Losing Website Leads After Hours: What an AI Chatbot Changes

You can read the rest of this article in any order, but the answer sits in three numbers you can pull from your own site in the next half hour. The first is the share of your sessions that arrive outside your staffed hours. The second is the share of enquiries that never get a reply at all. The third is the monthly message allowance on a Starter, Standard or Pro plan. Once you have those three, the rest of this page is just arithmetic.

Why your contact form is not covering the gap

A contact form is a recorder, not an answer. The visitor types a question, your inbox grows by one row, and the visitor walks away without knowing whether you ship to their address, whether you are open tomorrow, or what the price is. The form is useful for two things: tracking what people asked, and giving you evidence later of how much traffic left without a reply. It does not serve the visitor who sent it.

The distinction matters because a lot of sites have quietly assumed the form counts as cover. It does not. The site still looks closed. The page still says "we will get back to you within one working day" at 11pm on a Sunday. The visitor still books elsewhere, or closes the tab, or sends the message to a competitor. The form gives you a record of a lost opportunity. That is the only honest reading.

An AI chat agent is different in kind. It answers. It can say yes, we ship there, here is the price, here is the next delivery slot. It can answer the same question at 2am that it would answer at 2pm, and it does not need to be prompted. The form is a passive listener. The agent is a published answer, on demand, at any hour.

Step one: find your own out-of-hours share

Open your analytics. Most sites have Google Analytics 4, Matomo, Plausible, Fathom or similar. Find the report that shows sessions by hour of day. Most of these tools show it as a bar chart with hour on the x-axis and sessions on the y-axis. You are looking for the bars outside your staffed window.

If your staffed hours are 9am to 5pm in your local time zone, that is eight hours. The other sixteen hours are out of hours. The seven hours before opening and the nine hours after closing, plus the entire weekend. Count the sessions in each.

Do not take a number from an article for this, including this one. Your own share depends on your audience, your time zone and your opening hours, and no published average describes it. You are not guessing this. Your analytics has it. Pull it.

The reason this share matters more than total traffic is that it answers a question you cannot answer any other way: how many visitors arrived when there was nobody on the other end. Every other conversion metric, bounce rate, pages per session, scroll depth, all of them blur staffed and unstaffed hours together. This one does not.

Step two: count the enquiries that got no reply

The second number is harder to find but it matters more. It is the share of contact-form submissions, sales emails or live-chat requests that never received a reply from your team. A lot of sites do not measure this at all, which is why the problem persists.

The method is straightforward. Take your inbox or CRM and the enquiries from the last 30 or 60 days. For each one, check whether your team sent a reply. If your reply is missing, the enquiry is unanswered. The number of unanswered enquiries divided by the total is your unanswered share.

Be honest about what counts as a reply. An out-of-office auto-responder does not. A "thanks, we will get back to you" does not. A real answer does. The two-step method works because the gap is large enough that you do not need statistical precision. Even a rough count gives you the right order of magnitude.

If the unanswered share is high, two things are true at once. Your team is overloaded, and your form is collecting evidence of an unsolved problem. The form is making the situation worse, not better. Every unanswered submission is a visitor who got the same treatment they would have got from a closed shop. They came, they asked, they heard nothing, they left.

Step three: work out the size of the window

Multiply. The two numbers above give you a rough count of sessions that arrived when nothing useful happened. Multiply the out-of-hours share by total sessions to get the size of the unstaffed window in sessions. If you also know the share of sessions that start a chat, or the share of sessions that fill in a form, you can convert sessions into enquiries.

This is the volume the agent would handle. Each conversation with the agent is one message or one thread, depending on how the platform counts. Asyntai counts messages, with the published allowance per plan. The conversion above gives you the rough monthly volume to plan against.

Which Asyntai plan fits that volume

Asyntai publishes four self-serve tiers and one contracted tier. The published message allowances are 100 on Free, 2,500 on Starter at $39 a month, 15,000 on Standard at $139 a month, and 50,000 on Pro at $449 a month. Enterprise is a signed order form with a negotiated allowance, not a self-serve purchase.

Each published plan has a per-message cost you can work out yourself. The maths blocks below use only the published price and the published allowance, no other inputs.

Two things to notice. The published per-message cost falls as you move up the tiers. And it falls by a factor of nearly two between Starter and Pro, even though the monthly price is over ten times higher. The published plans are priced for volume, not for occasional use. If your out-of-hours volume is small, the right plan is small.

Pick the plan using your own numbers

Take the out-of-hours enquiries you worked out in step three. That is your monthly message volume. Match it to the published allowance.

• Up to 100 messages a month: Free plan, $0, useful for trying the agent and seeing how it handles your real questions.
• 100 to 2,500 messages a month: Starter plan, $39 a month, the right tier for most small sites once they have measured their actual volume.
• 2,500 to 15,000 messages a month: Standard plan, $139 a month, the right tier once your out-of-hours volume passes the Starter cap.
• 15,000 to 50,000 messages a month: Pro plan, $449 a month, for sites whose visitor base or enquiry intensity pushes past Standard.
• Over 50,000 messages a month: Enterprise, signed order form, negotiated allowance.

This is a one-step process if you trust your measurement. Two steps if you do not. The conservative move is to start one tier below your measured volume, watch the meter for a month, and move up if the allowance runs out. Asyntai's published plans are month-to-month, so the tier change costs nothing to undo.

What an agent actually does at 11pm

This is the part most comparison articles skip. The interesting question is not what an agent costs, it is what happens when a visitor arrives at 11pm and starts a chat. The flow is simple to describe and easy to underestimate.

1. The visitor loads a page on your site. A small script tag, or a Code Injection block on Squarespace, loads the chat widget.
2. The widget opens. The visitor types a question. The question is sent to the agent.
3. The agent reads your published content, plus any documents you have pointed it at, and writes a reply.
4. If the question falls outside the agent's scope, or the visitor asks for a human, the conversation routes to your team with the transcript attached.
5. Every line is logged in the audit log. Nothing in that log can be edited or deleted after the fact.

Three details matter. First, the visitor does not need to wait. The reply is seconds, not hours. Second, the visitor does not need to fill in a form. The chat is the form, but it returns an answer instead of a confirmation. Third, PII redaction is configurable, so you can decide what gets written to the log before the visitor types a single character. The output is filtered by a separate classifier, not by a prompt instruction, which is the safer place for that control.

What changes the answer

The arithmetic above holds for a site whose out-of-hours volume is roughly constant through the month. Some sites have spiky volume. Seasonal sites, sites with a weekly campaign, sites that run a monthly product drop. The published plans are monthly allowances, not daily. A spike that doubles your volume for a week will not double your monthly cost if the rest of the month is average. If the spike is sustained, you move up a tier.

Other things that change the answer:

• Average messages per conversation. A short factual exchange is one or two messages each way. A longer exchange can be ten or more. Count messages, not conversations, because the published allowance is messages.
• Deflection of staffed-hour traffic. Once the agent is on the site, it answers some of the questions that would otherwise have arrived at 11am. That is not a loss. It frees your team. It does mean your total message volume will exceed your out-of-hours volume from step three.
• Languages. If your visitors write in more than one language, the agent can answer in more than one language. The message cost is the same.
• Handover behaviour. If you set the agent to hand off aggressively, your team gets more conversations and the agent's message count is lower. If you set it to handle more questions itself, your team gets fewer and the agent's message count is higher. Pick the behaviour, then check the meter.

The mistake most people make

The mistake is measuring the wrong thing. Sites measure traffic and assume traffic is the cost driver. Traffic is not the cost driver for an agent. Messages are. A site with twice the traffic and a quarter the message volume pays less, not more, because most of those extra visitors never open the chat. This is also why the per-message cost falls as you move up the published plans: high-volume sites tend to use the allowance more efficiently.

The second mistake is to count the contact form as a solution. It is not. The form is a record of a question. The agent is an answer to one. Until you measure the unanswered share, the form looks like it is working. Once you measure it, the form looks like a list of visitors you let down.

The third mistake is to pick a tier based on what feels right. The published plans are priced to be predictable at a known volume. The right plan is the one whose published allowance is closest to, but above, your measured volume. The cheapest plan above your volume is the one you want. Anything else is either overpaying for headroom you do not need, or underbuying and getting cut off mid-month.

A worked example using only published numbers

Suppose a site measures 8,000 sessions a month and finds an out-of-hours share of 50%. Suppose further that 1.5% of sessions open a chat, and the agent answers three quarters of them itself. The monthly message volume is worked out below.

240 messages a month fits inside the Starter plan, which has a published allowance of 2,500 messages at $39 a month. If the share rises to 70%, the same arithmetic gives 336 messages a month, still inside Starter. If the site adds a marketing campaign that lifts chats to 4% of out-of-hours sessions, the volume rises to about 896 messages a month, still inside Starter. The plan choice changes with measurement, not with guesswork.

The same example at higher intensity

Suppose the same site is in a vertical where every visitor asks something: a complex B2B purchase, a regulated service, a high-consideration transaction. Suppose the chat-open rate is 8%, the average conversation is 6 messages, and the agent handles 80% of the conversations itself. The arithmetic is:

1,920 messages a month still fits inside the Starter plan's 2,500-message allowance. The per-message cost on Starter is $0.0156, so the implied cost of covering those out-of-hours messages is roughly $30 of the $39 monthly bill. The remaining $9 covers the staffed-hour deflection that the agent also handles. This is why the measurement matters more than the marketing copy. The plan is the same; the workload is very different.

What to do on Monday morning

1. Pull sessions by hour of day from your analytics. Compute the share outside your staffed hours. Write it down.
2. Pull your last 60 days of enquiries. Count the ones that got a real reply. Compute the unanswered share. Write it down.
3. Combine the two to estimate monthly out-of-hours message volume.
4. Match that volume to the published Asyntai allowances: 100, 2,500, 15,000, 50,000.
5. Pick the cheapest plan above the line. Start one tier below if you are not sure.
6. Install the script tag, or the Code Injection block on Squarespace, and watch the meter for a month.
7. Move up or down a tier at the end of the month based on actual usage, not on feeling.

This is the whole plan. The arithmetic is the plan. The plan choice falls out of the numbers you can already pull from your own site. The contact form is not the cover. The agent is. Everything else is a question of which published allowance fits your measured volume.