The best way to learn how someone will chat with your customers is to watch them chat with a customer. A résumé tells you where a person worked. A short test tells you how they write, what they do when they are stuck, and whether they stay kind when a customer is not.
Below are five tests. Each one takes a few minutes. Give all five to every applicant, in the same way, so you compare people on the same work.
Before you start
A test is only as good as its setup. Do these four things first.
- Write each test as a real customer message. Use your own products, your own policies and the way your customers really write, typos included.
- Give a short policy sheet. Half a page with the rules a new agent would have on day one: what they can offer, what they cannot, and who they pass things to.
- Set a time limit that matches the job. If your customers expect quick replies, time the test too.
- Decide what a good answer looks like before you read any. Otherwise the first strong answer becomes your standard, and you start grading people against each other instead of against the job.
Test 1: The refund they cannot promise
The setup. A customer wants something the agent is not allowed to give. For example, a refund needs a manager's approval, or the order is past your return window.
I ordered this three weeks ago and it already stopped working. I want my money back today.
What you are testing. Can they hold a rule without sounding cold, and do they give the customer a real next step?
A good answer:
- Says sorry for the trouble in one line, not three.
- Does not promise the refund.
- Says what happens next, and when: "I've passed this to our returns team. They make refund decisions, and they'll email you."
- Asks for what the team will need, like the order number or a photo.
A weak answer:
- "No problem, I'll refund you now." This is a promise they cannot keep. It turns one unhappy customer into an angrier one later.
- "Sorry, that's our policy," with nothing after it.
- A long paragraph that hides the answer at the bottom.
Test 2: The angry customer
The setup. The customer is upset and rude, but the problem itself is simple, such as a late delivery or a double charge.
THIS IS THE THIRD TIME I'M WRITING. Nobody at your company reads anything. Fix it or I'm posting reviews everywhere.
What you are testing. Do they stay calm, and do they get to the fix quickly?
A good answer:
- Names the frustration once, plainly: "I'm sorry you've had to write three times. Let's fix this now."
- Starts on the problem in the same message.
- Does not argue, match the tone, or react to the review threat.
A weak answer:
- Ignores the feeling and pastes a script.
- Apologizes in every line but does nothing.
- Gets short or sarcastic.
- Offers a discount or credit to calm things down, when the policy sheet does not allow it.
Test 3: The question they cannot answer
The setup. Ask something your policy sheet does not cover, something no one could know on day one.
Does your app work with the accounting software we use?
What you are testing. Honesty. A confident wrong answer does real harm, because the customer acts on it.
A good answer:
- Does not guess.
- Says what they will do, and by when: "I don't want to give you a wrong answer. I'm checking with our product team and will reply here by the end of the day."
- Or sends the customer to the right place, if one exists.
A weak answer:
- Makes up an answer that sounds sure.
- "I don't know," and nothing more.
- Closes the chat with no next step.
In your instructions, tell applicants they may do anything a real agent could do on day one, including checking with a teammate. Then see who uses it.
Test 4: Two chats at once
The setup. Run two customer conversations at the same time. Make one simple ("Where is my order?") and one that needs care, such as a billing mistake. If the job means three chats at once, test with three.
You do not need special software. Two people on your team can each play a customer in separate chat windows, sending messages a minute or two apart.
What you are testing. Can they keep both customers informed without mixing them up?
A good answer:
- Sends short holding messages when needed: "Thanks, checking that now."
- Never mixes names, order numbers or details between the chats.
- Closes the simple chat quickly and gives the harder one proper care.
A weak answer:
- Leaves one customer waiting a long time with no word.
- Puts a detail from one chat into the other. In some businesses, such as a medical office, this is a privacy problem, not just a slip.
- Rushes both and does neither well.
Test 5: The tricky written instruction
The setup. Give a written task with one detail that is easy to miss. This tests reading, which is a big part of chat support.
Reply to the customer below. Offer the free shipping code only if the order has three or more items. End by asking if there is anything else you can help with.
Customer: "Hi, I just ordered two shirts. Do you have any shipping discounts?"
What you are testing. Do they read every line, and do they follow the rule even when breaking it would please the customer?
A good answer:
- Does not offer the code, because the order has two items.
- Explains kindly, and says what would qualify if your policy allows it.
- Ends with the question, as asked.
A weak answer:
- Offers the code anyway.
- Skips the closing question.
- Answers a different question than the one asked.
Your policies, saved replies and account notes are all written down. Someone who misses a line in a test will miss lines on the job too.
How to score the answers fairly
Use the same four points for every test:
| Point | What to look for |
|---|---|
| Correct | Followed the policy sheet and made no promises they cannot keep |
| Clear | Short, easy to read, answer first |
| Kind | Tone fits the customer's mood |
| Complete | Answered every question and gave a next step |
Score each point from 1 to 3. Look at speed after quality: a fast wrong answer is worse than a slower right one. Spelling matters, but a small slip matters less than a false promise.
If you can, hide names while you score. That way you judge the writing, not the person's background.
Tests also let you start with the strongest few instead of reading every application in full. When the people behind HireFlow hired a support team lead for their own business, 153 people applied in the first four days, 71 finished the written interview, and 3 were worth a call. They hired from those three.
If you would rather not build tests by hand, HireFlow gives each applicant a realistic customer chat picked for your kind of business and measures typing speed and reply time as people write. Apart from the "Are you 18 or older?" question every application starts with, a person on your team makes every decision.
In short
- Give every applicant the same five tests, with the same policy sheet and time limit.
- Test a rule they must hold, an angry customer, a question they cannot answer, two chats at once, and a written instruction with a catch.
- Decide what a good answer looks like before you read any.
- Score each answer on correct, clear, kind and complete, then look at speed.
- Meet the strongest few, and use the interview to ask about their answers.