How to test your AI before going live

Before your AI starts taking real calls, you can put it through its paces. I'm going to walk you through exactly how I recommend testing, what to look for, and how to report issues so your onboarding team can fix everything in one clean pass.

This takes about 30 minutes and it's the single best thing you can do to make sure your AI sounds sharp from day one.

Before you pick up the phone, sit down and write out a list of the scenarios your AI will actually face. Think about the types of callers your firm gets, and make sure you cover both people you want and people you don't.

Here's a practical example of what this looks like:

Your list should be diverse and include a mix of:

The more realistic your list, the more useful your testing will be.

When you call in pretending to be each of those test leads, you're checking four things.

Does the AI correctly figure out whether this caller is someone your firm can help?

Call in as someone who clearly qualifies and confirm the AI moves forward. Then call in as someone who doesn't qualify and make sure the AI handles it properly, whether that's a polite decline, a referral, or a redirect.

Things to watch for:

Once the AI decides a caller qualifies, does it ask all the questions you need? For example, if you handle med mal cases, the AI should be gathering details about the injury, the medical provider, the timeline, and anything else your team needs to evaluate the case before the first call with an attorney.

Call in as a qualified lead and let the AI run through its full intake. Pay attention to whether it misses anything your team would need.

If your AI is set up to book appointments or sell consultations, test what happens when the caller pushes back. Say things like "I'm not sure I'm ready" or "Can I just talk to a lawyer first?" or "That sounds expensive." A well-configured AI should handle these smoothly without getting stuck or giving up too easily.

If your AI is set up to transfer qualified leads to your intake team, test the handoff. Call in as a qualified, urgent lead (like a criminal defense case where someone is in custody) and see what happens. Check whether the transfer fires at the right moment, whether your team gets enough context, and whether the caller experience is smooth.

Here's the part most firms get wrong. They test a few calls, find a couple of issues, report them one at a time, get fixes, test again, find more issues, report those. This back-and-forth drags out onboarding and each fix only sees a sliver of the picture.

Instead, do all your test calls in one sitting.

After every call, jot down what went wrong (or what went right). Once you've gone through your full list, compile everything into one message to your onboarding team.

A good issue report looks like this:

When your onboarding team sees everything at once, they can spot patterns, fix root causes instead of symptoms, and get your AI dialed in much faster. One round of thorough feedback beats five rounds of scattered notes.

Aim for at least one call per scenario on your list. If you have three practice areas, you want at least one qualified and one unqualified caller for each, plus any appointment or transfer scenarios. Most firms end up making 8 to 15 test calls.

Yes, if your firm has web chat set up through Lexidesk. The same test cases apply. Just type as if you were the caller and check the same four areas: qualification, intake questions, objection handling, and any handoffs.

That's normal and that's the whole point of testing. Better to find 12 issues in testing than to have 12 real leads get a bad experience. Send the full list to your onboarding team and they'll work through it.

Yes. Once your onboarding team confirms the updates, optionally, you can through the same test cases one more time to verify. This second round is usually much faster because most things will be working correctly.