Demos are designed; checklists are yours
Every vendor demo shows the same happy path. The differences that will actually cost or save you money live in the pricing fine print, the limits, and the exit doors — none of which appear on a demo call. Evaluate with a written checklist and make every vendor answer the same questions.
1. Price it at YOUR team size, three ways
- Today's team — including the part-timer who covers lunch.
- Next year's team — per-agent pricing compounds quietly; a flat plan does not.
- Peak season — do extra seats for December cost a year's commitment or a month's?
Write the three totals down per vendor. Per-agent tiers plus paid add-on modules routinely double the sticker price by the time your real usage is priced in.
2. Read the free plan's fine print
"Free" hides in the limits: how many agents, how many conversations per month, how long is chat history kept, whose branding is on the widget, and — the one that bites later — what happens to your data if you stay free forever. A free plan with a monthly chat cap stops being a plan the day your site gets traffic; a free plan with unlimited chats is an actual tier.
3. Ask exactly how AI is charged
AI pricing models differ more than any other line item: per resolution, per session bundle, per agent add-on, or included. Ask each vendor: "if the bot answers 500 questions next month, what does that cost?" — and get the number in writing. A metered bot quietly converts your growth into their revenue; know which model you are signing.
4. Confirm the exit before the entrance
- Can you export transcripts, contacts, and knowledge-base articles — yourself, without a support ticket?
- Is there an API for the data, or only a dashboard?
- What deletes when you cancel, and when?
A vendor's answer to "how do I leave?" tells you more about the relationship than any feature grid.
5. The 30-minute hands-on test
Sign up for the free tier of your two finalists and run the same script on each: install the snippet on a test page, send yourself a chat from a phone, answer it from the console, trigger the offline state, and find the transcript afterwards. Note where you got stuck — YOUR stuck points, not the reviewer's. Thirty minutes of this beats thirty comparison tabs, because the product you can set up without help is the product your team will actually run.
Score it and decide
Total cost at next-year size, free-plan honesty, AI pricing clarity, data exit, and the hands-on friction score — five rows, two finalists, one afternoon. (MyLiveChat's answers are public: flat plans, a free tier with unlimited chats, AI without per-resolution metering, and a documented API — start the hands-on test free and hold us to the same checklist.)
Five questions vendors hope you skip
Feature grids are built to be compared favourably. The questions that actually separate products are the operational ones, and they are rarely on the comparison page.
- What happens to my transcripts if I stop paying? Are they exportable, for how long, and in what format? A CSV of the last ninety days is a very different answer from a full archive.
- How is the AI metered, exactly? Per conversation, per message, per resolved chat, or on your own provider key? The unit determines whether a busy month is an inconvenience or a bill nobody approved.
- What does a seat mean when someone leaves? Can you reassign it mid-term, and does the outgoing person's chat history stay with the account?
- Which domains does the widget load from? You need this before you meet a content security policy, not after the widget silently fails to appear.
- Where is the data stored, and who can reach it? Ask for the answer in writing, and ask what the support team can see.
The trial week that tells you something
Most trials are spent in the settings screen, which measures nothing you care about. Put the product in front of real traffic for a week instead, on one page, with one agent, and you will learn more than a month of evaluation.
Design the week around the situations that will actually hurt. Have two people take a chat at once and see how the queue behaves. Close the laptop mid-conversation and watch what the visitor sees. Have someone open the widget on a three-year-old Android phone on a bad connection. Send a file. Refresh the page mid-chat and see whether the conversation survives. Then, at the end of the week, ask the agent — not the buyer — whether they would keep it, because the person living in the console has a far better read on daily friction than the person watching the demo.
Write down what you saw as you go. Evaluation memory is unreliable, and by the third product every widget feels roughly the same unless you kept notes.
After you choose: the first thirty days
The decision is not finished at signature. Set a review date a month out, and agree in advance what would make you change your mind — a response-time target you did not hit, an integration that never got built, adoption that stalled at one agent. Teams that skip this step tend to stay on a tool that is not working, because by then the switching cost feels larger than the annoyance.
Keep the exit warm too: export your transcripts once during the first month, purely to confirm the export works and that the format is something you could actually load somewhere else. It takes ten minutes and it converts a contractual promise into a verified fact, which is the only form of it worth having.