Sales Operations Manager

Handling complaints in Zendesk: structure beats good intentions. Complaints are a different category of work from support requests. How to mark, route, detect and escalate them - and why AI should detect complaints but never answer them. Key takeaways. * Complaints are a different category of work from support requests, and most instances treat them identically - that's the root of nearly every complaint process that 'doesn't work'. * A support request wants an answer; a complaint wants acknowledgement, ownership and a decision, often from someone with authority the first agent doesn't have. Routing both through the same queue with the same SLA produces exactly the 'passed around' experience customers describe. * Separate complaints as a distinct object - a ticket type, custom field or tag - so you can route, report and review them as a category and answer 'how many complaints did 3Li Global get last quarter?', usually the first question anyone asks. * Give complaints their own, tighter first-response SLA: the acknowledgement is doing most of the work even when full resolution takes longer. Complaints are a different category of work from support requests, and most instances treat them identically. That's the root of nearly every complaint process that "doesn't work." A support request wants an answer. A complaint wants acknowledgement, ownership and a decision - often from someone with authority the first agent doesn't have. Routing both through the same queue with the same SLA produces exactly the outcome customers describe as being passed around. Separate complaints as a distinct object. You don't need a new system. You need the instance to know a complaint when it sees one. Give it its own type or field. Whether that's a ticket form, a custom field or a tag matters less than consistency. Without a marker, you cannot route, report or review complaints as a category - and you cannot answer "how many complaints did we get last quarter?" which is usually the first question anyone asks. Give it its own SLA. Complaints justify a tighter first-response target than general requests, even where full resolution takes longer. The acknowledgement is doing most of the work emotionally. Give it its own queue and owner. Complaints handled by whoever is next in rotation get inconsistent outcomes. A named owner produces consistency, and consistency is what regulators, executives and customers all actually want. Detect them, don't wait to be told. Customers rarely write "I would like to make a complaint." They write in a recognisable register, and you can catch a meaningful share automatically. Zendesk's 2026 release includes intelligent triage for classification and Forethought AI agents covering intent identification (Zendesk, July 2026). Applied here, the useful pattern is flagging for human attention - not automating the response. Practical detection signals worth building on: * Escalation language and repeat contacts on the same issue * A reopened ticket, which is one of the strongest predictors * Explicit references to cancelling, regulators, reviews or legal action * Sentiment classification, treated as a prompt to look, not a verdict Route detections to a human queue. Auto-responding to a complaint with an AI message is the fastest way to escalate it. Never let an AI agent handle a complaint end to end. Worth stating plainly. The technology can generate a fluent, apologetic, well-structured reply. It should not be the one handling a distressed customer, because: * The cost of being wrong is asymmetric - a wrong answer on an order lookup is an inconvenience; a wrong answer on a complaint is a formal escalation * Complaints often need a decision (a refund, an exception, a goodwill gesture) rather than information * Customers who are already unhappy read an automated reply as being dismissed Use AI to detect, classify, summarise and prepare context for the human. Not to respond. Most complaint processes fail at the handover, not the intake. Decide in advance: What triggers escalation - a breach, a reopen, an explicit request, a sentiment flag, a named account. Write it down. Who receives it - a person or a role, not "management." Ambiguity here means complaints sit while everyone assumes someone else has it. What travels with it - the full history, what's already been offered, and what authority the receiving person has. Escalating a bare ticket means the customer re-explains, which compounds the original complaint. What the resolution authority is - how much can be approved without further sign-off. Teams without a stated limit escalate everything, which defeats the purpose. Close the loop deliberately. Two things distinguish complaint handling from support: Confirm the outcome explicitly. Not "your ticket has been solved," but what was decided and what happens next. Solving a complaint silently reads as being ignored. Check whether it recurred. A complaint that produces a repeat contact three weeks later wasn't resolved - it was closed. Track reopen rate on complaints specifically; it's the metric that tells you whether the process works. Report on causes, not volume. Complaint volume alone tells you almost nothing - it moves with traffic, seasonality and how well you detect them. More useful: * Cause categories, so complaints feed product and operations rather than dying in support * Time to acknowledgement as distinct from time to resolution * Reopen rate * Escalation rate, and whether it's rising The point of complaint reporting is to change something upstream. If your complaints data never reaches the team that could remove the cause, you're measuring for its own sake. The quality-review angle. Zendesk's 2026 release added a coaching dashboard tracking coaching impact against Internal Quality Score. Complaints are the highest-value sample to review. They're where tone, judgement and authority matter most, and where the gap between agents is widest. Reviewing a sample of complaint interactions monthly does more for consistency than any amount of general QA. The short version. Mark complaints as distinct, route them to a named owner, acknowledge fast, keep AI on detection rather than response, define the escalation path in advance, and report on causes. Most of that is structure, not software - which is why instances with the same platform produce very different complaint outcomes. Frequently asked questions. Why doesn't my complaint process work in Zendesk? How should I structure complaints in Zendesk? Should complaints have a different SLA? Free checklist The Zendesk implementation checklist. Everything to get Zendesk live cleanly - channels, SLAs, triggers, AI agents, data migration and a go-live test script.

Agents think in milliseconds, legacy infrastructure doesn't. LinkedIn, Walmart and Zendesk shared how they closed the gap at VB Transform 2026. 11:48 am, PT, July 17, 2026 Legacy infrastructure, not the models themselves, is what's actually slowing AI agents down. That was the shared conclusion of three infrastructure leaders - from LinkedIn, Walmart, and Zendesk - at VB Transform 2026. The panel brought together Animesh Singh, senior director of AI platform and infrastructure at LinkedIn, Desiree Gosby, SVP of corporate technology services and technology strategy at Walmart, and Sami Ghoche, VP of applied AI at Zendesk, each describing what actually broke when they moved agents from pilot to production. Each arrived at the same conclusion from a different starting point: None of the bottlenecks they hit were model problems. Keep Watching What tied their answers together was a shared premise: most enterprise infrastructure was built for how humans work, not for how agents work. The gap between those two speeds is where the real engineering happened. Gosby put it plainly when asked what she'd learned scaling agents inside Walmart's own workforce. The goal, she said, is to make sure "engineering doesn't once again become the bottleneck for what it is we're trying to do." Where the bottleneck actually was. Each company hit a different version of the same wall: infrastructure designed for how people work doesn't hold up once agents are doing the work instead. At LinkedIn, the first bottleneck wasn't a model, it was Kubernetes, which assumes containers spin up on demand, a process that takes seconds. Singh said that's too slow for agents. The fix was moving from on-demand provisioning to pre-provisioned pools of containers that swap agentic workloads in and out in real time. A second, harder problem surfaced once LinkedIn let agents control their own orchestration. A five-point evaluation system looked clean, but hallucination kept showing up anyway. Singh said the issue was structural, an LLM evaluating another LLM's output shares the same failure mode as the thing it's evaluating. "We built our own harness, our own control flow, and pushed the LLMs to the leaf instead of them orchestrating the loop," Singh said. Roughly 80% of the workflow is now scripted, deterministic code, with LLMs used only where reasoning is required, and each step's evidence is committed to disk before the system moves on. Walmart's bottleneck came from success. An agent harness put directly into employees' hands went viral internally, and what Gosby called "citizen developers" began building their own agents to solve problems that once required a formal engineering roadmap. The upside was real innovation. The downside was duplication, dozens of overlapping agents with no coordination. The fix wasn't reining in the harness, it was building governance to spot duplication, promote the best version of an agent, and get it into production without engineering becoming a chokepoint. Zendesk hit its bottleneck from the data side. Ghoche, who joined through Zendesk's acquisition of Forethought, which closed in March 2026, described sitting on what he called a public figure of 20 billion customer conversations in Zendesk's repository. The instinct is to hand that history to a large language model with a big context window and let it generate the agents a business needs. Ghoche said that doesn't work. "You can't really do that, so instead you have to really invest in the underlying data pipelines and all the data infrastructure that comes with that," he said. The role of open source. On open source, all three leaders landed on a similar instinct: own what you can, and lean on frontier labs only where they still have a clear edge. Ghoche said his own view is that most enterprises would prefer to own their models and infrastructure wherever that's possible, and that reasoning is what drives Zendesk's own approach. The exception is frontier reasoning work, where the labs still lead, though he said that slice of use cases is shrinking relative to everything else enterprises now do with AI. LinkedIn's answer was to build two subsystems specifically for independence. The first is what the company calls an AI gateway, a single interface that every outbound call to a model runs through regardless of provider. The second component is a memory subsystem built to hold context independent of any model provider. "Every single outbound call going to an LLM, whether it's on a public cloud or on-prem in our own data centers, follows the same semantics, the same API calls. We can quickly switch between different providers," Singh said. Walmart built its own internal gateway to stay vendor agnostic across three workload types: fully deterministic workflows, planner-and-reasoner workflows for open-ended tasks, and a hybrid of the two. Compliance-heavy work stays deterministic by design; governance, security and evaluation run through the gateway regardless of which model is on the other end. Gosby said the choice between a frontier model and an open-weight model comes down to whichever is most effective for the specific workload, not a fixed policy. Advice for the modernization journey. Three pieces of advice came up directly, each tied to the wall a leader had already hit. Invest in evals before anything else. Ghoche called it the thing common to every use case, internal or customer facing. "The thing that's common to all of these is evals. It'll force you to break the problem down, and once you have a robust set of evals, you can move a lot faster," he said, Own your agent harness from day one. Gosby's advice was to put the AI harness directly in employees' hands early, paired with the infrastructure to monitor what it produces. "It will unlock a huge amount of innovation," she said. Build for model and context independence. Ensuring flexibility is critical for success. "Build for independence, whether it's a frontier model of today versus an open source model of tomorrow," Singh said. "Keep that context within your enterprise so that you can reuse it when you ship the model or the harness tomorrow," Singh said.

Engineering an omnichannel AI support ecosystem for MacPaw. Executive summary. MacPaw, a leading software developer renowned for products like Eney, CleanMyMac, ClearVPN, and Setapp, manages a massive global user base. Navigating customer support across multiple distinct software ecosystems, multiple languages, and varying time zones created a complex operational bottleneck. To relieve engineering pressure and eliminate support backlogs, Triviat architected and executed a multi-channel AI integration strategy. By embedding custom AI layers directly into MacPaw's existing tech stack (Zendesk, Intercom, and Freshcaller), Triviat automated first-line support, eliminated language barriers, and maintained an intelligent human-in-the-loop framework. The resulting ecosystem delivered end-to-end automation for routine queries while drastically improving overall support efficiency. The challenge: Channel fragmentations & scalability at global volume. With a diverse product portfolio, MacPaw's support volume is as varied as its user base - ranging from deep technical VPN diagnostics to routine inquiries regarding Setapp billing. To maintain their high standard of customer satisfaction, MacPaw faced three distinct operational challenges: * The Multilingual Email Bottleneck: High volumes of incoming support emails in foreign languages strained agents, who had to manually translate text back and forth outside their primary ticket workflows. * Chat Congestion on First-Line Support: Inundations of repetitive, easily answerable queries on ClearVPN tied up Tier-1 agents, reducing their ability to tackle high-friction technical issues. * Voice Channel Elasticity: Managing an influx of phone calls on Setapp's dedicated US toll-free hotline during specific operational hours required a smart, responsive routing layer to handle call volumes efficiently without sacrificing the customer experience. MacPaw needed a partner who could handle the technical implementation end-to-end, building smart integrations tailored exactly to their existing workflows. The solution: seamless automation with a human touch. Triviat designed, built, and deployed three purpose-built AI agents to act as the intelligent "brain" across text, chat, and voice channels. 1. Multilingual Email translation assistant (Zendesk integration). To democratize communication within MacPaw's core mail system, Triviat integrated a localized language-detection helper natively into Zendesk. * How it works: The AI automatically scans incoming emails and identifies the customer's native language. If it isn't English, the system translates the text instantly so the agent can understand the context immediately. * The Loop: The support agent drafts the resolution natively in English. With a single click, the AI renders the final outbound response back into the customer's original language, maintaining a flawless localized conversation without the agent ever leaving Zendesk. 2. Autonomous first-line chatbot (ClearVPN via Intercom). For ClearVPN's fast-moving live chat ecosystem, Triviat engineered an AI chatbot backed entirely by MacPaw's comprehensive internal knowledge base. * Smart Containment: The bot acts as a first-line shield, successfully handling the majority of routine customer requests completely autonomously. * Frictionless Handover Logic: To ensure a premium customer experience, the bot operates on a strict human-in-the-loop principle. If a user explicitly requests a human, or if the bot reaches a standard 3-to-4 response threshold without a clean resolution, it silently steps down. The system automatically reassigns the ticket and updates the Intercom status from "In Progress" to "Open," signaling a live agent to step in with the full chat transcript preserved. 3. Voice AI assistant (Setapp via Freshcaller). Triviat breached the voice space by deploying a conversational voice bot for Setapp's toll-free US customer service line. * How it works: Integrated alongside Freshcaller, the AI greets inbound callers, processes natural language intent, and handles initial inquiries right away. * Intelligent Routing: If the inquiry requires complex troubleshooting, the AI seamlessly transfers the call to a live agent, while Freshcaller aggregates the call logs, metrics, and analytics for backend tracking. The impact: reliable integration, measurable efficiency. By entrusting the end-to-end technical implementation to Triviat, MacPaw successfully decoupled its support volume from direct human headcount. The integration delivered an immediate boost to operational efficiency while maintaining the positive, premium experience MacPaw users expect. Key operational impacts realized by the collaboration include: * Workflow Optimization: Human agents were freed from translating emails and answering repetitive FAQs, allowing them to focus strictly on complex, high-tier technical issues. * Smooth Escalations: The hybrid AI-to-human handover process proved highly reliable, ensuring customers never hit "dead ends" when interacting with automated systems. * Turnkey Execution: Triviat handled the entire technical pipeline seamlessly, adapting the AI architecture precisely to MacPaw's specific operational guardrails. The next phase. Following the success of this initial rollout, MacPaw and Triviat are already planning the next horizon of support automation. The upcoming phase will feature deeper database integrations - allowing the AI assistants to securely access account-level details like individual Account IDs, active license numbers, and subscription expiration statuses to handle account management queries autonomously. Furthermore, plans are underway to scale Triviat's conversational AI support across five to six additional legacy MacPaw products. The Triviat standard: architecture over hype. This project highlights Triviat's core philosophy: Triviat don't build isolated AI features; Triviat engineer cohesive automation pipelines. By handling the technical implementation end-to-end, its team ensures that complex integrations fit perfectly within an organization's existing workflows. Triviat focus on bridging the gap between cutting-edge LLM capabilities and practical, reliable business infrastructure - delivering scalable systems that work predictably from day one. Looking to integrate intelligent automation into your existing tech stack? Book a demo with its team today!