Guides

Your first week with Vesma

Last updated 2026-09-14

Setting Vesma up is not the hard part — the wizard takes about fifteen minutes. What decides whether it works is the order you do things in, because one of the steps waits on Meta and one of them is the difference between good answers and vague ones.

This is the order we would do it in, and what to look at on day seven to decide whether it is earning its place.

Day one: the six setup steps

Signing up drops you into a six-step wizard. None of it is throwaway — every answer changes how the AI behaves later.

Name your workspace. What your team sees, and what appears on your public pages.
Pick your niche. This one matters more than it looks. It tunes the assistant's guardrails and the message templates, and it renames things throughout the product — a booking becomes a site visit for real estate, a consultation for interiors, a demo class for coaching.
Business details. Used verbatim in replies and bookings, so a wrong address is a wrong address in front of a customer.
Add your first project. What the AI is selling. You can add the rest later, but it cannot quote what does not exist.
Upload your knowledge. What it answers from. Covered on its own below, because this is the step people rush.
See it work. Ask it a question the way a customer would. Do this honestly — try the awkward question, not the easy one.

Day one: the channel that needs nobody's approval

Put the website chat widget on your site on the first day. It is a script tag, it needs approval from nobody, and visitors get exactly the same AI that runs on WhatsApp with conversations landing in the same inbox.

The reason to do it first is not the traffic. It is that you get to watch the assistant answer real people within hours, while every other channel is still waiting on somebody else.

Day one: start WhatsApp, because it waits on Meta

WhatsApp is where most enquiries will end up, and it is the one step you cannot rush — it runs through Meta's official onboarding. Start it on day one so the waiting happens in parallel with everything else rather than after it.

One decision there is genuinely hard to reverse, and it is explained in Connecting your WhatsApp number. Read that before you pick which number to use, not after.

Day two: the hour that decides answer quality

Every answer the AI gives is grounded in documents you have approved. It does not know your business from the internet and it will not guess: asked something your documents do not cover, it says it will check with your team rather than inventing an answer.

That is the safety property that makes it usable at all, and it is also why a thin knowledge base produces a thin assistant. An hour spent here is worth more than any other hour in the first week. What to upload, and what it can and cannot read, is in What your AI answers from.

Days three to six: watch, do not tune

The temptation in week one is to read every conversation and adjust something after each. Resist it. You cannot tell signal from noise at five conversations, and each change resets your read on the previous one.

Do these instead:

Read the inbox daily, but only act on real handovers. When the AI hands a conversation over it is telling you something it could not answer. That is the highest-value feedback you will get, and usually the fix is a document, not a setting.
Add the missing answer to the knowledge base, not to the conversation. Answering it yourself once helps one customer. Uploading the document answers it for everyone afterwards.
Leave the follow-up cadence alone. The default is five touches over two weeks. You cannot judge it until the two weeks have passed.

Day seven: is it working?

Three things tell you, and none of them is the number of leads.

Median first reply. On the dashboard. This should be seconds. If it is not, the AI is not actually handling the conversations — something is escalating early.
How many conversations the AI finished without you. A high handover rate in week one is normal and is almost always a knowledge-base gap. A high handover rate in week three is a problem worth looking at.
Whether anything got booked. The whole system exists to produce appointments. One booking in week one from a standing start is a working system; zero with thirty enquiries is a signal to read the conversations.
What week one will not tell you

It will not tell you whether your ads are working, because the outcome loop needs qualified leads before it has anything to report, and Meta needs volume before it can learn from them. Judge that at week four, not week one — see Telling Meta and Google which leads were real.

Set it up and watch it answer

The free plan carries every feature. Put the widget on your site and you will see it against real enquiries the same day.