Public results from companies that put AI into production, and the pattern we build on: agents that act, humans who approve.
Real numbers from companies running AI agents today, with people still owning the hard calls. These companies are not GOFTUS clients.
Klarna
2/3
Klarna reported its AI assistant handled 2.3 million conversations in its first month and cut average resolution time from 11 minutes to under 2.
By 2025, Klarna said it was putting more emphasis on human agents for complex cases.
Octopus Energy
80%
Octopus Energy said AI-drafted customer emails scored 80% customer satisfaction versus 65% for human-written emails, and within around three months AI handled over a third of customer emails with staff reviewing drafts.
Staff reviewed AI drafts before they were sent.
Lyft
87%
Lyft reported that an AI customer-care assistant cut average resolution time by 87%, while complex cases route to human specialists.
Complex cases still go to human specialists.
The Permanente Medical Group (Kaiser Permanente)
15,791 hrs
7,260 physicians used AI scribes across about 2.5 million patient visits between October 2023 and December 2024. The scribe drafts the clinical note from the visit conversation.
Physicians review and edit every draft before it enters the record.
Stanford Health Care
162
In a five-week study, clinicians used about 20% of the drafts and reported lower task load and burnout scores. Time spent on messages did not change.
Every draft was edited by a clinician before it reached a patient.
Klarna
2/3
The assistant held 2.3 million conversations in month one and cut average resolution time from 11 minutes to under 2.
By 2025 Klarna was putting more emphasis on human agents for complex cases.
Bank of America
3B+
Erica now averages more than 58 million interactions a month across nearly 50 million users, from balance trends to investment questions.
When a person is needed, Erica books an appointment and hands the client to a banker.
NOBULL
50%
AI agents took over low-complexity tickets. New tickets fell 49% year over year and AI conversations scored 90 to 91% customer satisfaction.
AI performance is reviewed with the same QA process used for the human team.
Zalando
90%
Campaign production time fell from six weeks to a few days, and its Size & Fit AI cut size-related returns by more than 8%.
Built on years of Zalando's own product and customer data rather than generic content.
UPS
50%
UPS built Message Response Automation (MeRA) to draft responses to some of the more than 50,000 customer emails it receives every day.
Built to support contact centre agents, with plans to extend it to sales, HR and finance.
DHL Supply Chain
100k+
Agents run appointment scheduling, driver follow-up calls, transport status calls and high-priority warehouse coordination.
Routine communication is automated so teams can focus on exceptions and problem-solving.
Equity Residential
90%
More than 1.5 million customer interactions a year run through EliseAI, which Equity Residential says contributed to $14 million in payroll savings.
Figures quoted by Equity Residential in EliseAI's published customer testimonial.
Zillow (Follow Up Boss)
6.2M+
Follow Up Boss added a feature that drafts talking points and email copy from agents' call transcripts. Agents asked for it early, and it shipped in four days.
Agents review the drafts and send them from their own accounts.
Georgia Tech
66%
Deployed to more than 600 students in an online AI course. Answer accuracy ranged from 75% to 97% depending on the source material.
Jill only answers when verified course material supports the answer.
Georgia State University
21%
Admitted students with the assistant were 3.3 percentage points more likely to enrol. Fewer than 1% of 50,000 student messages needed staff attention.
Issues that needed a person were escalated to admissions counsellors.
Book a free call and we will tell you honestly whether AI automation helps, and where to start.
We reply within one business day.