AI in customer service: which tasks to automate first
Updated
Across the 5.2 million US jobs in customer service, the work scores 66 of 100 for AI exposure, weighted by how many people do each job. AI can take on about 1,126 hours a year per full-time person, 56% of a working year. The task that frees the most hours across the team is to keep records of customer interactions or transactions, about 186 hours a year for each person who does it.
- AI exposure, weighted by employment
- 66 out of 100 High exposure
- Hours AI can take on
- 1,126 hours a year per full-time person
- Task time AI can do today
- 51% AI speeds up another 40%
Tasks to automate first
The 15 tasks where AI can take on the most hours across customer service: how exposed each task is, how much of the job it takes and how many people do that job. Hours are per person in the role that does the task. How we score each task.
| Task and who does it | AI exposure |
|---|---|
| Keep records of customer interactions or transactions, recording details of inquiries, complaints, or comments, as well as actions taken. AI transcribes and summarizes interactions into case records automatically. Done by customer service representatives. | 85% AI can do it 186 hours a year for each person in the role |
| Complete contract forms, prepare change of address records, or issue service discontinuance orders, using computers. Structured form filling and address changes suit AI well. Done by customer service representatives. | 85% AI can do it 168 hours a year for each person in the role |
| Determine charges for services requested, collect deposits or payments, or arrange for billing. Billing systems calculate charges and process payments with little human input. Done by customer service representatives. | 70% AI can do it 141 hours a year for each person in the role |
| Schedule appointments and maintain and update appointment calendars. Scheduling bots handle booking and calendar updates routinely. Done by receptionists. | 75% AI can do it 112 hours a year for each person in the role |
| Operate telephone switchboard to answer, screen, or forward calls, providing information, taking messages, or scheduling appointments. AI voice agents can answer, screen, route calls and take messages. Done by receptionists. | 65% AI can do it 105 hours a year for each person in the role |
| Perform administrative support tasks, such as proofreading, transcribing handwritten information, or operating calculators or computers to work with pay records, invoices, balance sheets, or other documents. Proofreading, transcription and invoice processing are highly automatable. Done by receptionists. | 75% AI can do it 101 hours a year for each person in the role |
| Post charges, such as those for rooms, food, liquor, or telephone calls, to ledgers, manually or by using computers. Charges post automatically from point of sale systems; minimal human input. Done by hotel desk clerks. | 75% AI can do it 81 hours a year for each person in the role |
| Make and confirm reservations. Booking engines and email confirmations already handle most reservations. Done by hotel desk clerks. | 72% AI can do it 78 hours a year for each person in the role |
| Enter customers' transactions into computers to record transactions and issue computer-generated receipts. Routine digital transaction entry, largely handled by software. Done by tellers. | 68% AI can do it 62 hours a year for each person in the role |
| Keep records of room availability and guests' accounts, manually or using computers. Property systems already store availability and guest accounts digitally. Done by hotel desk clerks. | 70% AI can do it 76 hours a year for each person in the role |
| Identify transaction mistakes when debits and credits do not balance. Locating debit and credit imbalances is pattern work software does well. Done by tellers. | 65% AI can do it 57 hours a year for each person in the role |
| Prepare and verify cashier's checks. System-generated cashier's checks with verification a person reviews. Done by tellers. | 62% AI can do it 56 hours a year for each person in the role |
| Answer applicants' questions about benefits and claim procedures. Routine questions about benefits and claim steps suit chatbots. Done by eligibility interviewers. | 75% AI can do it 116 hours a year for each person in the role |
| Keep records of assigned cases, and prepare required reports. Case records and reports are heavily digital and templated. Done by eligibility interviewers. | 70% AI can do it 115 hours a year for each person in the role |
| Compile, record, and code results or data from interview or survey, using computer or specified form. Coding and recording responses is highly structured digital work. Done by interviewers. | 80% AI can do it 118 hours a year for each person in the role |
What stays with people
9% of the task time across these roles needs a person, and the output of the rest still needs someone to check it. The tasks below take the most of that time.
- 51% AI can do it
- 40% AI helps
- 9% Needs a person
- Greet persons entering establishment, determine nature and purpose of visit, and direct or escort them to specific destinations. Receptionists, 15% exposed
- Receive payment and record receipts for services. Receptionists, 22% exposed
- Monitor bank vaults to ensure cash balances are correct. Tellers, 15% exposed
- Count currency, coins, and checks received, by hand or using currency-counting machine, to prepare them for deposit or shipment to branch banks or the Federal Reserve Bank. Tellers, 15% exposed
- Clean and maintain lobby and common areas, such as restocking supplies and watering plants. Hotel desk clerks, 5% exposed
Adoption path
Start with the tasks where AI helps a person, then automate the routine ones, then give whole workflows to agents with a person approving the output.
Roles in customer service
16 roles, each linked to its task-by-task score. Of the roles with at least 2% of the jobs, customer service representatives score highest at 74, event planners lowest at 46.
- Customer service representatives, 49.6% of the team, score 74 out of 100
- Receptionists, 17.4% of the team, score 62 out of 100
- Tellers, 6.3% of the team, score 53 out of 100
- Hotel desk clerks, 5% of the team, score 54 out of 100
- Court clerks, 3.4% of the team, score 62 out of 100
- Eligibility interviewers, 3% of the team, score 65 out of 100
- Interviewers, 2.8% of the team, score 61 out of 100
- Event planners, 2.7% of the team, score 46 out of 100
- Reservation agents, 2.3% of the team, score 60 out of 100
- Public safety telecommunicators, 2% of the team, score 57 out of 100
- Library assistants, 1.6% of the team, score 49 out of 100
- Order clerks, 1.4% of the team, score 68 out of 100
- Postal service clerks, 1.4% of the team, score 38 out of 100
- Switchboard operators, 0.7% of the team, score 58 out of 100
- Gambling cage workers, 0.3% of the team, score 40 out of 100
- Telephone operators, 0.1% of the team, score 58 out of 100
Questions about AI in customer service
Which tasks should a customer service team automate first?
The first task to automate is to keep records of customer interactions or transactions (85% exposed). Next come the tasks to complete contract forms (85%) and to determine charges for services requested (70%). Together they free the most hours across the team.
How many hours can AI save a customer service team?
About 1,126 hours a year per full-time person, 56% of a 2,000-hour working year. For a team of 10 that is about 11,260 hours, the time of 5.6 full-time people. The figure is the share of task time AI can remove today, weighted by how many people do each job.
Which roles in a customer service team are most exposed to AI?
Of the roles with at least 2% of the jobs, customer service representatives score highest, at 74 of 100, and event planners lowest, at 46. Each role's page lists its tasks one by one.
Will AI replace jobs in customer service?
AI can do 51% of the task time across these roles today. Another 40% gets faster with AI while a person stays in charge. 9% needs a person. One task that stays with people is to greet persons entering establishment.
Where should a customer service team start with AI?
With the quick wins: approved assistants for the tasks AI helps with, under a written AI policy. Then automate the 15 tasks on this page AI can do end to end. An AI audit maps the tasks in your team and ranks them the same way this page does.
Next steps
- Automated resolution rate: measuring support AI by outcomes
- Multi-brand service desk automation: one desk for several customers
- Case study: support triage with replies ready for review
Find the tasks to automate first in your team
We map the tasks in your customer service team, score each one the way this page does and build the automations that pay back first.