If ten people click the LINE button, did ten new messages arrive?
We cannot tell from that click. It opens LINE; someone might send nothing, write later, or return to the website and use the form. A number labelled “contacts” can easily mean nothing more than “button clicks.”
When we updated measurement on INTECH’s Contact page, we started by separating those meanings. These are decisions from our own website, rather than a client success story. We do not yet have results showing how many additional customers this approach brings in.
Name the event after what you know
We track four distinct actions:
| Event | What it tells us | What it does not tell us |
|---|---|---|
| LINE click | Someone activated the link to LINE | Whether they sent a message |
| Email click | Someone activated the link to their email app | Whether they wrote or sent an email |
| Form start | Someone actually edited a form field | Whether they will finish or submit it |
| Accepted submission | The email provider accepted the request | Whether it reached the inbox or the team replied |
“Accepted” has a specific boundary here. We count it after Resend returns an acknowledgement with a message ID. Clicking Submit is not enough, and the event says nothing about a sale or completed service.
Delivery and response times would need another source of evidence. Clear names keep the report within what the system can verify.
Focusing a field is not filling it in
Someone might tap the name field and move on. A link from a service page might also preselect a service for them. Counting either action as a form start would increase the number before they had entered any details.
On this site, a start means the first user edit that leaves a non-empty value in a data field on that form visit. Focus alone, the consent checkbox alone, and a service value filled in by the website do not count.
We also keep the PDF download separate. It prepares a brief for someone to save or send themselves; downloading it does not deliver an inquiry to INTECH.
Let the server confirm acceptance
The interface explains what is happening, but confirmation that the email provider accepted an inquiry belongs on the server.
The request passes field validation and Turnstile verification before it goes to Resend. Only an accepted response with a message ID can increase the successful submission count. If a provider is unavailable, the form reports the problem and preserves the entered details for a retry.
A retry introduces another question: is this the same accepted message? We deduplicate that receipt rather than count it again simply because a response took too long. Measurement also stays separate from delivery. If the counter store fails after an inquiry has been accepted, the inquiry remains successful. An optional statistic should not turn a working contact channel into a failed one.
Keep the route in, not the customer’s story
We want to understand whether service pages, projects, or articles lead into Contact. Answering that does not require a name, email address, company, or the contents of a brief.
Measurement uses the immediately preceding page category and the page language. It excludes full URLs, query strings, and form contents. The inquiry itself still goes to the team so they can respond; it is separate from the statistics.
This is immediate attribution, not a complete browsing history. Someone who reads an article, opens a project, and then contacts us is attributed to the project page. That helps us assess connections between pages. It does not prove that a particular article won a customer.
We also honour Do Not Track and Global Privacy Control signals. The report will not include every visit, and we do not build individual histories to fill that gap.
Use the numbers to choose the next check
If form starts rise while accepted submissions stay low, first inspect the form and its dependencies. Are the questions difficult? Do error messages help people fix a field? Is verification working? The numbers alone do not show that visitors lost interest.
If LINE gets more clicks than the form, check that the link reaches the right account and that the team is watching that channel. A click count still cannot stand in for a conversation count.
Daily aggregates are retained for 90 days. Transformed identifiers help deduplicate events over 48 hours; they do not give us a count of unique people suitable for calculating a sales conversion rate. We read trends alongside the inquiries the team actually receives.
Before adding another chart, ask three things: when does this event happen, who can confirm it, and what does it leave unknown? Those answers make a report more useful than another impressive-looking number.
See the INTECH website project, or read about retries without duplicate records.