The Bot Needs a Hall Monitor

Plus: bank-wide AI risk training, agents without owners, customer-service fatigue, and the very human job nobody should forget to assign.

Jun 22, 2026

Your daily signal on AI and CX — minus the hype.

DCX Stat of the day: NatWest is offering all 60,000 employees a two-to-three-month course on the ethical risks of using AI. The Times

In this issue:

→ Every role gets an AI judgment test
→ Oversight becomes part of service
→ Bot fatigue gets more expensive
→ Repetitive work gets a teammate
→ Recovery needs a human owner

🔍 DEEP DIVE

Someone Still Has to Raise Their Hand

NatWest is taking AI training across the whole bank. All 60,000 employees are being offered a course on AI ethics and risk as the bank puts the technology deeper into customer service, operations, and everyday work.

Fine. Good.

But the training module is not the interesting part.

The customer will never see that. They will see the answer, the recommendation, the flagged transaction, the delayed handoff, the strange refusal, or the employee trying to explain what the system just did.

That is where this gets real for CX. AI literacy is not about making everyone sound fluent in model risk. It is about helping ordinary employees recognize the moment where a confident system is about to create a customer problem.

That moment is easy to miss because the machine sounds finished. It gives an answer. It writes the summary. It classifies the case. It recommends the next step. The employee has to know when to pause anyway.

Bottom Line: The human job in AI-enabled service is to know when the system sounds finished but the customer risk is still unresolved.

📬 Copy-Paste Take

AI training only matters if it changes what happens in the moment. Can an employee question the answer, slow down the workflow, escalate the edge case, and explain the decision to the customer? If not, the company has trained people to use AI without training them to protect the experience.

🧭 OPERATOR PLAYBOOK

Put the Pause Button Where Work Happens

Pick one journey where employees are already using AI behind the scenes.

Support summaries. Fraud review. Complaint classification. Product guidance. Financial advice prep. Eligibility checks. Case routing. Follow-up messages.

Then look for the place where the employee is most likely to trust the machine because it saved them time.

Audit that moment for four risks:

Input: What customer data is the employee feeding into AI?
Confidence: What makes the answer look more certain than it is?
Authority: What can the employee decide after using it?
Recovery: Who owns the fix if the AI-shaped work hurts the customer?

Then put the red flags inside the actual workflow.

Do not bury them in an annual training deck. Put them where the employee is about to send, decide, approve, deny, summarize, or escalate.

Ask your team: Where does an employee need permission to say, “Wait, this one needs a person”?

Signal: The best AI control may be a well-placed pause button with a clear owner behind it.

📊 MARKET REALITY CHECK

The Risk Budget Is Following the Bot

Here is the banking number that makes NatWest’s training move feel less like HR hygiene and more like operating reality.

KPMG’s 2026 Banking Technology Survey says 80% of banking executives expect AI to significantly disrupt their business and operating models in the next three to five years. The risk spending is already following: 84% are increasing cybersecurity investment specifically to address risks introduced by AI.

The threat list is not theoretical either. Banking leaders named AI-introduced code vulnerabilities, deepfakes, AI bots, and securing agentic technologies as emerging threats.

That changes the CX conversation. When banks add AI to fraud, payments, access, service, personalization, and internal work, cybersecurity is no longer just protecting the system. It is protecting the customer’s ability to trust what happened, understand what changed, and recover when something goes wrong.

Why it matters: AI risk is moving into the same rooms as customer access, payment trust, identity, and service recovery. If CX is not in that conversation, the customer impact will be designed around later.

AI adoption + customer access = risk design problem

🧰 TOOL WORTH KNOWING

Convey

What it does: Convey builds AI “teammates” for repetitive operational work. The company positions the product around outcomes rather than single-task agents, with early customer references including NBCUniversal, Samsara, TelevisaUnivision, Unity, Faire, and ChargePoint.

CX use case: Useful for teams with too much customer work stuck in the ugly middle: status checks, internal follow-ups, order updates, case prep, exception handling, and the manual cleanup nobody puts in the journey map.

Worth watching because: The customer experience often breaks in the work around the interaction. Someone is checking systems, moving data, reconciling status, sending updates, or cleaning up after a process that never should have required that much human glue. If AI teammates can take on the truly repetitive pieces, employees can spend more time on judgment, relationship, and recovery.

Bottom line: Convey is worth watching when the customer problem is hiding behind the conversation, in the operational drag that slows everything down.

The DCX AI Today - AI Tool Directory - If you lead a CX team and want a curated shortlist of tools worth evaluating, this is your starting point.

📡 90-SECOND CX RADAR

Customers Are Tired of the Bot Maze

Guardian readers described customer service in 2026 with a lot of anger and very little patience. About one in 10 responses called out automated chatbots as a hurdle to resolving problems, especially when the issue involved billing, fraud, broken products, deliveries, or medical needs.

That is the part companies keep underestimating. Customers do not grade automation on whether it answered something. They grade it on whether it got them unstuck.

Why it matters: If AI handles the clean task and blocks the messy one, the company has moved the cost of service onto the customer.

✅ YOUR MOVE

The useful AI question this week is simple: who is allowed to stop the machine?

Because that is where a lot of customer risk will sit. Not in the strategy deck. Not in the model announcement. In the small moment where an employee sees an answer that looks right enough and has to decide whether to keep moving.

Pick one AI-assisted workflow and run a failure drill.

What happens if the answer is wrong, the customer data is incomplete, the system goes down, or the employee cannot explain the recommendation?

If the team has no practiced response, that is useful information. It means the current operating model is depending on luck, employee judgment, and customer patience.

The customer does not need everyone to become an AI expert. They need someone with the authority to stop, explain, and fix the moment when AI is wrong.

Until tomorrow,