AI for Customer Service: What to Automate and What It Costs

AI for customer service works best on repetitive questions with a clear right answer: order status, account changes, policy lookups and first line troubleshooting. Start there, connect the AI to your real systems and help content, give every conversation an easy path to a human, and pay by resolved conversation rather than by seat. Keep complaints, refunds with judgment and anything legally sensitive with people.

If you want ideas like these working inside your own business, see our AI development and automation service.

What is AI for customer service, in plain terms?

AI for customer service means software that reads a customer's message, works out what they want, and either answers it, takes an action, or hands it to a person with the context already filled in. Modern systems use a large language model (the same kind of model behind ChatGPT) plus your own data: help articles, order records, account details and policies.

There are three common shapes:

• AI agent in front of the queue. The AI answers customers directly on your website chat, email or WhatsApp, and escalates what it cannot solve. This is what most people mean by an AI chatbot for a website.

• Copilot behind the agent. The AI drafts replies, summarises long threads and suggests the next step, while a human sends the final message.

• Automation in the back office. The AI tags and routes support tickets, detects urgency, and fills in forms, without talking to the customer at all.

The first saves the most money; the copilot is the safest place to start. If you want to understand how the AI looks things up in your documents, our plain English guide to RAG for business leaders explains the retrieval step that makes answers accurate.

Does AI in customer support actually work?

Yes, AI in customer support works, and a field study of 5,179 support agents measured by how much. That study, Generative AI at Work by Erik Brynjolfsson, Danielle Li and Lindsey Raymond, tracked those agents as a company rolled out an AI assistant. Access to the tool raised productivity, measured as issues resolved per hour, by 14% on average.

The gain was 34% for novice and low skilled agents and minimal for the most experienced ones. The authors suggest the AI spreads the habits of the best agents to everyone else. The study also found better customer sentiment and higher employee retention.

So a copilot pays off fastest where staff turnover is high, and since the study measured AI helping humans, treat claims about fully autonomous support with more care.

Gartner is bullish on the autonomous side. In a March 2025 prediction reported by CX Today, Gartner said agentic AI will autonomously resolve 80% of common customer service issues without human intervention by 2029, leading to a 30% reduction in operational costs. Note the word "common": it is a forecast about routine issues, not a result today.

Do customers want AI customer service?

Many customers do not want AI customer service, and that is the main risk to plan around. A Gartner survey of 5,728 customers, run in December 2023 and reported by DIGIT in July 2024, found 64% would prefer companies not use AI in customer service, and 53% would consider switching to a competitor if they learned a company was going to.

The concerns are specific, which makes them fixable:

• 60% worried it would get harder to reach a person

• 46% worried about AI displacing jobs

• 42% worried about AI giving wrong answers

• 34% worried their data would be less secure

Keith McIntosh, a senior principal at Gartner, said in remarks quoted by DIGIT that AI chatbots must tell customers they will connect them to an agent when necessary, and the chat should turn into an agent chat that picks up where the bot left off. In practice that means two rules. Never hide the human option, and never make the customer repeat themselves after a handoff.

What should you automate first with AI customer support?

With AI customer support, automate first the questions that are frequent, have one correct answer, and can be checked against a system of record. Pull your last 90 days of conversations, group them by intent, and score each group on three questions:

1. Volume. Is this in your top ten reasons for contact?

2. Determinism. Is there one right answer that lives in a system you can connect, such as an order database, a booking system or a policy page?

3. Risk. If the AI gets it wrong, is the damage small and reversible?

Groups that score high on all three are your first wave. For most businesses that list looks like this:

• Where is my order, delivery dates and tracking links

• Password resets, login help and account detail changes

• Opening hours, pricing, plan differences and policy questions answered from your help centre

• Booking, rescheduling and cancelling appointments

• Simple troubleshooting that follows a fixed checklist

• Collecting details before a human takes over, such as order number, photos and a description

Keep these with people, at least at first: complaints from upset customers, refunds or exceptions that need judgment, anything involving health, money or legal rights, and your highest value accounts. The AI can still help there by drafting and summarising, but a human should send the reply.

How much does AI for customer service cost?

AI for customer service costs about $0.99 per resolved conversation on outcome priced tools such as Intercom's Fin, plus seat fees on some plans and the work to connect your systems. As of November 2026, the Fin pricing page lists $0.99 per outcome, so your bill tracks how much work the AI finishes, with a minimum of 50 outcomes a month when you use Fin with your existing helpdesk, or $19 per helpdesk seat a month on top when you use Intercom's own helpdesk.

Read the definition twice. Fin counts a resolution when no further help is requested after its last answer, charges at most one outcome per conversation, and charges $9.99 for a lead qualification. "No further help requested" is not the same as "problem solved", so measure customer satisfaction on billed resolutions, not just the count.

Here is a hypothetical worked example to show the shape of the maths. A business gets 3,000 support conversations a month. After the first wave of automation, the AI closes 40% of them, which is 1,200 outcomes at $0.99, about $1,190 a month. If each of those conversations used to take a human six minutes at a fully loaded cost of $30 an hour, that work was worth $3,600 a month. The saving only holds if those customers were actually helped; if many come back through another channel, the numbers flip.

Also budget for integration work, help content fixes, and someone who owns the AI's quality week to week. Our guide to build vs buy for custom AI covers when an off the shelf tool is enough and when a custom build pays back.

What does a WhatsApp chatbot for business cost?

A WhatsApp chatbot for business costs whatever the AI tool charges plus Meta's per message fees: replies inside the 24 hour service window are free for the first 1,000 a month per number, and templates are billed by category and country. According to Meta's WhatsApp Business Platform pricing documentation, Meta has charged per delivered message since 1 July 2025, and only for messages from businesses to users. Messages customers send you are free.

The key concept is the customer service window. When a customer messages you, a 24 hour window opens and resets every time they message again. Inside that window you can send normal free form replies, which Meta calls service messages. Outside it you can only send approved templates, which are billed by category (marketing, utility or authentication) and by the recipient's country code.

Pricing changed again on 1 October 2026. Service messages, which had been free since November 2024, are now free only for the first 1,000 each month per business phone number, and charged after that. Utility templates sent inside an open window are no longer free either, except inside the free entry point windows that follow a Click to WhatsApp ad. If you planned a WhatsApp support bot on older pricing assumptions, rerun the numbers. Rates can change on the first day of each quarter, so check the official page before you commit to a budget.

Who is liable when an AI chatbot gets it wrong?

The business that runs the chatbot is liable for what it says, at least on the clearest ruling so far. In a case brought by Jake Moffatt against Air Canada, decided on 14 February 2024 by British Columbia's Civil Resolution Tribunal, the airline's website chatbot told a grieving customer he could apply for a bereavement fare after travelling.

As reported by Daily Hive, Air Canada argued it could not be held liable for information from its agents or representatives, including a chatbot. The tribunal called that "a remarkable submission" and said the chatbot "is still just a part of Air Canada's website". Air Canada was ordered to pay CA$812.02 in damages, interest and fees.

The sum was small, but the principle applies to any business: your AI's answers carry the same weight as your website. So:

• Ground every policy answer in the current policy text, and link to it.

• Block the AI from inventing exceptions, discounts or promises it cannot check.

• Log every conversation so you can see what was said and fix it fast.

• Review a sample of resolved conversations every week, not just the escalated ones.

How do you roll out AI customer support without hurting the experience?

Roll out AI customer support in stages, and let real conversations decide when to widen its scope. A sequence that works:

1. Clean your help content first. The AI repeats whatever your help centre says, outdated parts included.

2. Start as a copilot. Let the AI draft replies for human agents for two to four weeks. Every edit an agent makes shows you where the AI is weak.

3. Turn on autonomous answers for one or two intents. Pick the safest high volume questions from your scoring exercise.

4. Make the human handoff obvious. A visible "talk to a person" option, and a handoff that carries the full conversation and the details already collected.

5. Measure the right things. Track resolution rate, repeat contact within seven days, customer satisfaction on AI handled conversations, and escalation reasons. A rising resolution rate with rising repeat contacts means the AI is closing conversations, not solving problems.

6. Expand one intent at a time. Add actions such as rescheduling or address changes only once the read only answers are reliable.

The same approach works for phone support with a voice AI agent. For connecting support tools to the rest of your stack without custom code, our comparison of n8n vs Zapier for workflow automation compares the two main no code options.

Key takeaways

• Automate frequent, factual questions that can be checked against a real system; keep judgment calls human.

• The strongest evidence (5,179 agents, 14% more issues resolved per hour) is for AI helping people, with the biggest gains for newer staff.

• 64% of customers in Gartner's survey would prefer companies not use AI in service; their top concern (60%) was losing access to a person.

• Per outcome pricing ties cost to work done, and WhatsApp service replies are free only for the first 1,000 a month per number since 1 October 2026.

• You are responsible for what your chatbot says.

Frequently asked questions

Can AI fully replace a customer service team?

Not today for most businesses. AI can handle a large share of routine questions, and Gartner predicts agentic AI will resolve 80% of common issues autonomously by 2029, but complaints, exceptions and sensitive cases still need people. Plan for a smaller team doing harder work, not no team.

What is the difference between an AI chatbot and an AI agent in customer service?

An AI chatbot mainly answers questions. An AI agent can also take actions in your systems, such as changing a booking or issuing a return label, so it needs tighter permissions and testing.

Is an AI chatbot for a website worth it for a small business?

An AI chatbot for a website is usually worth it once you get enough repeat questions to keep a person busy for part of every day. Below that, a good FAQ page and fast email replies may serve customers better. Per outcome pricing helps small teams because a quiet month costs little.

Should customers be told they are talking to AI?

Yes. Hiding the AI backfires when customers find out. Say clearly that it is an AI assistant, and show how to reach a human at any point.

AISmith builds and integrates AI agents for customer service on web chat, WhatsApp and voice. To see how AI search engines such as ChatGPT and Perplexity describe your business, run our free GEO checker.