From asking for help to getting it done

People don’t ask for help because they want another answer. They want the problem solved. Together, DevRev and OpenAI are working to make that possible for employees and customers – without making them navigate the company behind the request.

Think about the last time you needed access to a tool at work. You may have found the policy, opened a ticket, waited for an approval, and checked back to see what happened. The answer was only the beginning. The work was everything after it.

That’s the opportunity we see in our work with OpenAI. OpenAI brings frontier AI models. DevRev brings Computer, which connects the context, permissions, and workflows needed to move from a request to an authorized next step. We call the AI that helps carry out that work an AI Operator.

An AI Operator should know when it can help directly, when it needs approval, and when a person should take over. It should make the handoff easier, not make someone repeat their story.

Employees want momentum, not another wait

If an employee needs software access, the useful response isn’t a stack of policy links. It’s help finding the right request, starting the approved process, and seeing where it stands. If approval or privileged access is required, the right person gets the request and the context to act. The existing IT system remains the system of record.

What better service looks like

A request keeps moving with less back-and-forth, while people remain in control of decisions that need their judgment.

Customers want to solve the problem, not wait on hold

Someone asking about an order, an entitlement, or a product problem wants more than an instant answer when the next step still needs to happen. With the right account and service context, an AI Operator can answer a question, take an approved action when the workflow allows it, or bring a support specialist in without making the customer start over. A simple question might need an answer; a more involved request needs a path to resolution.

The test isn’t whether a chat ended. It’s whether the customer’s issue was resolved, whether they had to come back, and how the experience felt.

Getting it right matters as much as getting it done

Useful action starts with trustworthy context. An AI model on its own doesn’t know an employee’s permissions, a customer’s entitlement, or which action a company allows. Computer connects that information to the request and keeps actions within the boundaries set by the business.

In our Enterprise-Bench evaluation, Computer reached 94.3% accuracy versus 63.6% on the tested L1–L2 tasks, while using 4.4× fewer tokens per correct answer. Those results measure answer quality and efficiency on that benchmark – not an automatic resolution rate for every customer workflow. We validate the outcome of each deployed workflow separately.

The BILL customer story shows what this can look like in practice: BILL reported a 70% AI resolution rate in a proof of concept using real customer queries. It’s one customer’s proof of concept, not a promise of the same result everywhere.

Measure the work, not the conversation

We believe the value of AI should be tied to outcomes people can verify. For each workflow, that means agreeing on what counts as a completed request, establishing a baseline, and measuring quality, repeat contacts, human effort, and cost alongside the result.

That approach can also inform how AI services are priced over time: around verified resolutions rather than chat volume alone. The details – including scope, exclusions, and the role of human approvals – need to be worked out with each customer. We’re developing this approach, not announcing a universal resolution-based price today.

Building with the people who make it work

Enterprise service doesn’t change because someone installs a model. It changes when a team chooses the right workflow, connects the systems involved, tests real requests, and improves what happens when an exception comes up. That’s work we’re excited to do with customers and partners.

Partners know where a customer’s requests get stuck: the handoff nobody owns, the approval that takes too long, the policy that is easy to find but hard to act on. They help choose a workflow worth solving, connect it to the systems the customer already uses, and define what a successful resolution looks like before the first request goes through. Then they stay close to the results – checking quality, handling exceptions, and improving the service as needs change. That is how a promising AI pilot becomes something people can rely on.

Distribution matters too, but being discoverable is not the same as being useful. Customers can already explore solutions in the DevRev Marketplace, and OpenAI’s Partner Network brings partners together to build and deliver AI solutions. We see a future OpenAI Marketplace as another potential path for customers to discover DevRev’s employee and customer self-service solutions. Any availability, purchasing options, program benefits, or treatment of existing OpenAI commitments would depend on applicable terms. What matters most is the working service partners and customers build together – and the outcome it delivers.

The self-service enterprise makes progress feel natural

A person asks for help. The right context is there. The next step happens safely. And if a human needs to step in, they can pick up where the AI left off. That’s how we move from answering questions to getting work done.