What AI Can and Can’t Do Today

Few topics in customer service generate more excitement, or more confusion, than artificial intelligence. Some vendors suggest AI can handle nearly every customer conversation. Some headlines suggest contact center jobs are about to disappear. Neither is the full picture.

This guide gives you a practical view of what today’s AI does well, where it still falls short, and how to make good decisions about using it.

What we mean by “AI”

In contact centers, “AI” usually refers to software that can:

• Understand everyday language, whether typed or spoken, instead of requiring customers to press buttons or use exact keywords
• Generate responses, writing answers, summaries, or suggestions in natural language
• Find patterns across thousands of conversations that a person could never review alone

Much of the recent progress comes from large language models, the technology behind modern AI assistants. They’re far more capable than the scripted chatbots of a few years ago, but they still have real limits.

Where AI does well

Answering common, predictable questions. Order status, store hours, password resets, appointment changes, and balance inquiries follow a clear pattern. AI can handle many of these quickly, at any hour, with no wait.

Summarizing conversations. After a call or chat, AI can write a summary for the customer’s record in seconds. This cuts after-call work and makes notes more consistent.

Helping agents find answers. During a conversation, AI can search your knowledge base and suggest the right article or response, so agents spend less time hunting for information.

Routing contacts to the right place. AI can understand why a customer is calling, from what they say or type, and send them to the right team without a long phone menu.

Reviewing large numbers of conversations. AI can scan every call and chat for trends, such as rising complaints about a product, or for compliance issues like missed disclosures. People can then focus their attention where it’s needed.

Translating. AI can help agents and customers communicate across languages, though accuracy should be checked for important or sensitive conversations.

Where AI still falls short

Emotional and sensitive situations. Upset, grieving, or frightened customers need empathy and judgment. AI can sound polite, but customers often notice when understanding is missing, and a poor response can make a bad situation worse.

Unusual or complex problems. AI works best on situations it has seen many times. Problems that require investigation, creativity, or working across several departments still need a person.

Decisions that need judgment. Making an exception to a policy, deciding how to make things right for a loyal customer, or recognizing that something “doesn’t add up” are human strengths.

Getting facts right every time. Language-based AI can produce answers that sound confident but are wrong, sometimes called “hallucinations.” This risk is lower when AI is limited to your own approved information, but it never disappears completely. Wrong answers about billing, policies, health, or legal matters can be costly.

Working with messy information. AI is only as good as the information behind it. If your knowledge base is outdated, incomplete, or contradictory, AI will repeat those problems, faster and at a larger scale.

A simple way to decide where AI fits

Contact type	AI on its own	AI helping an agent	Person only
Simple, frequent questions (hours, order status)	✓ Good fit
Routine tasks with clear rules (rescheduling, address changes)	✓ Good fit, with a path to a person
Product or technical questions	Sometimes	✓ Good fit
Complaints and upset customers		✓ Helpful	✓ Often best
Sensitive matters (health, finances, bereavement)		Limited	✓ Best fit
Exceptions, disputes, and judgment calls			✓ Best fit

Questions to ask before you buy

1. What specific problem are we solving? “Use AI” isn’t a goal. “Answer order-status questions without a wait” is.
2. How will we measure success? Pick measures in advance, such as resolution rate, customer satisfaction, and how often customers ask for a person.
3. Where does the AI get its information? Can it be limited to your own approved content? Who keeps that content up to date?
4. How do customers reach a person? There should always be a clear, easy path.
5. How are mistakes caught? Who reviews AI conversations, and how often?
6. What happens to customer data? Where is it stored, who can access it, and is it used to train the vendor’s systems?
7. What does it really cost? Ask about setup, ongoing fees, usage charges, and the staff time needed to maintain it.

Rules and disclosure

Laws around AI in customer service are developing quickly. Some rules already apply, such as limits on automated and artificial-voice phone calls in the United States, and some states and countries require businesses to tell customers when they’re talking with AI. Being upfront is good practice in any case: customers tend to trust businesses more when they know who, or what, they’re dealing with. Check current requirements for your industry and location before launching, and see our Compliance Basics section.

The bottom line

AI is a powerful tool for handling routine work and helping agents do their jobs better. It isn’t a replacement for well-trained people, good processes, and accurate information. The contact centers getting the most from AI are the ones that start with a clear problem, keep humans involved, and measure the results honestly.

What to read next

• Chatbots, voice bots, and agent assist explained: the main types of contact center AI
• In-house, outsource, or AI?: how AI fits alongside your other options
• Knowledge bases and call scripts: why accurate information comes first