Agentic AI is next - but only if customers trust it

Can we trust AI to power customer communications? A support leader's playbook for 2026

AI is no longer optional in customer support. It's in your queue, your inbox, and your IVR. The real question is whether customers trust it enough for you to scale it.

Recent research from Sinch shows clear momentum and clear friction. Support leaders who focus on trust, routing, and measurable outcomes will win the next cycle.

What customers and leaders are saying

• 98% of companies across retail, healthcare, tech, and financial services are using or plan to use AI in customer communications.
• 63% plan to adopt AI voice assistants in 2025; 46% are investing in AI chatbots; 35% made AI and automation a strategic investment this year.
• 34% of leaders worry about consumer perception of AI in communications.
• 42% of consumers would trust AI trained on a company's support docs. 52% trust AI for basic answers like order status.
• Generational split for AI support: Gen Z 72%, Millennials 58%, Gen X 39%, Boomers 20% willing to use it.

Channel preference still matters. Only 5% of consumers pick AI chatbots as their first choice for service. Email leads (31%), followed by live chat (22%) and phone with a human (19%).

Build the right foundation before you scale

AI is only as good as the system around it. Beyond the bot, you need reliable delivery, intent routing, and strong data handling.

• Messaging APIs that scale across SMS, email, voice, and in-app.
• Clear data flows with consent, encryption, and regional compliance.
• Observability: transcripts, redaction, analytics, and model performance tracking.
• Fail-safes: graceful handoffs, queue awareness, and channel fallback.

This groundwork is what enables the next step: agentic AI that acts on behalf of customers with oversight.

The trust playbook for AI support

• Disclose clearly: label AI, explain why it's used, and give an easy switch to a human.
• Start with high-fit use cases: order status, password resets, appointment scheduling, shipping updates, returns, payment reminders.
• Design smart transfers: detect frustration, complexity, or high risk and route to a human with full context.
• Ground the model: retrieve answers from approved knowledge (product docs, policy pages, support runbooks). Keep a strict fallback when confidence is low.
• Protect data: auto-redact PII, set retention rules, and restrict training on sensitive content.
• Measure what matters: CSAT, FCR, containment with satisfaction, average handle time including transfer, and resolution quality.
• Continuously review: sample transcripts, monitor hallucination rate, and update guardrails weekly.

Industry notes you can use

Healthcare

• Leaders report: 54% automate patient communications via chatbots, 52% use predictive communication, 51% analyze patient data.
• Opportunity gap: only 32% use AI for scheduling, yet 57% of AI-comfortable patients want it.
• Top concerns: data privacy/security 55%, accuracy 40%, regulatory compliance 39%.
• Consumer trust is mixed: 35% would use a provider's AI chatbot; 40% wouldn't; 25% are unsure. Accuracy (64%), privacy (40%), and "too impersonal" (43%) drive hesitation.

Action: lead with utility. Offer scheduling, pre-visit checklists, prescription refills, non-emergency FAQs, and post-visit follow-ups. Make a human path obvious. Add empathetic language and confirm understanding before giving next steps.

Trust lift: willingness rises above 40% when faster care is on the table. Use that value proposition, then deliver on it.

Financial services

• Leaders report: 53% use AI chatbots for support, 53% analyze client data, 49% provide basic financial advice.
• Concerns: data/security 41%, consumer perception 37%, accuracy 35%.
• FinServ is ahead on chatbots: 59% already use them (vs. 52% average).
• Consumers are cautious with advice: 43% wouldn't engage, 21% unsure, 36% willing. Another study found two-thirds (66%) of Americans have used AI for financial advice, jumping to 82% among Gen Z and Millennials.

Action: focus AI on balances, transactions, card controls, dispute status, and education. Keep advice basic with clear disclaimers. Route portfolio or lending scenarios to licensed humans.

Governance: adopt a risk framework and model oversight. The NIST AI Risk Management Framework is a solid starting point.

Retail and e-commerce

• Top use cases: shipping/delivery notifications 48%, personalized offers 45%, automated CS with chatbots 45%.
• Concerns: data privacy 48%, accuracy 44%, customer trust 37%.
• Comfort zones: 52% of consumers will use AI for order tracking and delivery; Gen Z 67%, Millennials 63%.
• Returns via messaging chatbots: 76% are willing to try.
• Tension: 40% feel uneasy sharing preferences with AI, yet 70%+ want recommendations that actually make sense.

Action: personalize with zero- and first-party data. Recommend based on behavior, cart, and purchase history. Cap frequency, explain "why you're seeing this," and make opting out painless.

Trust moment: be proactive with delays, backorders, and substitutions. Clear updates prevent support tickets and earn loyalty.

Agentic AI is next

Agentic AI doesn't just reply; it acts. It schedules, amends orders, follows up on unresolved cases, and keeps customers updated across channels. That autonomy requires clear rules, permissions, and transparent logs.

As one Sinch leader put it, the shift is from cutting costs to creating context. You're greeted with history, preferences, and next best actions. That's useful - if you can explain how and why it's happening.

Readiness checklist for agentic AI

• Define allowed actions: what the agent can read, write, update, and trigger.
• Set permissions and approvals for high-risk moves (refunds, account changes, PHI/PII).
• Connect systems: CRM, order management, support platform, identity, and messaging.
• Enable identity verification for sensitive workflows.
• Log everything: prompts, retrieved data, decisions, actions, and outcomes.
• Add safety rails: confidence thresholds, sandbox testing, auto-handoff on uncertainty.
• Measure: satisfaction after agent actions, error rate, reversals, and impact on backlog.

KPI stack to prove trust

• Containment with satisfaction (not just deflection).
• Time to first response and time to resolution, including transfers.
• Escalation quality: context passed, re-explanations avoided.
• Disclosure acknowledgment rate and opt-outs.
• Accuracy and hallucination rate from transcript review.
• Privacy/security incidents and compliance exceptions.
• Revenue impact: conversion assist, save rate, repeat purchase, churn.

Quick wins you can ship this quarter

• Add "talk to a human" at every step and route intelligently.
• Ground answers in your support docs with retrieval; update weekly.
• Launch AI for order status, returns, appointment scheduling, and basic billing.
• Proactively message shipping delays and appointment reminders.
• Audit data flows and enable PII redaction by default.
• Stand up a transcript QA loop with weekly tuning.

Level up your team

Your stack is only as strong as the people running it. Get your support org fluent in prompts, routing logic, evaluation, and safety.

For practical upskilling, explore courses by role at Complete AI Training. Keep an eye on new programs here: Latest AI Courses.

Bottom line: earn trust with clarity and control. Pick the right jobs for AI, keep humans close for the hard parts, and measure everything. Do that, and customers won't just accept AI - they'll prefer it.