At Workday Rising 2026, the Randstad Digital team, along with every other attendee, met Sana. As Workday describes it, Sana is a “single agentic experience that knows your company, learns how you work, and earns more autonomy as it goes,” serving as the AI front-end of the platform and its agentic AI layer. Where earlier assistants stopped at answering questions, Sana is designed to complete the work behind them. An employee states what they need in plain language, and Sana can determine the steps and carry them out across HR, finance and IT, whether it’s a leave request, a password reset or an onboarding checklist.
For health plans, the appeal is evident. Member services, claims and care management teams, working under persistent turnover and pronounced seasonal hiring peaks, generate a steady volume of the repetitive requests Sana is meant to handle. This shift toward AI-driven execution is a fundamental one, though, and it raises questions about data security, HIPAA compliance and consumption-based pricing that are best settled before launch so payors can make the most of their Workday investment.
1. how will sana change the cost of workday?
Workday prices its AI through Flex Credits, a usage-based model in which organizations pay for the actions agents complete, not only for the number of seats. Sana for Workday and the Sana Self-Service Agent come with an initial allocation of credits, so the question is what happens as usage grows beyond it.
the risk
Agents that run in the background across tens of thousands of employees can generate a large volume of activity, and without rate limits and sound architectural controls, that activity can cause customers to question what the impact on their budget could be. Overlap is a second exposure. Organizations also risk paying double when multi-query integrations overlap with existing tools (e.g., triggering queries in Microsoft Copilot that route downstream into Sana).
the action
Before turning on autonomous, multi-step agents, put three things in place. Establish consumption monitoring so you can see usage as it happens, assign budget limits by department and audit where your existing licenses duplicate what Sana will do.
2. can we trust it with regulated data, and where should people stay in charge?
Sana operates within Workday’s role-based access controls, the permissions that determine who may see and do what. Workday’s Trusted Agent Platform applies the same principle to software agents, which run within the permissions, approvals and audit trail of the user they act for. That is a sound foundation, and it’s the reason a health plan can reasonably consider autonomous workflows at all.
the risk
A health plan holds sensitive employee data and member information. It cannot accept data leakage, fabricated answers or visibility granted to the wrong people, and a gap in any one of these carries regulatory consequences as well as operational ones. Workday’s controls govern what happens inside Workday. They don’t tell you how well your own access tiers are set up, or how protected health information is handled once an agent begins working across systems.
the action
Before connecting anything autonomous, re-validate your core data quality, access tiers and security policies. Map where protected health information sits and which of those systems an agent could reach, then have your security and compliance teams test those paths in a non-production environment first. Enforce a strict three-tier oversight model:
- In the loop: A person makes the decision. Hiring, salary adjustments and executive approvals belong here.
- On the loop: The agent acts while a person monitors thresholds and resolves exceptions. Expense monitoring and other high-volume tracking are the familiar examples.
- Out of the loop: The agent runs unattended. Background data integrity checks and routine transaction sampling fit here.
Agree on these boundaries internally and document them before deployment, so that they reflect considered policy and not a reaction to an incident.
3. are our own processes ready for an agent to run them?
Technology creates capability but does not by itself create results. The point applies directly to Sana. An agent can only be as effective as the workflows and data beneath it, and it has no means of compensating for a process that was already difficult to follow.
the risk
Many health plans carry years of accumulated complexity in their HR and finance operations: approval chains that persist out of habit, compensation rules layered on by successive reorganizations and master data that differs from one system to the next. Placing an autonomous agent on top of that structure does not repair it. The agent executes the same flawed steps faster and at greater volume, and it produces exceptions that no one planned for. Plans still consolidating regional businesses after a merger tend to carry more of this than most.
the action
Treat the announcements as an occasion to optimize your ERP, the core HR and finance system, before asking an agent to run it. Eliminate approval steps that no longer serve a purpose, rationalize legacy compensation rules and correct master data at the source. Simplify the core workflows first, and then decide which of them an agent should own.
4. will it work across the rest of our systems?
Workday now lets outside AI agents work with its data through the Model Context Protocol, a standard that lets AI tools read from and act on other software. Agents can use 275 tools that read data and 54 that write or delete it, and they can run on models from Anthropic, OpenAI, Google, Meta and others.
the risk
There are two exposures. The first is shortcuts. Workday warns that connecting an agent through a single, broad service account, or copying Workday data into an outside data lake, can bypass approvals and leave security rules behind. The second is your other systems, such as ServiceNow, Salesforce, Microsoft Entra ID, electronic health records and payroll. An agent that acts in one without regard for the rest can leave conflicting records, and Workday’s controls don’t reach systems it doesn’t run.
the action
Map the end-to-end data pipelines connecting Workday to external enterprise platforms. Validate integration readiness and identity governance before deploying write-enabled external agents into production environments.
5. can it help us plan our workforce and costs together?
Workday has two tools here, and they do different jobs. Decision Intelligence can put headcount requests, financial actuals and payroll budgets in one place, so a leader can ask a question in plain language and model the answer. It is available now to customers of Workday’s HR and finance products. Work Intelligence can break each role into its component tasks and shows which could be automated, which an agent could support and which should stay with people.
the risk
Health plans are dealing with labor shortages, high turnover and thin margins. When staffing decisions are made apart from the financial picture, the result is costly backfills and misaligned labor spend. The pressure peaks during Medicare Annual Enrollment and open enrollment, when contact center capacity has to scale fast and budgets are closely watched.
the action
Leverage natural-language scenario modeling to align headcount planning with dynamic financial targets. Identify which operational and corporate tasks can be augmented or automated, freeing up staff capacity to focus directly on patient/member outcomes.
conclusion.
The integration of agentic workflows is a massive operational leap. Workday is building the engine, but enterprise execution requires careful calibration. At Randstad Digital, we work closely with payors to translate these platform updates into secure, financially sound realities. The goal is not just to turn the AI on, but to build an architecture that supports your organization’s broader digital transformation. To get started, contact our experts today.