Practical Lessons from Early AI Adoption in Support
This article was created in collaboration with Embrace and draws from our recent webinar, The AI Playbook for Zendesk Support Leaders, where support leaders shared practical, real-world lessons from using AI inside Zendesk.
AI is taking on a larger role across support operations, especially as teams explore how to simplify workflows and improve speed without major system changes.
With that in mind, we partnered with Embrace.ai to host a webinar that brought together experienced operators to show what practical AI adoption looks like inside real support environments.
The conversation featured:
Tim Guinan, Sr Manager, Digital Customer Experience at WP Engine
Jacob Harper, Director of Global Support Operations at Symplicity
Both speakers walked through what drove them toward AI, the challenges they faced along the way, and the measurable impact they have seen so far.
This article distills those insights into a practical take-home resource for readers.
Why Explore AI in the First Place
Neither WP Engine nor Symplicity pursued AI because it was trendy. They were responding to real operational pressure.
Symplicity needed to support a global 24/7 environment across 13 different product lines without continually adding headcount.
WP Engine had a vast but hard to search knowledge base of more than 1,000 technical documents that were hard for agents to search quickly during live chats.
They needed to transform it to a conversational AI resource for both internal employees and external customers.
While their contexts differ, the underlying need was similar.
Both teams turned to AI to simplify knowledge retrieval, reduce repetitive tasks, and create more scalable workflows.
Impact After Implementation
Both teams shared clear, quantifiable improvements:
WP Engine was able to deploy AI support via Slack to more than 200 support technicians providing an “always on” access to the extensive knowledge base. This resulted in WP Engine reducing handling time by two minutes per chat
For the more than 200,000 WP Engine external customers, the effectiveness of AI-powered search resulted in a 45% reduction in the volume of chats generated by customers who were previously unable to self-serve
Symplicity cut the first reply time 38%, from four hours to two and a half
Symplicity achieved 12 to 19 percent ticket deflection in its first quarter using AI
During the first 3 months of using AI powered support, CSAT jumped by 12 points
Moving AI inside Zendesk increased WP Engine’s agent usage by 50%
Both companies saw fewer repetitive tickets as customer-facing AI took over basic guidance
The ROI was operational, measurable, and fast.
None of these improvements required restructuring Zendesk or hiring additional staff. They came from tightening workflows, improving knowledge sources, and rolling out AI in intentional phases.
A Three-Phase Approach to Rolling Out AI
Both speakers emphasized that success depended on a phased approach that protected quality and minimized risk.
1.Crawl. Internal Only
Each team began with a small internal pilot group to build trust and tune the model. Documentation updates played a major role here because good AI depends on good content.
2. Walk. Customer-Facing Rollout
In the Walk phase, internal use had reached a level of reliability that allowed teams to expand externally. AI became the first step in their Help Centers or was embedded directly into product dashboards. Rollouts were intentionally slow and driven by ongoing feedback.
3. Run. Full Operational Expansion
With stability established, they expanded into broader workflows:
One-click ticket summaries
AI-powered QA scoring
Recommended next steps in Zendesk
Embedding AI directly inside the customer-facing product experience
This stage focused on removing friction across the entire support ecosystem, not just deflecting tickets.
10 Lessons Support Leaders Can Apply Right Away
Here are the top takeaways that emerged from the conversation, applicable to any team exploring AI.
1.Start with your internal team before involving customers
Early trust from agents is essential. Internal rollouts reveal issues before customers can feel them.
Tim and Jacob trained the AI using real tickets, real workflows, and real feedback before exposing it to customers. This dramatically reduced risk and increased confidence in the tool.
2. Don’t rush the implementation
Both organizations wanted to move quickly but found that slower, phased rollouts created better long-term adoption and fewer surprises.
3. Documentation is the backbone of good AI
If your AI answers are wrong, your documentation is probably the real issue.
WP Engine focused heavily on keeping articles clean and up to date so the AI stayed reliable without constant overrides.
AI models only perform as well as the knowledge behind them. Fixing source content is more effective than patching answers.
4. Change management matters more than technology
Agent concerns about replacement are real. Consistent communication reduces friction.
Both teams framed AI as a support layer that accelerates agents, not a system that replaces human judgment.
5. Keep early pilot groups small and strategic
Select agents who provide thoughtful feedback and influence their peers. Quality feedback early on leads to much faster refinement.
6. Put AI directly inside the workflow
Usage spiked when WP Engine moved AI from Slack into Zendesk. Keeping the tool where work happens increased adoption immediately.
7. Use one-click actions to drive adoption
Quick wins matter. Features like one-click summaries and rapid knowledge retrieval helped teams see value instantly, which built long-term confidence.
8. Monitor feedback continuously
Negative ratings on AI answers are opportunities to refine documentation and improve agents. Both teams monitored feedback closely to guide updates.
9. Don’t rely on overrides when documentation is outdated
Overrides solve short-term problems but create long-term maintenance gaps. Updating original documentation keeps everything clean and reduces future errors.
Both teams were also intentional about which ticket types were appropriate for automation.
10. Choose partners who prioritize flexibility and collaboration
Both teams emphasized the value of working with vendors who iterate quickly, separate data sources cleanly, and don’t push long-term lock-in.
Closing Reflections
AI is already helping Zendesk support teams work more efficiently, deflect routine questions, and streamline agent workflows.
The companies seeing the most success are not the ones pushing rapid adoption. They are the ones treating AI as an iterative process grounded in documentation, feedback, and trust.
If you want to explore these ideas in more depth, you can watch the full webinar recording that this article is based on.
