Sixty-six percent of customer service organizations now use agentic AI, according to Salesforce’s 2026 survey of 3,075 service professionals, and 91% of service leaders told Gartner they feel executive pressure to implement AI. Those two numbers are real, and they describe adoption and pressure. They do not describe results. Gartner’s own survey of 275 service leaders, run in July and August 2026, found that just 4% had achieved the ROI they expected from AI.
This page collects the customer service AI figures that can be traced to a primary document: Gartner press releases, Forrester’s 2026 predictions, the U.S. Bureau of Labor Statistics, the Brynjolfsson, Li and Raymond field study, Klarna’s annual filing, Salesforce’s earnings release and the vendor surveys themselves. Every row names its source, its sample and its period, and says whether it is a survey, a forecast, a company-reported figure or a vendor claim.
An earlier version of this page carried a benchmark table built on a Zendesk “enterprise median” and a set of vendor deflection claims. The Zendesk figures could not be found in the Zendesk report and have been removed; the checks table near the end lists each one. For the measurement side, see the AI customer support metrics and deflection framework and the ROI calculator.
- 01Adoption is high; proven return is not.Salesforce’s 2026 survey has 66% of service organizations using agentic AI. Gartner’s 2026 surveys have 24% of leaders showing positive financial returns across their AI use cases and 4% reaching the ROI they expected.
- 02Customers are not using company chatbots more.Gartner found customers about three times more likely to use third-party GenAI than a company-provided chatbot, and use of company chatbots statistically unchanged since 2022.
- 03Resolution claims are vendor-defined.Intercom states a 76% average resolution rate for Fin and Klarna reports its assistant handled 80% of chats in 2025. Each is the company’s own metric, so compare them with the definition, not as a benchmark.
- 04The workforce story is redesign, not mass layoffs.In Gartner’s survey, 85% of leaders are adding responsibilities to agent roles and 31% have implemented or plan AI-driven layoffs through 1Q27. The BLS still projects a 5% decline in customer service representative jobs from 2025 to 2035.
- 05The best controlled evidence is modest and uneven.A large field study in customer support found 14% more issues resolved per hour, with 34% for novice agents and minimal change for the most experienced.
00 — How to readHow to read the tables.
Each row carries a type. Official statistics and Official (company-reported) come from a government agency or a company’s own filing. Survey is a sample of respondents, and the sample and period are stated. Forecast is a prediction with its target year. Vendor claim is a company describing its own product. Academic study is a published paper. A vendor survey measures what its respondents say, and the vendor sells the product it asks about.
01 — AdoptionAdoption: widespread use, and pressure from the top.
Adoption and pressure are separate measurements. Gartner’s 91% is the share of leaders reporting executive pressure to implement AI, not the share that has done so. Salesforce’s 66% counts organizations that use agentic AI, from a survey run by a vendor that sells those agents.
The customer side looks different. Gartner’s February and March 2026 survey of 3,566 customers found that use of company-provided chatbots has been statistically flat since 2022, while use of third-party GenAI tools has nearly doubled in a year. Companies are deploying faster than customers are choosing to use what they deploy.
For the operating side of that deployment, see the 30-60-90 day launch plan and the guide to 24/7 support without added headcount.
| Figure | What it measures | Type | Source |
|---|---|---|---|
| 66% | Share of customer service organizations using agentic AI, Salesforce survey of 3,075 service professionals, Mar 9 to Apr 4, 2026 (the vendor sells AI agents for service) | Survey | Salesforce, May 20, 2026 |
| 39% | Share of customer service organizations using agentic AI in 2025, as reported in Salesforce’s 2026 release (“up from 39% in 2025”) | Survey | Salesforce, May 20, 2026 |
| 85% | Share of service organizations using at least one form of AI, Salesforce 2026 survey of 3,075 service professionals | Survey | Salesforce, May 20, 2026 |
| 91% | Share of 321 customer service and support leaders reporting pressure from executive leadership to implement AI, Gartner survey, October 2025 | Survey | Gartner, Feb 18, 2026 |
| 79% | Share of service leaders saying investment in AI agents is essential to meet business demands, Salesforce State of Service seventh edition, survey of 6,500 | Survey | Salesforce, Sep 10, 2025 |
| 70% | Share of organizations with AI service agents reporting measurable value within 60 days of deployment, Salesforce 2026 survey (self-reported, among adopters) | Survey | Salesforce, May 20, 2026 |
| 97% | Share of customer service leaders with AI who say it is changing their approach to workforce planning, Salesforce 2026 survey | Survey | Salesforce, May 20, 2026 |
| 74% | Share of consumers saying that, due to AI, they now expect customer service to be available 24/7, Zendesk CX Trends 2026 (6,182 consumers, 22 countries, June 2025) | Survey | Zendesk, Nov 18, 2025 |
| 3x | How much more likely customers are to use third-party GenAI tools than company-provided chatbots to resolve service issues, Gartner survey of 3,566 B2B and B2C customers, February to March 2026 | Survey | Gartner, Jul 8, 2026 |
| 58% | Share of customers who use GenAI and say they have used it to complete a task on their behalf (74% in B2B), Gartner survey of 3,566 customers, February to March 2026 | Survey | Gartner, Jul 8, 2026 |
| 38% / 2% | Growth in AI spending by customer service leaders against growth in overall service and support budgets, Gartner survey of 199 leaders, April to May 2026 | Survey | Gartner, Aug 26, 2026 |
| 12% | Median share of their 2025 budget that service and support leaders invested in AI, the highest of 10 business functions, Gartner survey of 1,303 senior leaders, January to April 2026 | Survey | Gartner, Jul 8, 2026 |
02 — Vendor ClaimsVendor claims against filings.
Resolution, deflection and containment are different events, and each vendor defines its own. Intercom says Fin’s confirmed-resolution rate averages 76% across customers; a company’s own newsroom or blog post is a claim, not a benchmark. No independent cross-vendor deflection median was found for this update, and the figure an earlier version of this page used could not be sourced.
The firmest company-reported numbers come from filings. Klarna’s 2025 Form 20-F states that its assistant handled 80% of customer service chats in 2025. The same filing shows customer service and operations expenses of $207 million, up 2% on 2024, while volumes rose 32%. Both facts are company-reported, and the second is the more useful: it reads as cost held roughly flat while volume grew, not as a cost collapse.
Salesforce’s Agentforce ARR is a revenue figure for a product, and its Agentic Work Units count tasks executed, not customer outcomes. They show how much the product is used, not how well it resolves. For the common ways deflection numbers mislead, read the anti-patterns guide.
| Figure | What it measures | Type | Source |
|---|---|---|---|
| 76% | Average resolution rate across Fin customers, stated by Intercom for its own product in March 2026 (confirmed resolutions as the vendor defines them; more than 12,000 teams) | Vendor claim | Intercom, Mar 12, 2026 |
| 80% | Share of Klarna’s customer service chats handled by its AI assistant in the year ended December 31, 2025, from Klarna’s own chat logs | Official (company-reported) | Klarna Group plc, Feb 26, 2026 |
| 700+ | Full-time agents whose work Klarna estimates its AI assistant does, based on the average monthly reduction in chat and telephone conversations in 2024 | Official (company-reported) | Klarna Group plc, Feb 26, 2026 |
| $39M | Cost savings Klarna reports its AI assistant delivered in 2024 | Official (company-reported) | Klarna Group plc, Feb 26, 2026 |
| $207M | Klarna customer service and operations expenses in 2025, up 2% on 2024 while volumes rose 32% and transactions 25% | Official (company-reported) | Klarna Group plc, Feb 26, 2026 |
| 2,831 | Klarna full-time employees at December 31, 2025 (3,422 a year earlier), a reduction Klarna attributes to leveraging AI | Official (company-reported) | Klarna Group plc, Feb 26, 2026 |
| $1.5B+ | Salesforce Agentforce annual recurring revenue at the end of fiscal Q2 2027 (quarter ended July 31, 2026); a product revenue figure, not a measure of customer outcomes | Official (company-reported) | Salesforce, Inc., Aug 26, 2026 |
| 7.0B | Agentic Work Units (discrete tasks executed by AI agents in production, as Salesforce defines them) delivered to date across Agentforce and Slack | Official (company-reported) | Salesforce, Inc., Aug 26, 2026 |
03 — SatisfactionSatisfaction and trust.
No independent, cross-company CSAT comparison of AI against human agents turned up in the primary sources read. Salesforce’s respondents rank customer satisfaction as the KPI that improved most after deploying AI agents, but the release publishes no score. Zendesk’s report offers one customer story, which is a vendor example.
What the surveys do measure is expectation and trust. Gartner found 54% of customers trust human agents more than AI for recommendations against 32% who trust AI more, and Zendesk found 95% of consumers want an explanation for AI decisions while only 37% of CX organizations offer one. Salesforce’s service professionals think customers trust AI more than the Metrigy consumer index it cites does.
| Figure | What it measures | Type | Source |
|---|---|---|---|
| #1 KPI | Customer satisfaction ranked first among KPIs improved after deploying AI agents, ahead of rep productivity, handle time, retention and first-response time, Salesforce 2026 survey (respondent-reported, no score published) | Survey | Salesforce, May 20, 2026 |
| 34% to 70% | SeatGeek’s AI agent satisfaction scores, as described in Zendesk’s own report (a customer story from the vendor) | Vendor claim | Zendesk, Nov 18, 2025 |
| 54% / 32% | Share of 5,801 US customers who trust human agents more than AI for product or service recommendations, against the share who trust AI more, Gartner survey, January to February 2025 | Survey | Gartner, Apr 28, 2026 |
| 95% | Share of consumers who want to know why AI makes some of the decisions it does, Zendesk CX Trends 2026 | Survey | Zendesk, Nov 18, 2025 |
| 37% | Share of CX organizations that currently offer agents or customers an explanation of AI decisions, Zendesk CX Trends 2026 | Survey | Zendesk, Nov 18, 2025 |
| 65% / 44% | Share of service professionals who say customers fully trust AI (Salesforce 2026) against the 44% of consumers who trust AI with service, a figure Salesforce cites from the Metrigy Consumer CX Index | Survey | Salesforce, May 20, 2026 |
| 85% | Share of CX leaders who say one unresolved issue is enough to lose a customer, Zendesk CX Trends 2026 | Survey | Zendesk, Nov 18, 2025 |
04 — ROIReturn and cost, as measured.
Gartner’s numbers are the most direct measure of return. In its July to August 2026 survey, 4% of service organizations had achieved the ROI they expected, and its separate January to April 2026 survey of 1,303 leaders found 24% showing positive financial returns across their AI use cases. Service leaders put a median 12% of their 2025 budget into AI, the highest of ten functions.
Gartner’s explanation is portfolio design: AI-forward organizations invest about 1.8 times more in data and knowledge foundations than in the use cases themselves, and Gartner predicts that by 2028 organizations with connected portfolios will get twice the value of those running standalone use cases. That is a forecast, and a research firm’s view of what separates better results.
There is no primary source for a universal “dollars returned per dollar spent” figure. If you need one, build it from your own volumes and costs; the ROI calculator guide shows how.
| Figure | What it measures | Type | Source |
|---|---|---|---|
| 4% | Share of customer service organizations that have achieved the ROI they expected from AI initiatives, Gartner survey of 275 service and support leaders, July to August 2026 | Survey | Gartner, Nov 4, 2026 |
| 24% | Share of service and support leaders demonstrating positive financial returns across their AI use cases, Gartner survey of 1,303 senior leaders, January to April 2026 | Survey | Gartner, Jul 8, 2026 |
| 2x | Business value Gartner predicts organizations with connected AI portfolios will achieve by 2028 against those pursuing standalone use cases (a forecast) | Forecast | Gartner, Nov 4, 2026 |
| 1.8x | How much more AI-forward service organizations invest in data and knowledge foundations than in the AI use cases themselves, Gartner, July to August 2026 survey | Survey | Gartner, Nov 4, 2026 |
| 30% | Reduction in operational costs Gartner predicts from agentic AI resolving common service issues by 2029 (a forecast) | Forecast | Gartner, Mar 5, 2025 |
05 — ProductivityHandle time and agent productivity.
The strongest evidence on support work is academic. Brynjolfsson, Li and Raymond studied 5,179 agents at one company after a generative AI assistant was introduced in stages: issues resolved per hour rose 14% on average, 34% for novice and low-skilled agents, with minimal effect for the most experienced. The OECD’s review of experimental studies puts average gains in customer support, software and consulting at 5% to over 25%.
Company figures run in the same direction but are not controlled: Klarna reports two minutes to resolve a query with its assistant against 12 for human agents as of September 2024, and Salesforce’s respondents say reps using AI spend 20% less time on routine cases.
| Figure | What it measures | Type | Source |
|---|---|---|---|
| 14% | Increase in issues resolved per hour after support agents at one company got a generative AI assistant, measured across 5,179 agents (34% for novice and low-skilled agents) | Academic study | National Bureau of Economic Research, revised Nov 2023 |
| 5% to 25%+ | Range of average productivity gains in experimental studies of generative AI among customer-support, software-development and consulting workers, OECD review | Academic study | OECD, Jul 8, 2025 |
| 2 vs 12 min | Average time to resolve a customer query at Klarna: AI assistant against human agents, as of September 2024, from Klarna’s chat logs | Official (company-reported) | Klarna Group plc, Feb 26, 2026 |
| 20% / 4 hours | Share of time reps using AI spend less on routine cases, and the weekly time Salesforce estimates this frees, 2025 survey of 6,500 | Survey | Salesforce, Nov 13, 2025 |
| 1 hour | Average reduction in daily agent workload Forrester expects from AI automating narrow tasks such as creating FAQs (a prediction for 2026) | Forecast | Forrester, Nov 10, 2025 |
06 — AutonomyForecasts: how much will AI resolve?
Forecasts disagree in tone. Gartner predicted in March 2025 that by 2029 agentic AI will autonomously resolve 80% of common customer service issues, with a 30% reduction in operational costs. Salesforce’s respondents expect AI to handle half of cases by 2027, up from 30% estimated in 2025. Forrester is more cautious: it predicts only one in four brands will see a 10% rise in successful simple self-service by the end of 2026, and that service quality will dip as companies deal with deployment complexity.
All four are predictions or expectations, and the wording matters. Gartner’s 80% applies to common issues. Salesforce’s 50% comes from service teams, not from measured case data. None of these is a result.
| Figure | What it measures | Type | Source |
|---|---|---|---|
| 80% | Share of common customer service issues Gartner predicts agentic AI will resolve autonomously by 2029 (a forecast) | Forecast | Gartner, Mar 5, 2025 |
| 30% to 50% | Share of service cases handled by AI today (as estimated by service teams in 2025) and the share they expect by 2027 (respondent expectation) | Forecast | Salesforce, Nov 13, 2025 |
| 1 in 4 | Share of brands Forrester predicts will see a 10% increase in successful simple self-service interactions by the end of 2026 (a prediction) | Forecast | Forrester, Nov 10, 2025 |
| Dip | Forrester’s prediction that service quality will dip in 2026 as companies wrestle with the complexity of AI deployment (a prediction) | Forecast | Forrester, Nov 10, 2025 |
07 — WorkforceThe workforce: redesign over elimination.
Gartner’s 2025 survey of 321 leaders found that 85% are adding responsibilities to agent roles, 75% are moving agents into new roles, and 63% are reducing headcount gradually through attrition. Thirty-one percent have implemented or plan AI-driven layoffs through 1Q27.
Government data points the same way over a longer horizon. The BLS counted 2,666,000 customer service representative jobs in 2025 and projects a 5% decline to 2035, about 141,800 fewer, while still expecting 289,500 openings a year to replace people who leave. Klarna’s filing shows the same pattern in one company: headcount fell from 3,422 to 2,831 in 2025.
| Figure | What it measures | Type | Source |
|---|---|---|---|
| 85% | Share of 321 service leaders adding new tasks and responsibilities to frontline agent roles, Gartner survey, September to October 2025 | Survey | Gartner, Apr 28, 2026 |
| 31% | Share of service leaders who have implemented or plan frontline workforce reductions through layoffs in response to AI through 1Q27, Gartner | Survey | Gartner, Apr 28, 2026 |
| 63% | Share of service leaders reducing frontline headcount gradually through attrition, Gartner survey, September to October 2025 | Survey | Gartner, Apr 28, 2026 |
| 75% | Share of service leaders shifting agents into entirely new roles within the service organization, Gartner survey, September to October 2025 | Survey | Gartner, Apr 28, 2026 |
| 58% | Share of service leaders aiming to upskill agents into knowledge management specialists, Gartner survey of 321 leaders, October 2025 | Survey | Gartner, Feb 18, 2026 |
| 2,666,000 | US customer service representative jobs in 2025, Bureau of Labor Statistics | Official statistics | U.S. Bureau of Labor Statistics, read Nov 5, 2026 |
| -5% | Projected change in employment of customer service representatives, 2025 to 2035 (a projection, down about 141,800 jobs) | Official statistics | U.S. Bureau of Labor Statistics, read Nov 5, 2026 |
| 289,500 | Average annual openings projected for customer service representatives over the decade, all from workers leaving the occupation | Official statistics | U.S. Bureau of Labor Statistics, read Nov 5, 2026 |
| $44,770 | Median annual pay of US customer service representatives, May 2025 | Official statistics | U.S. Bureau of Labor Statistics, read Nov 5, 2026 |
08 — ChecksFigures you will still meet elsewhere.
These figures appeared on this page before the rebuild or circulate widely. Where the source does not contain the number, the table says so.
| Often quoted as | What the source says |
|---|---|
| Zendesk’s enterprise median tier-1 deflection is 41.2%, with a top quartile of 58.7% | Neither figure appears in Zendesk’s CX Trends 2026 report or launch release. Every place that repeats them is a secondary or vendor page. Removed. |
| AI-handled tickets score 4.10/5 CSAT against 4.30/5 for humans (Zendesk) | Not in the CX Trends 2026 report. The report’s satisfaction example is a single customer story (SeatGeek, 34% to 70%), labelled here as a vendor claim. |
| AI customer service returns $3.50 for every $1 invested (Intercom / Fin) | No Intercom publication read for this update contains it. Gartner’s survey of 275 service leaders (July to August 2026) found 4% had reached their expected AI ROI. |
| Gartner finds AI deflects 45% of queries but only 14% are resolved | Not in any of the Gartner customer service releases read for this update. The releases that exist describe ROI, spending, workforce and customer behaviour, not those two numbers. |
| Fin’s average resolution rate is 51% or 67% | Those are older figures. Intercom’s March 12, 2026 post states 76% across customers and defines the metric itself, so treat it as a vendor claim. |
| Only 44% of consumers trust AI to handle customer service (Salesforce) | Salesforce’s 2026 release cites the Metrigy Consumer CX Index for the 44%. Salesforce’s own figure is that 65% of service professionals say customers fully trust AI. |
| 66% of service organizations run AI agents, “including pilots” | The Salesforce wording is that 66% “now use agentic AI”, from a 2026 survey of 3,075 service professionals. The release does not say pilots are included, and the vendor sells the product. |
| Klarna’s AI replaced 700 agents and will add $40M in profit | Klarna’s 2025 Form 20-F says the assistant does the equivalent work of over 700 full-time agents and delivered $39 million in cost savings in 2024. The $40M was a February 2024 projection. |
| Decagon 80%, Ada 70% to 80%, Sierra 70% deflection or resolution | These came from vendor case-study pages that were not read at the vendor’s own page in this update, and the metrics are defined differently by each vendor. Removed rather than repeated. |
09 — Next StepsWhat to do with this data.
Define the metric before you compare. Ask any vendor whether its resolution figure counts confirmed resolutions, containment or deflection, and what happens to customers who return within a few days. Then set a baseline from your own tickets, which is the only benchmark that matches your mix of intents.
Plan for the workforce changes Gartner describes rather than the layoffs vendors imply: knowledge management, quality review and escalation design become more important as automation takes simple work. The agentic AI customer support playbook covers team structure, and the AI agent productivity statistics page covers the wider evidence on productivity.
If you are evaluating platforms or planning a rollout, our CRM and automation services team can help you set baselines and test vendor claims against your own data.
10 — The SourcesWhere each figure comes from.
Government statistics
- U.S. Bureau of Labor Statistics, Occupational Outlook Handbook: Customer Service Representatives. read Nov 5, 2026.
Academic and public research
- National Bureau of Economic Research, Generative AI at Work (Working Paper 31161; Brynjolfsson, Li and Raymond). revised Nov 2023.
- OECD, Unlocking productivity with generative AI: Evidence from experimental studies. Jul 8, 2025.
Company filings and official disclosures
- Klarna Group plc, Form 20-F for the fiscal year ended December 31, 2025. Feb 26, 2026.
- Salesforce, Inc., Second quarter fiscal 2027 results (Form 8-K, Exhibit 99.1). Aug 26, 2026.
Research firms and analysts
- Gartner, Survey Finds 91% of Customer Service Leaders Under Pressure to Implement AI in 2026 (press release). Feb 18, 2026.
- Gartner, Survey Finds 85% of Service and Support Leaders Are Expanding Human Agent Responsibilities (press release). Apr 28, 2026.
- Gartner, Survey Finds Customers Are 3x More Likely to Use Third-Party GenAI Than Company-Provided Chatbots (press release). Jul 8, 2026.
- Gartner, Survey Finds AI Spending by Customer Service Leaders Has Surged by 38% (press release). Aug 26, 2026.
- Gartner, Survey Reveals Just 4% of Customer Service Leaders Have Achieved the Expected ROI of Their AI Initiatives (press release). Nov 4, 2026.
- Gartner, Predicts Agentic AI Will Autonomously Resolve 80% of Common Customer Service Issues by 2029 (press release). Mar 5, 2025.
- Forrester, Predictions 2026: AI Gets Real For Customer Service, But It’s Not Glamorous Work (blog post). Nov 10, 2025.
Vendor surveys and vendor reports
- Salesforce, State of Service: AI Agents Edition (survey of 3,075 service professionals, Mar 9 to Apr 4, 2026) and its launch article. May 20, 2026.
- Salesforce, State of Service, seventh edition (survey of 6,500 service professionals, Apr 25 to Jun 6, 2025) and two launch articles. Sep 10 and Nov 13, 2025.
- Zendesk, CX Trends 2026 report (6,182 consumers and 5,115 business respondents in 22 countries, June 2025) and launch press release. Nov 18, 2025.
- Intercom, From resolutions to outcomes: Evolving how Fin delivers value (company blog). Mar 12, 2026.
Adoption is high; proven return is still thin.
The sourced picture is consistent. Most service organizations are using AI and most leaders feel pressure to do more, while Gartner’s 2026 surveys find few can show the return they expected. Controlled evidence shows real but uneven productivity gains, and company filings show cost held roughly flat while volume grows.
Use the figures here as reference points with their labels attached, and measure your own resolution, re-contact and cost per resolved ticket before committing to a vendor’s number.