Contact center automation: from IVR menus to AI agents
Gartner expects AI agents to resolve 80% of common customer service issues by 2029 and cut operating costs 30%. Its own survey shows the cost side is more complicated than the forecast.

Contact center automation has gone through three stages. Interactive voice response (IVR) menus routed callers by keypad. Chatbots answered simple questions from scripts. AI agents, built on large language models, now understand a request in plain language, look up the customer's account, take an action and confirm the result.
Gartner's March 2025 forecast is the number most vendors quote: agentic AI will, by 2029, resolve 80% of common service issues with no human involved, and cut operating costs by 30%. Gartner describes agentic AI as software that finishes tasks by itself, for example cancelling a membership, which separates it from earlier tools that only wrote text or summarized calls.
What the forecast leaves out
The forecast is a prediction. Gartner's own survey of 321 customer service leaders, run in October 2025, describes the present: 91% said they are under pressure from executive leadership to implement AI. Only 20% had reduced agent headcount because of AI. Gartner also predicts that by 2028 more than half of customer service teams will spend twice as much on technology while keeping their talent, and says most organizations understate the talent AI needs.
For a center with $5 million in annual operating costs, a 30% cut is $1.5 million. Treat that as the ceiling of a forecast. The number to plan against is smaller if technology spending doubles first and agents are retrained instead of removed. Nearly 80% of surveyed organizations plan to shift some agents into new roles, and 58% plan to train agents as knowledge management specialists.
What changes with AI agents
| Scripted chatbot | AI agent | |
|---|---|---|
| Understanding | Keywords and fixed flows | Free-form language, voice or text |
| Actions | Answers from a knowledge base | Checks orders, changes bookings, issues refunds through connected systems |
| Handoff | Starts over with a human | Passes a summary and context to the human agent |
| Failure mode | "I didn't understand that" | Confident wrong answers if not grounded and tested |
Three fixes before any pilot
The first is the knowledge base. Gartner's recommendation is to invest in knowledge management and notes that many organizations hold backlogs of outdated articles with inconsistent review. An AI agent answers from that material, so stale articles become stale answers.
The second is integration. An agent can only resolve what it can reach. If order, billing and account data sit in systems without APIs, it can answer questions but not fix problems. Most of the effort goes into connecting those systems, setting limits on what the agent can do, and testing it on real conversations.
The third is consent for voice. In February 2024 the FCC confirmed that the consent rules for artificial or prerecorded voices under the Telephone Consumer Protection Act cover AI-generated voices, including cloned ones. Calls to cellular lines that are not marketing need prior express consent, and marketing calls need prior express written consent. The FCC stopped short of banning the technology, but an outbound AI voice campaign needs legal review first.
What to automate first
Start with high-volume requests that follow clear rules: order status, password resets, appointment changes, address updates and simple billing questions. Keep humans for complaints, retention-sensitive cancellations, vulnerable customers and anything that depends on judgment. Agent-assist tools, which transcribe calls, suggest answers and write the after-call summary, are a lower-risk first project because customers never talk to a bot. For the platform choices behind this, see the contact center tech refresh.
What IT teams should do
- Rank contact reasons by volume and pick three simple ones to automate first.
- Audit the knowledge base: retire outdated articles and name an owner for each topic.
- List the back-end systems an AI agent would need and whether each has an API.
- Get legal sign-off on consent before any outbound AI voice use.
- Budget for retraining staff alongside the technology, not instead of it.
Sources
- Gartner: Agentic AI will autonomously resolve 80% of common customer service issues by 2029 (March 5, 2025)
- Gartner: 91% of customer service leaders under pressure to implement AI in 2026 (February 18, 2026)
- Gartner: Customer service leaders must blend human strengths with AI in 2026 (December 17, 2025)
- MacTech: Gartner says over 50% of customer service organizations will double technology spending by 2028
- Orrick: Did the FCC really ban AI-generated voice calls?