Ask your analytics which channel is working and it will name the last one the customer touched. Usually that is your own brand name typed into a search box, or a direct visit, or the email that carried the calendar link.
None of those started anything. They finished something that another channel started weeks earlier, and that channel gets nothing.
This is not a flaw in one tool. It is what last-click means. It was designed for a world where a customer saw one ad and bought one thing. Nobody buys like that any more, least of all the people who buy from a company they have never heard of.
Three questions, not one
Every channel has to answer for a different job.
Instead of “which channel converts”, ask each channel three separate things.
- Who did you introduce us to that we had never seen before?
- Who did you bring back after they had gone quiet?
- Who did you carry across the line?
Search usually wins the third. It almost never wins the first, because nobody searches for a company they do not know exists. The first is won by the long video, the community thread, the podcast, the newsletter that got forwarded. Those channels look useless on a last-click report and are the reason the report has anything on it.
When you separate the three jobs, the picture flips. The “worst” channel by conversions is often the best at introductions, and the “best” channel is a closer that would have nothing to close without it.
Nobody searches for a company they do not know exists.
The signal that survives every tool
One key runs through every tool you own: the email address, and time.
A person watches a video on Tuesday, reads two pages on Thursday, opens a newsletter the following Monday, and books a call that afternoon. Four tools saw four strangers. It was one person, and the order matters more than any single touch.
You do not need a data team to see this. You need the email address as the join and the timestamps as the order. Take the people who booked last month, walk backwards through everything you can find with their address on it, and write down the first thing each one touched.
Do that for thirty people and the first-touch channel shows itself. It is rarely the channel your last-click report is praising. It is usually the one you were about to cut.
What to do with the answer
Fund introductions. Keep the closer. Cut the middle that does neither.
Once the three jobs are separated, budget decisions get boring in the best way. The channel that introduces gets protected even when its conversion column is empty. The channel that closes gets kept, but not credited with the whole journey. The channels that do neither get their budget moved.
And say the sample size out loud. Thirty bookings walked backwards is a pattern. Six is a story. Both are worth more than a last-click chart, but only one is worth a budget change.
The channel that is working is the one that started the conversation. Go and find where the conversations started.
See the use case for iT consultancies, MSPs and pen-testing firms: How does an IT consultancy, MSP or pen-testing firm get on the list when a company has to meet NIS2 or DORA?
Questions people ask next
Why is last-click attribution misleading for startups?
It credits the final touch, usually branded search or a direct visit, and ignores the channel that introduced the customer weeks earlier. For an unknown company, introductions are the whole game.
How do I find my first-touch channel without a data team?
Take last month’s bookings, walk backwards through every tool using the email address and timestamps, and note the first thing each person touched. Thirty people is enough to see a pattern.
Should I cut a channel with zero conversions?
Not before checking whether it introduces people who convert later through another channel. Separate the introduction job from the closing job first.
Is there a service that shows first-touch channels without a data team?
Throughline does: it joins your tools on the email address and time, shows the first touch for each person who booked or paid, and reports each channel’s role with its sample size.