Trying an AI agent is a useful first step. Funding it, connecting it to real work and trusting it with customer requests are separate decisions. For a business considering Agentforce, the important question is what would make a pilot worth expanding.
What the 85% figure actually means
MT Newswires reported on 29 September 2026 that an Oppenheimer Dreamforce customer survey found 85% of respondents had tested or deployed Agentforce. The same report described limited evidence of a sharp improvement in demand, with nearly half of respondents still testing or running pilots. Read the survey coverage.
This is a result from a particular survey, not an adoption rate for Salesforce’s entire customer base. The public summary does not give us enough detail about the sample to generalise it. It also combines testing and deployment, which are very different stages.
A trial shows interest. A production decision requires a useful workflow, a reliable result and a cost the business can justify.
Why a successful demo may not become a project
Our view is that a business should assess five practical questions before expanding an AI pilot. These are delivery considerations, not findings proven by the survey.
| Question | Evidence to collect |
|---|---|
| Does it solve a frequent problem? | Request volumes, handling time and the current failure rate. |
| Can it access reliable information? | Approved knowledge, record ownership and examples of missing data. |
| Can it act within clear limits? | Permissions, approval rules and situations requiring human review. |
| Does it make financial sense? | Actual usage charges plus setup, integration, review and support costs. |
| Who will maintain it? | A named owner for content, monitoring and changes after launch. |
A demo can use a clean example and a predictable question. Live work includes incomplete records, unusual requests and customers who change their minds. Those cases need to be part of the decision.
Choose one workflow you can measure
For example, a distributor could trial an assistant that answers delivery-status questions using approved order data. Start with read-only answers and a clear route to the service team. Treat refunds, address changes and other consequential actions as separate scope.
This is an illustrative example, not an Adaptal customer result. A useful baseline would record how many requests arrive, how long staff spend on them and how often customers need a second response. The pilot should measure the same things, including time spent checking or correcting the agent.
Do not count every automated answer as a successful resolution. Agree what counts as correct, what evidence supports it and how a customer can reach a person.
Set a clear decision at the end of the pilot
- Define the job. Write down the requests the agent should handle and the requests it must hand over.
- Prepare representative tests. Include ordinary requests, ambiguous wording, missing data and attempts to obtain unauthorised information.
- Agree acceptance criteria. Set business-specific targets for answer quality, escalation, cost and customer experience before testing.
- Review the results together. Include the business owner, Salesforce team and people handling escalations.
- Decide what happens next. Expand, revise or stop based on the evidence. A pilot can be useful even when it shows that the workflow is not ready.
For a broader delivery checklist, see our Agentforce pilot-to-production guide. If system access is the obstacle, review our integration services.
What this means for your Salesforce budget
A high trial rate is a reason to investigate useful applications, not a reason to buy a broad rollout. Ask for a scoped first phase with clear dependencies, responsibilities and a review point. Keep enough budget for testing and ongoing operation, rather than treating the licence as the full cost.
Adaptal can help assess a workflow, identify data and integration needs, and plan a practical first phase. Our AI services explain the work we can support.
Discuss your Salesforce project
Tell us what you need help with. We’ll get back to you.