Bain & Company estimates that agentic AI could create a $100 billion US market for software that automates the coordination work between enterprise applications. The opportunity is not primarily another system of record. It is the manual interpretation, checking and follow-up that employees perform because existing systems do not complete an outcome together.
Workers routinely move between ERP, CRM, support platforms, vendor tools and email. They compare structured records with unstructured messages, decide whether to approve or escalate, and then update several systems. Bain argues that agents can turn much of this labour into software spending.
Vendors are already capturing an estimated $4 billion to $6 billion, leaving more than 90% of the potential US market untapped. Canada, Europe, Australia and New Zealand could add a similar amount.
The opportunity lives between systems of record
Traditional robotic process automation works well when rules are stable and inputs are predictable. It struggles when a workflow contains ambiguity, exceptions or information distributed across several applications.
Agentic AI can interpret messages, retrieve context, coordinate actions and operate inside policy guardrails. A supplier enquiry, for example, may require checking a contract, inventory, an invoice and previous correspondence before deciding whether to respond or escalate.
Bain calls cross-workflow decision context the next competitive advantage for SaaS companies. Established vendors already own integrations, domain expertise and customer trust. AI-native challengers can attack the same opportunity with products designed around completed outcomes rather than screens and seats.
The largest addressable areas reflect both workforce size and technical suitability. Sales represents about $20 billion, while cost of goods sold and operations account for roughly $26 billion. Customer support and R&D or engineering have the highest estimated task automation potential, at 40% to 60%.
Automation potential depends on verification and risk
Not every workflow deserves the same degree of autonomy. Bain identifies output verifiability, consequences of failure, digitised knowledge, process variability and integration complexity among the factors that determine whether an agent can take over.
Code that compiles, an invoice that reconciles and a support ticket that stays resolved provide clear feedback. Legal advice, tax filings and security incidents carry greater consequences and more subjective judgement, so they require tighter human supervision even when models appear capable.
Data can be the limiting factor. Agents need machine-readable records, documented policies and access to the informal decision logic that often exists only in experienced employees’ heads. Authentication, APIs and exception handling become harder as a workflow crosses more systems.
That explains why the most valuable opportunities may also be the most difficult. No single platform controls the result, so the agent must assemble context and authority across organisational boundaries without losing accountability.
SaaS economics shift from access to outcomes
When an agent resolves an issue or processes an invoice, pricing by user seat becomes less meaningful. Bain expects greater use of consumption- and outcome-based models tied to work completed.
That transition changes product strategy and sales incentives. Vendors must identify automatable subprocesses instead of labelling an entire department as an AI opportunity. They also need data tied to outcomes, cloud-native orchestration, AI engineering talent and enough funding to absorb model inference costs.
Companies can build internally, acquire missing capabilities or partner. Bain points to different approaches including ServiceNow’s acquisition of Moveworks and Salesforce’s partnership with Workday. The right choice depends on the workflow context and data advantage each company already possesses.
Strong foundations will include machine-readable hand-offs and records of each agent’s decisions and outcomes. Those traces improve evaluation, governance and future product performance.
The $100 billion estimate should be treated as a model of potential rather than guaranteed revenue. Businesses will not transfer labour budgets to software unless agents produce dependable results at a lower total cost.
Still, the direction is clear. The next SaaS battleground is not another interface placed beside existing applications. It is the intelligence that connects them, resolves their exceptions and turns fragmented information into a verifiable business outcome.
