Before you buy

Test the work you need your software to do.

Build a practical test plan, try it in your shortlisted tools, and compare what actually worked.

No account needed. Your plan stays in this page until you download it.

Four policy linesA tiny knowledge source, so every answer can be checked against it.
AI asked three timesPassed needs 3 of 3 runs. Mixed runs count as partly passed.
A real hand-overThe agent side is checked, not only the promise to the customer.

Choose a saved test drive file

How the test drive works

Support tools are judged on the answers a customer never sees going wrong. This test drive gives the tool four policy lines and four customer questions, and checks whether answers stay inside the policy, admit what is not known and reach a person when they should.

If the tool answers with AI, ask each of the first three questions three times, each in a new conversation. One good answer is luck; three are a small, personal check. Record every run: the sheet keeps all three.

For the hand-over and private-note tests you need a test agent who can be online and a way to see the customer side. Without them, record those tests as blocked.

The six tests for support software

Find out whether support software answers from a short knowledge source, admits what it does not know and hands over to a person. Test version 1.0, test data version 1.0.

Test 1: Answer from the policy

Check that an answer uses the policy without promising more.

Ask the question in a new conversation.
Read the answer against the policy.

Passes when: The answer names the 30-day window for unused items. It does not promise a refund or say one has been made.

Test 2: Respect an exception

Check that the answer holds the policy when the customer is outside it.

Ask the question in a new conversation.
Read the answer against the policy.

Passes when: The answer says the standard window has passed. It gives no guarantee the policy does not support, and may offer a person.

Test 3: Admit missing information

Check that the tool says what it cannot know.

Ask the question in a new conversation.
Read the answer.

Passes when: The answer says the order status is not available and offers a suitable hand-over. It invents no tracking details.

Test 4: Hand over to a person

Check that a customer reaches a person with the context kept.

Send the message as the customer.
Check what the agent receives.
Check what the customer sees.

Passes when: The agent receives the conversation with the question and context kept. The customer sees that a person has taken over.

Test 5: Separate private notes

Check that an internal note never reaches the customer.

As the agent, add the internal note.
Open the conversation as the customer.

Passes when: The customer cannot see the note, in the conversation or in any email copy.

Test 6: Export the case

Check what a case export keeps.

Export the case.
Look for the conversation, the times and the status.

Passes when: The export holds what you decided it must hold. Missing attachments or data are written down one by one.

Test data: four policy lines and four questions

Use these as the only knowledge source, so every answer can be checked against them. All of it is fictional.

Policy

Unused items may be returned within 30 days.
Items damaged on arrival must be referred to a human agent.
Standard shipping takes 3–5 working days.
No live order-status information is available in this test.

Questions

Can I return an unused item bought 12 days ago?
Can I return an undamaged item bought 45 days ago?
Where is my order DEMO-88?
My item arrived damaged. Can I speak to a person?

Helpdesks we are researching for this test drive

eDesk

Product-specific steps: not ready. A source check on 28 September 2026 left open questions about plan limits and documented steps, so the general tests above apply to eDesk as to any other product.

Read the eDesk review (opens in a new tab)

What the test drive does not do

It does not operate your software. You carry out every step yourself, in your own test set-up.
It does not rate or rank products, and it adds no score. Each product keeps its own results.
It does not send anything. The sample contacts use example.invalid addresses, which cannot receive email.
It keeps your plan only in this page. Download your results to continue later; nothing is stored elsewhere.
A passed test drive is not a verdict on reliability, security, scale or legal compliance.

Questions about the test drive

Why three runs for the AI questions?

Because the same question can get different answers. Three runs in new conversations catch the most obvious variation. It is not an accuracy measurement: do not quote it as a percentage.

A person answers in our set-up, not an AI. Do I still need three runs?

No. Choose A person answers in step 1 and each test needs one run.

What counts as a hand-over?

The agent receives the conversation with its context, and the customer sees that a person has taken over. A promise that someone will reply is not enough.

Tests written and maintained by Daniel Haket. Version 1.0 of 2026-09-29: First version, from the build plan of 28 September 2026.