platform embedded in every AI approval can keep earning revenue even when lawmakers move a deadline.

If you want more recent data on this point, please see our latest AI governance market report.

Who actually pays for AI governance inside a company?

AI governance budgets today are spread across privacy, legal, IT, data, risk and security teams, which makes the category valuable but unusually messy to sell.

The IAPP’s profession research found primary responsibility split across privacy and legal/compliance at 22% each, IT at 17%, data governance at 10% and security at only 5%. Half of AI governance professionals in the study sat inside ethics, compliance, privacy or legal organizations.

The broader involvement is even more distributed. More than half of respondents expected privacy, IT, security and legal or compliance teams to take on additional AI governance responsibilities. Only 10 of 671 respondents said their organization would need no additional AI governance staff over the following year.

For vendors, this creates a very different sales process from cybersecurity. A security company usually knows that the CISO organization controls the budget. An AI governance purchase might require agreement from legal, the chief data office, enterprise risk, IT, information security and an AI product team.

It also explains why the large enterprise platforms have an advantage. OneTrust already has privacy and risk buyers. ServiceNow already controls workflows. IBM already sells into technology and risk organizations. AuditBoard already works with audit and GRC teams. They can introduce AI governance through a relationship that already exists.

Pure-play vendors have to earn a new budget line or persuade several departments that one shared platform is worth funding. The prize can still be large, especially in highly regulated enterprises, but the sales motion is inherently cross-functional.

Can standalone AI governance software become a big business?

Standalone AI governance software is now a real enterprise category, although generic governance workflow products will have a hard time staying independent.

Gartner’s first Magic Quadrant for AI Governance Platforms is a useful marker. It evaluated 13 vendors rather than treating governance as a small feature hidden inside model-management software. The capabilities Gartner now looks for include AI discovery, inventory, policy management, evidence collection, risk workflows and agent governance.

The category has also moved well beyond spreadsheets. Credo AI, ModelOp and Holistic AI all try to become a central record of AI systems across different models, vendors and business units. Once approvals, controls, incidents and evidence are attached to that record, replacing the software gets painful.

The threat comes from companies that already own adjacent enterprise workflows. ServiceNow expanded its AI Control Tower in 2026 so that it can discover and manage AI running outside ServiceNow itself. The company announced 30 new integrations covering AWS, Google Cloud, Microsoft Azure and enterprise applications such as SAP, Oracle and Workday.

OneTrust is also pushing from traditional privacy and risk management into agent inventory, runtime observability and technical guardrails. IBM combines watsonx.governance with a much broader GRC, AI and services stack.

A startup selling a registry, risk questionnaire and regulatory library now faces competitors that can bundle those functions into contracts customers already have.

The standalone opportunity still looks attractive when the product owns something deeper: cross-platform neutrality, AI-specific lifecycle automation, technical evaluations, agent governance, proprietary risk intelligence or a workflow painful enough to justify a dedicated system. Simple compliance workflow software looks much more vulnerable.

If you want more recent data on this point, please see our latest AI governance market report.

How do AI governance software companies charge customers?

AI governance software is currently sold mostly through annual enterprise contracts, while pricing is starting to follow the number of AI systems, evaluations and controls rather than just user seats.

IBM gives us one of the few public pricing windows in this market. Its watsonx.governance Risk & Compliance Basic plan starts at $3,500 a month, while Advanced starts at $6,450 a month. Those starting prices annualize to roughly $42,000 and $77,400 before larger deployments and add-ons.

IBM’s AWS offer is even more useful because it shows what the vendor chooses to meter. A 12-month $42,000 package includes five AI use cases, 12,000 evaluations and 25 concurrent users. Extra use cases can be purchased separately, and IBM also uses resource-unit pricing elsewhere in the product.

That gives us a clue about where AI governance pricing can go. Seats make sense when people are operating a workflow. They make less sense when an enterprise has 50 reviewers overseeing several thousand AI systems and agents.

A governed asset is a more natural unit. So is an evaluation, an automated assessment, a protected application or eventually a runtime transaction. The amount of governance work rises as the AI portfolio grows, even if the governance team stays the same size.

Most specialists still hide pricing behind enterprise sales conversations, which makes sense given the variation in business units, integrations, jurisdictions and deployment requirements. For now, the most plausible long-term model is a substantial annual platform commitment with expansion tied to AI portfolio size and automated activity.

Public pricing example	Starting price	Included usage	What it shows
IBM Risk & Compliance Basic	$3,500/month	Basic instance, module and user allowance	Enterprise software floor
IBM Risk & Compliance Advanced	$6,450/month	Broader capacity and functionality	Module and deployment expansion
IBM watsonx.governance on AWS	$42,000/year	5 AI use cases, 12,000 evaluations, 25 concurrent users	Asset and activity-based pricing
IBM Cloud resource pricing	Variable	Resource units consumed	Usage can sit on top of the platform contract

Why are AI governance vendors adding consulting and managed services?

AI governance vendors add consulting and managed services because many customers still lack the people and internal rules needed to make the software useful.

Installing a platform does not decide which department owns an AI agent, what counts as a high-risk use case, which evidence satisfies legal, or who can approve an exception. Companies often have to design those processes while they implement the tool.

ModelOp’s partnership with Angularis.ai shows one version of the model. The companies package ModelOp’s lifecycle-governance software with an external team that can conduct risk assessments, monitor systems and produce compliance reporting. The customer can effectively rent part of an AI risk function rather than build the entire team internally.

Credo AI has taken a related approach by adding forward-deployed AI governance experts around its platform. That looks increasingly similar to what successful enterprise AI companies do elsewhere: put experienced people close to early deployments, configure the system around real workflows, and gradually automate the repeated work.

There is a healthy and an unhealthy version of this model. Implementation services can shorten deployments, train customers and improve software retention. Managed governance can also generate recurring revenue for companies that genuinely want to outsource part of the job.

A platform that permanently needs a large consulting team for every customer has a scalability problem. Better to see services turn messy governance work into repeatable product workflows than become the main engine of revenue.

For now, hybrid software plus expertise looks especially practical. The category is still young, internal governance teams are understaffed, and enterprise processes differ enough that software alone often arrives before the organization is ready for it.

Can AI auditing and ISO 42001 certification become recurring businesses?

AI auditing and ISO 42001 certification can produce recurring revenue because companies need independent proof repeatedly, although the market should remain smaller than enterprise governance software.

There is already a clear separation between operating a governance program and verifying it. A company can use its own platform to document policies and controls, but customers, regulators or boards may still want somebody independent to say whether those controls actually work.

ISO/IEC 42001 gives that business a repeatable cycle. Schellman’s certification process uses an initial two-stage audit, followed by surveillance audits and full recertification every three years. The customer therefore returns even if nothing dramatic happens.

Other certification bodies are building the same capability. TÜV Rheinland now offers ISO 42001 certification, and large organizations have begun obtaining the credential. That creates surrounding work for readiness assessments, training, evidence preparation and remediation.

Demand for external assurance also appears to be ahead of companies’ confidence. Grant Thornton’s 2026 AI Impact Survey, cited by IBM during its Think conference, found that 78% of business executives were unsure their organization could pass an independent AI governance audit within 90 days.

The economics resemble security assurance more than SaaS. Each engagement requires skilled people, independence and professional judgment, so margins cannot scale like a pure software product. Automation can still improve the business considerably by collecting evidence continuously and reducing audit preparation.

The attractive structure is software generating evidence every day and an independent auditor checking it periodically. Those two businesses can grow together without needing the same company to own both.

Is AI security becoming the most valuable part of AI governance?

The technical security edge of AI governance has produced the strongest disclosed acquisition values we can see so far.

Palo Alto Networks paid total purchase consideration of $634.5 million for Protect AI, according to its filings. Protect AI had focused on securing AI models and applications, giving Palo Alto Networks a way to add AI-specific protection to a much larger cybersecurity platform.

Check Point later acquired Lakera, which protects generative-AI applications and autonomous agents against threats such as prompt injection, data leakage and model manipulation. Check Point’s filings put total consideration at about $201.8 million.

The pattern is still moving. Fortinet announced the acquisition of Virtue AI only days ago. Virtue AI works on automated validation, agent red-teaming and runtime protection, including monitoring autonomous systems as they use tools.

These buyers already have enormous CISO relationships and security budgets. A product that blocks an unsafe agent action or detects an attack fits naturally into those budgets because the customer can connect the product directly to production risk.

Pure policy-management software has a harder value story. It may reduce legal exposure or save review time, which can be substantial, but the result is harder to observe minute by minute.

We should still keep AI security and AI governance conceptually separate. Governance decides what an AI system should be allowed to do and what evidence the organization needs. Security catches malicious or unsafe behavior technically. Commercially, though, the boundary is getting thinner because enterprises increasingly want governance decisions enforced in software.

The acquisition evidence currently favors vendors close to that enforcement layer.

If you want more recent data on this point, please see our latest AI governance market report.

Are runtime guardrails a better business than AI governance dashboards?

Runtime AI governance currently has better expansion economics than a dashboard because the product can keep working every time an AI system or agent acts.

A traditional governance platform might review a use case when it is created, again when it changes and periodically during its life. Runtime controls can inspect behavior continuously. A busy application may generate millions of opportunities to apply the same policy.

OneTrust is moving aggressively in that direction. Its current AI governance product can discover agents, monitor runtime behavior, convert policies into machine-readable controls, filter prompts or outputs, restrict actions and enforce policies in MCP environments. Holistic AI has also moved toward runtime enforcement alongside its inventory, compliance and testing products.

That creates several possible expansion units beyond seats: protected applications, agents, evaluations, requests or enforcement volume. We do not yet see enough public contracts to say that consumption pricing has become the standard, so that remains a likely economic direction rather than an established market fact.

Runtime governance also creates better evidence. A questionnaire tells the risk team what an AI application is supposed to do. Live telemetry shows which tools it used, what data it touched and whether it tried something outside policy.

The technical burden is much higher, though. An inline control has to be fast and reliable. False positives can break real workflows. An enforcement layer that goes down can become an operational problem itself.

The strongest governance companies do not necessarily need to build every firewall themselves. Owning the policy and evidence layer while connecting to multiple technical enforcement systems could be just as valuable. The commercial step that counts is getting governance closer to actual AI behavior instead of leaving it inside periodic reviews.

Can AI inventory and third-party AI risk become standalone businesses?

AI inventories and third-party AI risk are valuable entry products, but by themselves they already look too easy for larger governance platforms to bundle.

The problem is genuine. Credo AI’s current governance research found that 40.7% of surveyed senior leaders considered third-party AI systems and vendors their biggest governance challenge, ahead of regulatory compliance and manual workflows.

That makes sense when we look at how companies use AI now. A business can control the models its own engineers build and still inherit hundreds of AI features through SaaS products, APIs and software vendors. The company often carries the compliance or reputational risk without seeing the vendor’s training data, internal evaluations or model updates.

Inventory is the first step because a company cannot govern AI it does not know it has. The problem for a startup is that inventory is becoming standard functionality.

IBM recently added a Governance Graph that maps relationships among AI use cases, agents, models, MCP servers, tools, risks and controls. IBM also connects third-party risk information from companies including Dun & Bradstreet, RiskRecon, SecurityScorecard and RapidRatings. Other major governance vendors are building similar discovery and registry functions.

A standalone scanner can still win customers quickly, particularly through shadow-AI discovery. Its long-term position is weaker if the product stops after finding the asset.

The valuable part begins once that asset enters a permanent record. The platform can assign an owner, assess the vendor, determine which controls apply, request evidence, trigger approvals, watch for changes and reopen the review later.

Third-party AI risk could eventually support a more unusual model if vendors begin reusing verified assurance data across customers. Thousands of companies currently ask the same major SaaS providers similar questions. A shared, continuously updated evidence network would create more defensibility than another vendor questionnaire.

For now, discovery and third-party risk look strongest as high-value modules inside a broader control plane.

Does vertical AI governance work better than horizontal platforms?

Vertical AI governance can beat broad platforms when customers need domain-specific evidence and testing that a generic GRC product cannot easily copy.

Financial services is an obvious example because banks already have formal model-risk processes. Insurance has similar requirements around pricing, underwriting and claims. Healthcare adds clinical risk, privacy and regulated products. Hiring introduces employment discrimination rules and specialized bias testing.

Monitaur has leaned into this logic by focusing its governance and model-assurance work heavily around regulated decisions, including insurance. The product can speak the language of model owners, validators and risk teams rather than asking every customer to design the workflow from scratch.

FairNow followed a similar route around practical AI compliance and became particularly visible in employment-related governance. AuditBoard’s decision to acquire FairNow shows how a focused AI governance product can become valuable to a much larger GRC platform.

The advantage comes from depth rather than simply loading sector regulations into a database. A strong vertical product knows which tests should be run, which documentation reviewers expect, what an acceptable exception looks like and how existing domain systems need to connect.

Horizontal platforms still have the larger theoretical market because almost every large company will govern AI across several departments. They can also add industry templates over time.

For a new company, however, “AI governance for everyone” is a difficult pitch when ServiceNow, IBM, OneTrust and other incumbents already cover the horizontal market. Solving one expensive and highly specific governance problem can create a much clearer route into customers.

Vertical AI governance looks particularly attractive where the same workflow recurs across many companies and getting it wrong can cost real money.

If you want more recent data on this point, please see our latest AI governance market report.

Are AI governance agents changing how vendors make money?

AI governance agents are starting to turn governance software from a workflow tool into something closer to paid labor replacement.

Credo AI made its Govern AI Assistant, or GAIA, generally available in 2026. The agent can help intake AI use cases, identify risks and suggest relevant controls using the company’s governance libraries. Credo AI is also preparing interfaces that allow customer-built agents to interact with the governance platform.

ModelOp launched MADE in June 2026, an agentic framework designed to move AI projects through enterprise delivery and governance processes. Its product can orchestrate reviews across existing systems while tracking models, agents, costs, risks and business value.

The commercial change is subtle but important. Traditional GRC software mainly helps employees perform governance work. An effective governance agent can perform parts of that work itself.

A company reviewing 5,000 AI systems may eventually care less about how many employees have a software login and more about how many assessments, tests, reviews and evidence tasks the platform completes automatically.

That could push pricing toward a hybrid model: a predictable annual platform contract, followed by additional charges as the amount of automated governance rises. IBM’s existing use-case and evaluation-based pricing already shows that enterprise buyers can accept activity-based units alongside normal software contracts.

We should be careful here because the agentic products are still new and vendors disclose little about customer-level economics. There is not enough evidence yet to claim that AI governance has already shifted to outcome pricing.

The direction is clearer than the endpoint. If governance teams remain relatively small while the number of AI systems keeps climbing, vendors that automate actual work have a better expansion mechanism than vendors that sell more dashboard seats.

Where is AI governance consolidation heading?

Recent AI governance M&A points toward GRC platforms owning organizational governance while cybersecurity platforms absorb the technical enforcement layer.

AuditBoard’s acquisition of FairNow is the cleanest example on the GRC side. FairNow brought AI inventory, assessments and compliance workflows into a company that already sells audit, risk and compliance software. The buyer can now treat AI as another enterprise risk inside an established platform and an existing customer relationship.

Cybersecurity has followed a more aggressive route. Cisco acquired Robust Intelligence after the startup built AI red-teaming and firewall technology. Financial terms were officially undisclosed, although 451 Research estimated the transaction at around $350 million.

As we saw above, cybersecurity buyers have continued purchasing AI-native protection companies since then, and Fortinet’s very recent Virtue AI deal shows that the consolidation is still active.

Governance workflows fit naturally beside GRC, privacy and audit. Red-teaming, attack detection, agent permissions and runtime protection fit naturally beside cybersecurity.

That leaves independent governance companies in the middle. Their best defense is becoming the neutral layer that works across all of those systems.

A large enterprise may use several cloud providers, several foundation-model companies, multiple cybersecurity products, internal models and hundreds of third-party AI tools. It still needs one place to understand which AI exists, what each system is allowed to do, who approved it and which evidence supports that decision.

Neutrality can therefore be a real advantage. A governance company does not need to replace ServiceNow, Palo Alto Networks or AWS if it becomes the layer connecting their data into one enterprise-wide decision process.

The weaker pure plays will probably be absorbed or bundled away. The strongest ones have a chance to become the system of record sitting above an increasingly fragmented AI stack.

So what are the main business models in AI governance today?

The strongest AI governance business models today are recurring governance platforms and technical runtime controls, with managed services, assurance and vertical products forming durable businesses around them.

The core software opportunity is the AI control plane: one place where companies inventory AI, assess it, approve it, assign controls, collect evidence and keep following the system after deployment. This can support high-value annual contracts and expansion as the customer adds more AI.

Runtime governance and AI security may have even stronger economics in some cases because the product operates continuously and sits close to a large existing cybersecurity budget. The acquisition market has already assigned substantial strategic value to that technical layer.

Managed governance also makes sense right now because enterprises have more AI work than experienced governance people. We expect the strongest service offerings to become increasingly software-enabled, with humans handling judgment and automation handling repetitive intake, evidence and monitoring.

Independent auditing and ISO 42001 certification should develop into a meaningful recurring assurance market. Vertical products can build excellent businesses where one industry requires specialized testing and evidence that horizontal platforms struggle to reproduce.

The weakest model is generic compliance software built mainly around regulatory content, questionnaires and static policies. Those features are becoming widely available, regulations keep changing, and large GRC vendors can bundle them cheaply.

What we are seeing now is a shift from governance as documentation toward governance as an operating layer. The businesses with the best prospects are the ones that get paid whenever AI is added, reviewed, changed, tested or used. That creates recurring work even after the first regulatory deadline has passed.

AI governance business model	Recurring revenue potential	Defensibility	Current position
Enterprise AI governance control plane	Very high	High if deeply integrated and vendor-neutral	Strongest standalone software model
AI security and runtime governance	Very high	High where controls sit in production	Strongest technical model and strongest M&A evidence
Governance inside GRC, IT or privacy platforms	Very high	Very high distribution advantage	Likely to capture a large share of horizontal demand
Managed AI governance	High when sold as an ongoing service	Medium	Strong today because internal talent is scarce
Independent assurance and ISO certification	Naturally recurring	High trust barrier, lower software scalability	Durable specialist business
Vertical AI governance	High inside the right niche	High when expertise and testing are hard to copy	Attractive startup route
Compliance-only tools	Moderate	Low	Most exposed to bundling and regulatory changes

If you want more recent data on this point, please see our latest AI governance market report.

Who is the author of this content?

NEW MARKET PITCH TEAM

We track new markets so founders and investors can move faster

We build living "market pitch" documents for emerging markets: AI, synthetic biology, new proteins, and more. Instead of outdated PDFs or hallucinated LLM answers, our clients get a clean, visual, always-updated view of what's really happening: key players, deals, regulations, and signals that matter. Learn more about us.