A survey headline can help an outsourcing buyer ask better questions. It cannot tell that buyer how many people should staff tomorrow's queue. The missing link is the actual work: the requests arriving, the service promise and the capacity needed when an automated response falls short.
In a November 5 release, Gartner reports a survey of 275 service and support leaders conducted in July and August 2026. Twenty-one percent said their organisations had reduced human agent headcount because of AI. This is a share of responding leaders reporting an organisational change, not the percentage of all customer-service jobs eliminated.
Keep the denominator attached
The survey is evidence about what those respondents reported. The public release does not provide the underlying respondent data or enough detail to establish how closely this sample resembles an individual client's operation. It does not measure Myaa's workforce, a particular outsourcing provider or every service team worldwide.
That distinction changes the conversation with a provider. A client cannot convert a proportion of organisations into a proposed reduction in contracted hours. It first needs evidence about its own workload. Otherwise, a statistic about other businesses becomes a staffing assumption without an operational bridge.
For example, imagine a retailer whose outsourced team handles routine delivery questions alongside damaged-order complaints. Even if automation helps with the first category, the client and provider still need to understand the second. A smaller queue might contain more difficult work. This is a hypothetical planning example, not a finding from Gartner's survey.
Measure the work that reaches people
Our recommendation is to agree on a workload record before renegotiating capacity. Describe the types of requests, how they are counted and which outcomes would qualify as a completed interaction. Keep those definitions consistent across the client and provider.
A useful review could follow a request beyond its first automated answer. Did the customer return about the same issue? Did someone reopen the case? Was a refund, correction or human explanation still required? These are suggested questions for a local test, not claims that any particular product measures them correctly.
Look at the work left with the human team as well. If an assistant handles straightforward requests, staffing discussions should examine the remaining case mix rather than assuming every hour has become equally easier. Ask the provider to explain what its team is now responsible for and what evidence supports the proposed service commitment.
Make the client–provider agreement specific
The client should define the result it is purchasing. The provider should explain the resources and operating assumptions behind delivering it. A price discussion becomes more useful when both sides can inspect the same accepted workload record instead of debating a general AI adoption forecast.
Consider agreeing a limited review period before a lasting capacity change. Specify the comparison, the exceptions that need human coverage and who can restore capacity if the service commitment is missed. That proposal is an operational option, not a prediction about savings.
It is also worth deciding where any freed time should go. A team may choose faster replies, more detailed follow-up or additional coverage. Those choices have different implications for the service being bought. A vague instruction to become more efficient leaves the client and provider free to imagine different benefits.
Use the survey as context
A survey can challenge an overly simple story about automation. Its most useful role in a supplier meeting may be to prompt a request for better local evidence, rather than to supply the answer.
Before changing an outsourcing arrangement, ask for the documented relationship between observed work and the proposed capacity. Keep measured outcomes separate from assumptions about future automation. Then choose a service plan that can be explained in terms of the customer's actual needs.
Evidence limits: the central survey claims come from Gartner's public release, without independent access to its raw data. The planning examples and client–provider review suggestions are original analysis, not reported survey results.
AI-generated conceptual editorial image; fictional service planning scene, not a Gartner survey participant or Myaa client.