Every AI vendor has a number. Seventy per cent is popular. Some go with sixty. Others push eighty. It depends on who is doing the pitching.
The number is not dishonest. But it is incomplete. How much of your customer service AI can actually handle depends on what your customers ask about, and that is a question most organisations do not have good data on until they start measuring.
AI automation in customer service means an AI system handles enquiries end-to-end, from first contact to resolution, without a human needing to intervene. How much of that is realistic depends on the type of questions your customers are actually asking.
What kinds of enquiries can AI actually handle?
Simple, repetitive enquiries are where AI performs best. If the answer to a question is always the same regardless of who asks, AI can deliver that answer as well as a human agent. And faster.
Enquiry types AI handles well:
- Opening hours, pricing and contact information
- Order status and delivery updates
- Account details and invoice enquiries
- Technical support for known issues
- Cancellations and straightforward rescheduling
Enquiries that require human judgement:
- Complaints with strong emotional charge
- Complex cases spanning multiple systems
- Exceptions requiring business judgement
- Negotiations and retention conversations
- Cases with legal or regulatory implications
Why does automation rate vary so much between companies?
Case distribution varies significantly across industries and business models. A telecoms company with straightforward support queries can automate a very different proportion from a B2B software company where every case is unique.
The deciding factor is not the AI system's capability. It is your case mix. And most companies do not know the answer until they start measuring it.
How do you calculate your automation potential?
A structured review of case data gives a clear picture quickly.
Step one: map your case mix. Look at the last 500 to 1,000 cases and categorise them. What proportion are recurring standard questions? That is your baseline.
Step two: identify the ten most common case types. In most organisations, the ten most common types account for 60 to 70 per cent of total volume. If AI can handle eight of them, the foundation for high automation is in place.
Step three: start with the lowest-hanging fruit. Open with opening hours, order status, and invoice enquiries. Measure deflection rate, meaning the proportion of cases resolved without escalation to a human. That is your real metric, not the vendor's percentage.
lynes' AI-supportagent and AI-telefonist are built to handle repetitive and rule-based cases efficiently, with built-in logic for escalation when a case requires it. A well-planned deployment typically delivers 30 to 50 per cent deflection within the first 90 days.
What do real figures look like?
Among lynes customers who have deployed the AI-telefonist, we typically see 30 to 60 per cent of inbound calls handled without a human agent needing to step in, depending on case type and industry. For written channels via the AI-supportagent, the picture is similar.
Automation rate is not a fixed number. It is a function of how well the AI is configured, how clearly escalation rules are defined, and how similar the cases are to each other. The number improves over time if you actively work with it.
What should never be fully automated?
There are cases where an AI response can be technically correct but humanly wrong. A customer who calls upset and wants an explanation needs to be met by someone who can genuinely show understanding. AI can identify that a case requires human handling and transfer it, but the initial reading of tone and emotional weight is critical.
lynes AI systems are designed to read context and escalate proactively when a case is complex or emotionally charged. That is just as important as the automation rate itself.
Curious about the difference between an AI chatbot and an AI agent in this context? We have covered it in detail in AI Chatbot vs AI Agent: Which One Does Your Customer Service Need?
Frequently asked questions
How quickly can you reach 50 per cent automation?
With clear case mapping and a well-configured AI agent, 50 per cent deflection is achievable within 90 days for organisations with a predictable case mix. Organisations with more complex cases may need 6 to 12 months of continuous configuration.
What happens to cases AI cannot handle?
AI systems with well-defined escalation rules automatically transfer to a human agent, with full context from the previous interaction. The customer does not have to start over, and the agent sees immediately what has happened.
Is automation worthwhile for a small business?
Yes, if the case volume justifies the setup cost. Many SMB customers using lynes deploy the AI-telefonist primarily to cover out-of-hours enquiries, which delivers meaningful value without requiring high overall volume.
Do chatbots and AI agents count the same in automation statistics?
No. A chatbot that tells the customer to call instead resolves nothing and should not count. What counts is when a case is resolved to the customer's satisfaction without human involvement. The distinction is explained in AI Chatbot vs AI Agent.
TL;DR
Seventy per cent automation is achievable for organisations with a straightforward case mix. Realistic first steps are 30 to 50 per cent deflection within 90 days. The key is case mapping, well-defined escalation rules, and continuous configuration.
Want to know what is realistic for your organisation? Book a call and we will work it out in ten minutes.
