What customers actually want from AI: price monitoring and 24/7 support
Customers are open to AI in retail and service, but they want it to be practical. A new survey from the IBM Institute for Business Value and the National Retail Federation points to two clear priorities: price monitoring and always-on customer support.
For support leaders, this is a nudge to build AI that solves day-to-day problems reliably, not theatrics. Trust is still thin, so execution matters more than hype.
Key findings from 18,000 consumers
- Top request: a deal-hunting agent that tracks prices across brands, applies discounts and loyalty rewards, and alerts shoppers at the best time to buy.
- Next priority: a 24/7 customer service agent that handles inquiries, resolves issues, and provides personalized help across all touchpoints.
- Adoption today: 45% have used AI for help, 41% for product research, and about one-third to find reviews.
- Trust gap: only 24% trust AI recommendations outright. Another 17% validate with social content, and 22% cross-reference sources when researching products.
Why this matters to customer support leaders
Customers are willing to use AI if it helps them make smarter decisions and get faster resolution. They reward consistency and clarity, not gimmicks.
- Start with a 24/7 agent that can answer, resolve, and escalate-cleanly.
- Make the agent proactive: notify customers about delays, outages, refunds, and fixes before they ask.
- Personalize with guardrails: use account context and order history, but explain how the answer was formed and offer sources when possible.
- Offer instant handoff to a human with full context when confidence drops or emotions rise.
Trust is the bottleneck
Consumers will trust AI when it proves reliable and transparent. Dee Waddell of IBM Consulting puts it simply: when AI uses accurate, up-to-date product data to surface options that match style, budget, and availability, it stops feeling experimental and starts feeling indispensable.
That usefulness only holds if recommendations are truthful, consistent, and free of hallucinations. If customers sense manipulation or guesswork, trust disappears fast.
Data foundations that make AI dependable
- Source-of-truth data: one place for product, policy, pricing, and order data that the agent can rely on.
- Shared standards: keep systems in sync (catalog, orders, logistics, CRM) so answers don't conflict by channel.
- Clear governance: define what the agent can and cannot say, when to cite sources, and when to escalate.
- Real-time signals: inventory, shipping delays, promotions, and known incidents should flow into the agent instantly.
- Feedback loops: capture corrections from agents and customers, then retrain or fine-tune on verified updates.
Blueprint for a 24/7 AI support agent
- Scope the top intents: order status, returns and exchanges, warranty, delivery issues, account access, payment problems, and product questions.
- Ground answers: connect the agent to product data, policies, and order systems via APIs-no freeform guessing.
- Set confidence thresholds: below a threshold, summarize context and transfer to a human instantly.
- Enable proactive service: push alerts for delays, partial shipments, back-in-stock, or price adjustments with clear next steps.
- Keep receipts: show reasoning summaries or source links so customers can verify claims without extra effort.
- Omnichannel consistency: deploy the same logic across chat, email, voice, and social, with identity and history intact.
How to reduce hallucinations and confusion
- Use retrieval-augmented generation with strict grounding to approved content.
- Limit the agent's scope to documented policies and current data. If it's not in the system, it shouldn't answer.
- Maintain a banned claims list (e.g., medical, legal, price guarantees) and force escalation.
- Run red-team tests for known tricky scenarios, then add pattern-based safeguards.
What to measure (so confidence grows)
- First contact resolution (AI + human) and resolution time by intent.
- Containment rate with satisfaction threshold (not containment at the expense of CX).
- CSAT/NPS after AI interactions vs. human only.
- Deflection quality: how many AI-resolved cases avoid repeat contact within 7 days.
- Truthfulness signals: number of corrections, escalations due to low confidence, and confirmed misinformation incidents.
- Proactive impact: percentage of issues prevented or auto-resolved through alerts or credits.
For deeper context
The survey was fielded by the IBM Institute for Business Value and the National Retail Federation. For broader research and guidance, see their resources:
Build skills your team can use this quarter
If you're standing up an AI support agent or cleaning data pipelines, targeted training shortens the learning curve. Explore role-focused options here:
The takeaway: customers want AI that saves them money and solves their problems any time, without guesswork. Build for accuracy, transparency, and proactive service, and trust will follow.
Your membership also unlocks:
