The customer satisfaction index just posted its sharpest drop since the pandemic. Complaint rates hit an all-time high, roughly double where they were a decade ago. Gartner looked at 432 AI customer service use cases and found a quarter of them produce positive ROI and a quarter produce negative ROI, with another 42% that nobody can quantify at all.
And more than three-quarters of service leaders plan to increase AI investment this year anyway.
That is not stupidity. It is what happens when a function is under cost pressure, has a tool that demonstrably reduces cost per contact, and has no agreed way to measure what it does to the relationship. You can prove the savings in one quarter. The damage shows up in three.
September is when that bill starts arriving, because Q4 is when your customers actually need you. Here is what happened in August and what to do about it before the holiday season makes the decision for you.
The EU Now Requires You to Tell People They Are Talking to a Bot
Article 50 of the EU AI Act became enforceable on August 2, 2026. If you serve EU customers, disclosure is no longer a roadmap item.
The article covers four situations: AI systems that interact directly with people (chatbots, agents, voice assistants), machine-readable marking of synthetic content, notice when you use emotion recognition or biometric categorisation, and visible labels on deepfakes. August 2 also stood up the EU AI Office as the enforcement authority. Penalties run to €15 million or 3% of global annual turnover. The Commission’s announcement is here, and CX Dive has the practitioner read.
Two pieces of advice from that coverage are worth stealing outright. Forrester principal analyst Enza Iannopollo: “Forget about the regulators… Consumers want to know when they are exposed to AI,” citing over 70% of consumers in nearly every market Forrester surveys. And TTEC Digital’s Mahadevan Meikum Perumal: “The wrong thing to do is to limit it to a legal or an IT problem.” His recommended first step is an inventory of every AI system touching a customer, which almost nobody has.
Do the inventory even if you have zero EU exposure. Not for compliance. Because most organizations cannot currently produce a list of every place an AI is talking to their customers, and you cannot govern an experience you cannot enumerate. That inventory is the cheapest customer experience audit available and it takes an afternoon.
One flag for your legal team: multiple law firms report that the separate high-risk obligations under Annex III were deferred to December 2027 via the Digital Omnibus. We could not confirm that against a primary source. Verify before you plan around it.
Complaints Just Hit a Record, and Nobody Has Left Yet
The American Customer Satisfaction Index released its Q2 2026 national report on August 11. The national score fell to 76.1 from 76.7 in Q1, the sharpest drop since the pandemic. Customer complaint rates hit an all-time high. ACSI’s release is here; CX Dive’s coverage is here.
Forrest Morgeson, associate professor of marketing at Michigan State and ACSI’s director of research emeritus, notes complaints have “nearly doubl[ed] over the past decade,” and explains why that number is worse than it looks: “Customers are taking the time to get in touch with the company and complain, and it’s really an indicator that there’s a good bit of customer displeasure and dissatisfaction out there.”
Then the sentence every executive team should have on a slide: “once consumers do have choices, with this kind of dissatisfaction, they’re going to take them.”
That is a description of latent churn. Dissatisfaction has not converted to defection yet because consumers feel choice-constrained, not because they are happy. Morgeson attributes the decline to record corporate profits alongside price increases and weaker service, which is a polite way of saying companies have been testing how much worse they can get before customers move, and the answer so far has been “quite a bit.”
The operational implication is specific. Your retention metrics are currently lying to you by omission. Churn is a lagging indicator that only fires when a customer has both the desire and the option to leave. Complaint volume is the leading indicator, and it is at a record. If your dashboard shows stable retention and rising contact volume, you do not have a stable business. You have a queue of people who have not found a replacement yet.
This is also the argument for measuring sentiment continuously rather than annually. An NPS or satisfaction program that runs once a year cannot see a build-up like this. It can only confirm it afterward.
Gartner Ran the Numbers on AI Service, and the Numbers Are Bad
Two Gartner findings in August, and together they close a loop.
The first, from July and covered on August 17: across 432 analyzed AI customer service use cases, 25% produced positive ROI, 25% produced negative ROI, 42% had ROI that leaders could not quantify, and 11% broke even. Despite that, more than three-quarters of leaders plan to increase AI investment in 2026. Customer service leads all business functions in AI adoption, averaging nearly five use cases and around 13% of functional budget. Fifty-six percent of service and support leaders expect their incentives tied directly to AI outcomes this year. Headcount is a wash: roughly a quarter of organizations are growing service headcount and roughly a quarter are cutting it. CX Dive’s coverage is here.
Julie Geller of Info-Tech Research Group put her finger on the mechanism: “Containment is also too often mistaken for success. Delaying contact with a human agent is not the same as resolving the customer’s problem.”
The second finding, published August 4, tells you what customers think of the containment strategy. A Gartner survey of 3,566 B2B and B2C customers fielded February to March 2026 found 87% say access to a human agent is essential when a company uses generative AI in service. Also worth noting, because it complicates the easy narrative: 50% say interactions are actually easier when companies use GenAI, and 58% of GenAI users have had it complete a task on their behalf, rising to 74% in B2B. The release is here.
Eric Keller, Senior Director Analyst at Gartner: “Service leaders should not use GenAI as a mandatory first step for every issue.” And, counterintuitively: “if you make it easier to switch to a person, it helps with longer-term adoption.”
Put the two studies side by side and the failure mode is obvious. Containment is the metric because containment is easy to measure and maps directly to cost. So teams optimize for containment, which means making the human harder to reach, which is precisely the thing 87% of customers say is unacceptable, which produces the complaint volume the ACSI just recorded. The 42% of use cases with unquantifiable ROI are unquantifiable because the cost side is measured in the contact center and the benefit side, or the damage, shows up in retention twelve months later on someone else’s P&L.
Keller’s point is the way out, and it runs against instinct: an easy escape hatch increases AI adoption rather than undermining it. People will try the bot if they are not afraid of getting trapped in it. Design the exit first, then the deflection. That reframing is most of the difference between AI that improves the experience and AI that just moves the cost.
The FTC Put Personalized Pricing on Notice
On August 19, the FTC proposed an enforcement policy statement signaling it will treat undisclosed data-driven individualized pricing as a potential Section 5 violation, and opened it for public comment. CX Dive’s coverage is here.
FTC Chairman Andrew Ferguson: “When consumers see a listed price, they expect it to be [the] same price that everyone else sees, not the retailer’s estimate of how much they are willing to pay based on their personal data.”
The FTC cannot ban the practice outright, but non-disclosure may violate Section 5. The state layer is moving faster: Maryland has limited it, Connecticut has banned it, and Holland & Knight counts more than 40 bills across more than two dozen states. The supporting research is uncomfortable reading if you run pricing. Consumer Reports and the Groundwork Collaborative found up to 23% price variance on identical Instacart grocery items. A separate Consumer Reports investigation of Uber and Lyft found a median spread around 42%. A 2024 Consumer Reports study found two-thirds of US consumers opposed the practice.
Jeannie Walters of Experience Investigators: “Trust is hard to earn and quick to lose.”
Here is the part that catches teams who think this does not apply to them. Very few companies believe they do personalized pricing. Quite a lot of companies do something functionally adjacent: dynamic discount eligibility, retention offers triggered by cancellation intent, segment-specific promo codes, or a win-back price that regular customers never see. If your loyal customer pays more than the one who threatened to leave, you have built a pricing system that punishes loyalty, and you did it without ever calling it surveillance pricing.
The regulatory exposure varies. The brand exposure does not, and it lands the day a customer posts a screenshot comparison.
Panera Threw Out Surprise-and-Delight for Points
The week of August 18, Panera relaunched MyPanera nationwide, converting up to 70 million auto-enrolled members from a surprise-and-delight model to a conventional points program. Ten points per dollar, redemptions from 250 points for bagels and mini-bakery items up to 2,000 points for whole entrees. Tiers now exist: base, then MyPanera+ at $300 in annual spend (12 points per dollar on delivery, a birthday You-Pick-Two), then MyPanera+ SipClub, a drinks subscription with “Sip Club Saturdays.” Keke Palmer fronts the promotion, with 500 bonus points on a first purchase through September 1. It is part of the “RISE” turnaround under CEO Paul Carbone. CX Dive has the details.
Chief Digital Officer Joshua Fine’s rationale: “Customers want choice. They want to be able to choose what they redeem. They want flexibility.”
The old MyPanera was one of the few actually distinctive loyalty programs in food service. It gave you unpredictable rewards based on your actual behavior, which felt like being noticed. It was also opaque, unforecastable for the customer, and impossible to plan a visit around. Panera traded distinctiveness for legibility.
That trade is defensible and it is worth being clear-eyed about what it costs. A points program is a discount schedule with a delay. Every competitor has one, none of them create preference, and the customer can now compute exactly what your loyalty is worth per dollar. What Panera bought is a tier structure, and the tiers are the real product here: $300 in annual spend to reach MyPanera+, plus a drinks subscription that converts an occasional visit into a recurring habit. Subscription beats points, and Panera clearly knows it.
The broader lesson for anyone running loyalty: legibility is not the same as loyalty. If your program can be fully described as “spend money, get money back,” you have built a rebate, and rebates compete on rate. The programs that actually raise customer lifetime value either change behavior (subscription, habit) or change status (recognition, access). Panera got one of the two.
Retail Is Reorganizing Around the Customer Before Q4
Three moves in the last two weeks of August, all pointed at the same season.
Salesforce published its 2026 holiday predictions on July 20, and one number should reset your Q4 planning: it projects 20% of holiday ecommerce traffic will originate from AI chat agents, with reliance on AI assistants as a first stop up 200% between May 2025 and May 2026, and 50% of shoppers using an AI assistant at some point. It also predicts one in three ecommerce sites will have a branded, site-specific shopper agent live by Cyber Week, and reports that retailers with branded agents saw 59% higher holiday sales growth in 2025 (6.2% versus 3.9%). Other projections: 38% of holiday retail dollars will be “hybrid sales,” 79% of in-store shoppers will be on their phones in the aisles, retailers will spend an extra $3 billion globally subsidizing free shipping, and 35% of shoppers will trade down. The full set is here. These are Salesforce projections from Salesforce data, so weight them accordingly, but the directional claim is corroborated by everything else in this roundup.
Worth noting for anyone building a Q4 plan on benchmarks: as of late August, Deloitte, Adobe Analytics and NRF have not published 2026 holiday forecasts. If someone hands you a holiday number this week, ask where it came from.
Best Buy disclosed on its Q2 FY2027 call, covered August 27, that it is rolling out Ask Blue, a conversational AI shopping and support assistant using product details, reviews, availability and pricing, which can route to self-service or connect to a live support specialist. Incoming CEO Jason Bonfig, currently chief customer, product and fulfillment officer, takes over in the fall. Q2 comps were up 4.1% on revenue of $9.8 billion, up 3.6%. Coverage here. Note the design choice: the assistant connects to a human specialist as a feature, not as a failure state. Gartner’s Eric Keller would approve.
Kohl’s named Arianne Parisi as its first chief customer officer on August 25, promoting her from chief digital officer to a newly created role reporting to CEO Michael Bender, with omnichannel CX, loyalty, personalization and marketing under one leader. CMO Christie Raymond departs in September. Bender: “Bringing marketing and digital together under one leader will help foster a greater focus on the full customer lifecycle.” Q2 comps were down 0.9%. Coverage here.
The Kohl’s move is the one to watch, because it is a structural answer to a structural problem. Marketing owns acquisition, digital owns the site, service owns the aftermath, and the customer experiences all three as one company. Consolidating them under a single P&L owner is the only reliable way to stop the internal handoffs from showing up in the customer’s experience. Whether it works at Kohl’s is a separate question, given a negative comp, but the org design is right.
If 20% of your Q4 traffic really does arrive via an AI agent, the thing to check in September is whether your site is legible to one: structured product data, real availability, honest pricing, and content that answers questions rather than gesturing at them. That is a funnel and site readiness problem with about six weeks left to solve it.
The CX Vendor Landscape Is Consolidating Under You
A quick roundup of things that change who you are buying from and what breaks.
Qualtrics cut jobs globally on August 19, three months after closing its $6.75 billion Press Ganey Forsta acquisition. The company declined to give totals; a WARN notice covering Qualtrics Tower in Seattle implies at least 50 at headquarters, since Washington’s threshold is 50 per site. CEO Jason Maynard called the acquisition a “defining milestone” that “meant making hard decisions about what the organization needed to operate and function as a single uniform team.” Maynard joined from Oracle, became CEO in February 2026, and removed five senior executives in April. GeekWire has it. If Qualtrics is your XM platform, ask who your account team is in October.
Zendesk’s August release carries dates you need on a calendar. Copilot intelligent triage (topic, sentiment, language and entity classification) is now included at no extra cost in Suite and Support Professional and above, which is a real giveaway worth turning on. Auto assist now pulls from external sources including Confluence, Notion and web crawlers, and from similar solved tickets. New knowledge connectors for Dropbox, Jira, Freshdesk and Contentful. Live call transcription in Contact Center and Voice. Now the deprecations: AI agents Essential stops receiving fixes August 31, 2026 and is removed December 10, 2026. API tokens are being retired, with unused tokens auto-deactivating since July 28, no new tokens after October 27, and all disabled by April 30, 2027. Migrate to OAuth. Release notes here.
Salesforce’s acquisition of Fin (formerly Intercom) for roughly $3.6 billion, announced June 15, is still pending, expected to close in Q4 of Salesforce’s fiscal 2027. We looked and found no regulatory clearance or revised timeline after June, so treat any “the deal closed” claim with suspicion. The announcement is here. Fin is meanwhile shipping independently: on August 13 it launched “Evals and Releases,” a sandboxed evaluation system that runs multi-turn simulations against Fin, scores them with deterministic checks plus an LLM judge, bundles changes into releases, A/B tests on resolution rate, escalation rate and CSAT, and rolls back instantly. Details here. Whatever you think of Fin, that is the right shape for AI service governance, and most teams running an AI agent in production have nothing like it.
The financials say AI is finally material. NiCE reported Q2 on August 5: revenue $782.3 million, up 7.6%, cloud up 12.6%, and AI ARR of $362 million, now 15% of cloud revenue, with a record quarter for AI bookings. Release here. Freshworks reported August 4: revenue $237.4 million, up 16%, its first GAAP-profitable quarter of 2026, with net dollar retention slipping to 104% from 106%. CEO Dennis Woodside: “Freddy AI Copilot is now attached to over 71% of new enterprise deals.” Release here.
Both are attach-rate stories. AI is being sold into deals that were happening anyway. That is a healthy sign for vendor revenue and tells you nothing about whether the AI works, which brings us back to Gartner’s 42%.
The practical takeaway is the least exciting one: know your renewal dates, know your deprecation dates, and know who owns the migration. Consolidation means somebody else’s roadmap is now your project plan, and platform changes land on whoever configured your systems, which is usually a person who no longer works there.
Our Take on the September CX News
The month resolves into one sentence: the industry has been measuring cost avoidance and calling it experience.
Containment rate is a cost metric. Deflection is a cost metric. Average handle time is a cost metric. Attach rate is a vendor’s revenue metric. Agentic work units are a consumption metric. Every one of those is easy to instrument, lands inside one quarter, and belongs to one department. Meanwhile the ACSI is at 76.1 and falling, complaints are at a record, and 87% of customers are telling researchers that the thing they most want is the thing containment is designed to prevent.
That is not a paradox. It is an accounting artifact. The savings are legible and immediate. The cost is illegible and delayed, and it shows up as churn on a different P&L in a different fiscal year. Given those incentives, of course three-quarters of leaders are increasing AI investment while a quarter of use cases are actively destroying value. The system is working exactly as designed. It is just measuring the wrong thing.
Two people quoted in this roundup said the same true thing from opposite directions. Info-Tech’s Julie Geller: delaying contact with a human is not the same as resolving the problem. Gartner’s Eric Keller: making it easier to reach a person increases AI adoption. Both are describing a service design where AI handles what it is good at, the exit is obvious and fast, and success is measured by whether the customer’s problem went away rather than by whether they gave up.
The thing that makes September the deadline rather than a nice idea is Q4. If Salesforce is even directionally right that a fifth of holiday traffic arrives via an AI agent, and your own service AI is optimized to prevent people from reaching a human, you have built a Q4 where a machine sends customers to a machine that is trying not to let them talk to anyone. That is a fine way to hit your cost target and a memorable way to convert a record complaint rate into next year’s churn number.
So: inventory every AI touching a customer, EU exposure or not. Add resolution rate and escalation-to-human rate to the same dashboard as containment, and let them argue. Check whether your loyal customers pay more than your churn risks. And make the path to a person short and obvious, because that is what 87% of your customers just said they need, and because it turns out to be the thing that makes the AI work better anyway.
The cheapest CX advantage available in a market this dissatisfied is being noticeably easy to deal with. Nobody has to invent anything to do that.
September 2026 CX Events
Customer Connect Expo 2026 September 9-10, Georgia World Congress Center, Atlanta, GA customerconnectexpo.com A two-day CX and contact center expo drawing 2,000-plus attendees, 100-plus exhibitors and 50-plus speakers, with CX directors, contact center leaders and support heads as the core audience. The 2026 speaker list includes Zendesk, Microsoft, AWS, Foundever and CXPA. Good for evaluating vendors in volume without a week away from the desk.
Dreamforce 2026 September 15-17, Moscone Center, San Francisco, CA (Trailblazer Bootcamp September 12-14) salesforce.com/dreamforce Salesforce’s flagship, themed on the “Agentic Enterprise,” with heavy Service Cloud and Agentforce content across 1,600-plus breakouts. If you are evaluating agentic service, this is where the category’s biggest vendor makes its case in full. Streams free on Salesforce+.
UNBOUND 2026 September 16-18, Boston Convention & Exhibition Center, Boston, MA unbound.hubspot.com HubSpot’s rebranded flagship (formerly INBOUND), with 200-plus sessions across marketing, sales, service and ops. The service track is the relevant one here, particularly for mid-market teams running Service Hub. The site currently shows the event as sold out.
AI for Customer Support Summit San Francisco September 22-23, Hyatt Regency SFO, Burlingame, CA events.customersuccesscollective.com/location/supportsf The most contact-center-relevant event of the September calendar: AI-first support playbooks, smart escalation design, and human-plus-AI workflow redesign. Support and service directors and VPs from Planet, Sentry, Rapid7, PayPal and BitGo, with Intercom, Ada, Kustomer, Balto and Hiver as partners. If your team is wrestling with the containment-versus-resolution problem, this is the room.
Customer Success Summit San Francisco September 22-23, Hyatt Regency SFO, Burlingame, CA events.customersuccesscollective.com/location/sanfrancisco Co-located with the support summit above, and one ticket covers both. Roughly 700 attendees, 150-plus speakers, about 75% senior management, across two tracks on net revenue retention, churn prediction, agentic AI and scaling CS. Speakers from Asana, Docusign, Salesforce, Mastercard and Notion.
Chief Customer Officer Summit San Francisco September 24, Hyatt Regency SFO, Burlingame, CA events.customersuccesscollective.com/location/ccosanfrancisco An executive-only, single-day capstone to the San Francisco week, aimed at CCOs and heads of customer. Relevant to the Kohl’s story above: this is the peer group for anyone building or inheriting a consolidated customer function.
Shoptalk Fall 2026 September 29 – October 1, Nashville, TN fall.shoptalk.com Retail and ecommerce strategy for 3,500-plus leaders, roughly a third of them C-suite, with a substantial customer experience and retail tech track. Timed for 2027 planning, which makes it the right place to pressure-test whatever your Q4 teaches you.