Customers who spoke at Dreamforce 2026 described service teams that used to be judged on cost per contact and are now judged partly on revenue, after AI agents took over the routine questions.
What the customers described
In a post published October 1, Salesforce collected lessons from customers who presented at Dreamforce 2026, among them AT&T, Southwest Airlines, Canada Goose, Canon, Fisher & Paykel, Sammons Financial Group and Takeda. The post is Salesforce’s own, so every result in it is a result the customer chose to share and Salesforce chose to print.
Canada Goose gives the clearest numbers. Salesforce’s post says Agentforce now autonomously resolves 89% of the company’s routine messaging inquiries and 15% of its calls. Dennis Liut, the company’s Head of Global Customer Experience and Revenue, described the result this way: “And now we have an Experience Center that used to be a call center that is bringing in millions in revenue.” According to the post, the staff who once answered delivery-date and exchange questions now work as “Style Experts” focused on personalization and personal shopping.
Advertisement
Canon is making a similar move. The same post says Canon is pivoting its service organization from a cost center to a value center, with agents fielding the questions that do not need a person. Bill Duval, the company’s VP of Service Information Management and Technology, says the other payoff is the customer data those conversations produce, which can feed product development.
Where the revenue comes from
This part is our read, not the customers’. The revenue claim in the post is thin. Only Canada Goose cites a revenue outcome, the figure is “millions,” and the post gives no baseline, no time period and no share attributable to agents. The mechanism is easier to see than the number. Once agents absorb order-status and exchange questions, the conversations left for people skew toward customers who are deciding what to buy. A sales conversation was sitting in a service queue all along, and a staff title like “Style Expert” is what the change looks like on an org chart.
Canada Goose’s figures also describe different channels. Agents resolve 89% of routine messaging inquiries but 15% of calls, and the post does not say how large routine inquiries are as a share of total volume. That leaves the number a sales leader would want, revenue per remaining human conversation, unstated.
We took a similar look at a company-supplied figure in Salesforce puts a real number on Agentforce ROI. The same question applies here: a result is only as useful as the baseline it is measured against, and the post does not give one.
The data and control work comes first
The customers in the post spent as much time on controls as on results. AT&T built what it calls an “agent briefing”: check-in details captured in its retail app are packaged so a store expert knows who the customer is and why they came before the conversation starts. The post says that depends on data quality and governance work done up front. AT&T also kept the ability to switch agents on or off at any time, runs a trust layer, and uses its own LLM for anything touching proprietary IP.
Sammons Financial Group ran more than 200 guardrails and tests before going live and chose to stop the model from learning from live conversations, which removes drift in a compliance-sensitive setting. Andrew Walling, the company’s AVP of Capability Planning and Delivery, said of the safety design: “We have a kill switch, a supervisor agent that’s constantly listening to all of those phone calls, checking the values in the CRM against what was said on the line.”
Southwest Airlines takes a different route to the same discipline. Megan Rauber, its Manager of Customer 360, says the airline uses chat data to decide where to deploy agents next and has seen CSAT improve week over week as capabilities are added. Fisher & Paykel’s Chief Digital Officer, Rudi Khoury, advised teams to start early and learn from end users instead of testing internal hypotheses.
What it means for the sales leader
Three things follow for anyone who owns pipeline.
First, the people left in the service queue after agents take the routine work are the closest thing most companies have to a warm-lead team. Whether those conversations become pipeline depends on routing rules between service and sales that someone has to write. Ask the head of customer service which conversation types now reach a human and what happens when a customer shows buying intent.
Second, ask for a baseline. A revenue figure from a service team means little without the quarter before agents went live, the definition of “revenue” and the share that came from agent-handled versus human-handled contacts. We made a similar argument about trust claims in Dreamforce sold trust, the numbers are still missing.
Third, copy the control work. Sales organizations that let agents log activity, draft outreach or update opportunities face the same risk Sammons designed around, an agent saying one thing while the record says another. Sammons and AT&T both treated an off switch and a guardrail test suite as launch requirements. Takeda, per the post, is automating the administrative work that consumes its sales reps’ time while staying inside pharmaceutical regulation, which is the same discipline applied on the selling side.
This quarter, have service log a weekly count of conversations that ended in a purchase, an upgrade or a handoff to sales, open an opportunity record for each, and compare the total with the quarter before agents went live.
Source: Salesforce, Getting to ROI: How Brands Turn Cost Centers into Revenue Engines with AI
