Customers rarely think about security until it gets in their way. A forgotten password, an extra verification code or a stalled refund can turn a smooth journey into a frustrating one, and the brand takes the blame whatever the cause. That is why customer experience platforms are changing how they approach protection, building permissions, audit trails and policies into the systems teams already use so safeguards apply automatically and run quietly in the background.
The shift matters more as AI agents take on real work, from sorting service tickets to issuing refunds and approving suppliers. Each agent needs enough access to do its job and no more, and every action it takes needs to be recorded and reversible. Personalisation raises the stakes further, because deeper personalisation draws on more customer data, which makes the question of who and what can read that data part of the experience itself.
Three leaders share how this works in practice: Chris Jordan, Regional Vice President for ASEAN at monday.com; David Hall-Johnston, Chief Technology Officer for APAC at Zendesk; and Eric Weight, Head of Sales Engineering for APJ at Qualtrics. Their perspectives begin with Jordan, who looks at how security travels with the everyday work of the business.
Security lives where the work already happens
Most of the AI progress in retail is happening in work customers never see. Jordan of monday.com, points to service tickets being sorted, campaign content being drafted, supplier records being updated and store tasks being kept on schedule. Customers feel the results as faster replies, fewer errors and offers that suit them better.
That makes security a question of what each AI process is allowed to read and where its output goes. Jordan says agents should be held to the same permissions, security and governance rules as the employees who do that work, and the access they get should match the task in front of them.
“A refunds agent needs order and payment status. It does not need the full customer record,” Jordan said.
The principle becomes harder to apply as retail runs on a web of suppliers, marketplaces, agencies and apps, with every connection a possible route for data to leave the business. His answer is to keep AI inside systems the business already governs, so rules do not have to be rebuilt for each new tool or partner.
“Brands should keep AI inside systems they already govern, so any agent, including one a partner brings in, inherits the existing access rules. External agents should sign in and be verified like a person and should reach only the data their role allows,” Jordan said.
Governance also has to work after something goes wrong. Accountability stays with the business whether a person or an agent made the mistake, and a customer sees a pricing error the same way regardless of its cause. Jordan adds that leadership, IT and security, legal and business teams each have a defined role, with a named person answerable for every AI workflow.
“Teams need records showing what the agent did, when it acted and what data it used. They also need a clear escalation path and the ability to undo the action,” Jordan said.
When permissions, audit trails and reversibility sit inside the systems teams already use, they apply without anyone having to remember them. For the customer, that shows up as quick service and fewer mistakes, with protection working quietly behind it.
If security sits inside the work, the next question is how it shows up for the customer on the other side. David Hall-Johnston of Zendesk, looks at that point, where strong protection has to work against scams and impersonation without adding friction to the journey.
Security follows the work to the customer’s front door
Once security is built into the work, the next test comes when the work reaches the customer. Hall-Johnston sees security as the part of personalisation that happens behind the scenes. Customers want a brand to remember their preferences and understand their issue, but they also expect clear limits on how their information is used.
Those limits are harder to hold when shoppers bring their own AI assistants into the journey, because the brand now has a second party to account for.
“When a customer uses their own AI assistant, brands need to verify not just the customer’s identity, but whether the assistant is authorised to act on their behalf—and which actions that authorisation covers. Any agent helping with an order should access only the information and perform only the actions needed for that task, within clearly defined permissions,” Hall-Johnston said.
He adds that collecting more data does not make personalisation better by itself. Brands should collect only what they need, explain how it is used and build security into the experience.
Scammers add another layer, reaching shoppers directly through fake stores, hijacked accounts and phishing. Hall-Johnston treats impersonation and account abuse as customer experience issues as well as security issues, so brands need protection that is strong without making the journey harder.
“They should also offer strong authentication, detect suspicious account activity, and require additional verification for sensitive changes. Support teams need clear procedures so that someone cannot bypass those checks through a convincing conversation, whether they are speaking to a person or an AI agent,” Hall-Johnston said.
The same care applies to the tools around the conversation, such as plugins, payment tools and chatbots. Reviewing the underlying AI model is only part of the job, so Zendesk tests the application, its integrations and complete customer workflows through its Govern, Guard and Ground approach.
“Our testing uses realistic conversations across multiple turns, with repeated evaluations. That scrutiny must continue as tools and workflows change,” Hall-Johnston said.
The stakes are high, as consumers are willing to switch after just one bad experience. Safeguards built into the service conversation, and tested as it changes, keep help easy for customers.
Whether those safeguards earn trust shows up in what customers say and do. Eric Weight, Head of Sales Engineering for APJ at Qualtrics, looks at how that trust can be measured.
Trust grows when customers share more
Trust is where the earlier safeguards show their value. Weight points to a gap that sits at the centre of customer experience. Customers want tailored experiences, yet many are unsure whether brands will handle their data responsibly, and that doubt shapes how much they are willing to share.
For Weight, that gap turns personalisation and security into a single job. “Personalisation runs on customer data, and every piece of data a brand holds is something it has to protect. So the two can no longer sit with separate teams,” Weight said.
The answer is not always more data. Weight says brands can stay relevant without holding more than they need.
“There are ways to deliver great experiences with limited personal data. The most relevant experiences come from understanding a customer’s context in the moment, what they are trying to do and where they are in the journey, more than from building a detailed profile,” Weight said.
From there, he says, brands should ask only for what they need, explain why and let people control or delete what they have shared. When customers can see the exchange is fair, sharing data feels less like a risk.
That care has to reach the tools that now handle much of the journey, from chatbots to AI agents and third-party services.
“Every chatbot, agent or third-party tool in the journey handles customer data, so each needs the same discipline – limits on what it can access, rules on what it must never say or do, and a named owner when it goes wrong. That playbook and governance structure has to exist before an incident,” Weight said.
A bot judged on how many conversations it keeps away from people can count a customer stuck in a loop as a success, so Weight says brands should ask customers how the tools are working and act on what they hear.
The cost of failure is often silent, because many customers who have a bad experience say nothing and quietly move on. A data incident touches the very thing customers already worry about, so earning trust back depends on behaviour they can see.
When customers feel safe enough to share more, the safeguards have done their job.
The future of CX is security that feels like service
Looking ahead, the three leaders describe security that works the way good service does, present when it is needed and unnoticed when it is not. Their answers point the same way: built into daily work, light on friction and easy to prove.
For Jordan, the future starts with an inventory of every workflow where AI touches customer, supplier or pricing data, and ends with governance that runs on its own. “After that, make governance part of daily work. Run AI inside the systems teams already use, so permissions and audit trails apply automatically, and keep the number of standalone tools low,” Jordan said.
Hall-Johnston looks at the same future from the customer’s side of the screen, where protection should add as little friction as possible. “The goal is to remove unnecessary friction while preserving meaningful protection. Across shopping, payments, and support, brands should maintain records of who acted, with what permission, and what happened, while making it easy for customers to reach a person when something does not look right,” Hall-Johnston said.
Weight adds that trust itself should be measured, since a safeguard is only as good as the confidence it creates. “Measure trust directly, alongside satisfaction and NPS. A practical way is to ask customers three things: does the brand deliver what it promises, does it act fairly, and does it put customers ahead of short-term gain,” Weight said.
When security is built in, tested often and measured by trust, it stops feeling like a barrier and starts feeling like service.
**
Read side by side, the three leaders agree that the best customer experience is protected before the customer arrives. Their safeguards line up as one chain, starting in different places. Jordan starts inside the work, with access rules and audit trails that travel with every task. Hall-Johnston starts at the customer’s front door, where checks must hold without adding friction. Weight starts with trust, which only counts when it is measured.
Before an AI agent or CX platform goes live, businesses should settle three questions. What can each agent see and do? Can the business show what happened and undo it? And will customers feel the protection as ease? Brands that can answer all three can serve customers quickly with confidence, because freedom grows with the ability to control what sits behind the service.
