Funnel review

performance-review asks which posts worked. This skill asks what happened to the people a Reel sent down the path, where they stopped, and what to change. Each round changes one thing, so the next one can tell whether it helped: that loop is what improves a funnel, more than any single rewrite.

If no automation has run yet, there is nothing to read: use reel-funnel instead.

1. Collect the numbers

With Saypage connected: get_account, then list_automations. There the messages read opening_message_copy, opening_message_button_label, private_message_copy (the link message) and recipe_link_button_label; the gates follow_gate_enabled and email_gate_enabled; the destination page_id or custom_link.

Then, for the funnels that ran, ten at most: the most links_sent first, then the newest Reels (get_automation_activity allows about six calls a minute: when it refuses, wait the seconds it names):

Ask how many sales or bookings came in since the funnel started, and whether their tool shows the link's tag. If they have no number, say the path stops at the click, and that a tagged link would show the rest next time.

State only numbers that a response returned or the user gave. When a metric is in unavailable, pass on its reason.

Leave out the user's own tests. recent names each commenter: on a small funnel, accounts that are plainly theirs can be most of the numbers. An automation counts from its activated_at, so links_sent: 0 on a new one is no result.

2. Check that the numbers belong together

These tools count over different windows. Before putting page loads next to links_sent, all three must hold:

1. This automation alone sends this page, as its main link or an extra button (list_automations).
2. days covers the whole life of the funnel. get_page_stats goes back 90 days, while the activity summary covers 13 months.
3. The automation stayed on this post. An automation moved from another Reel carries that Reel's history in its summary.

If one fails, report page loads on their own, with their window, and draw no ratio across them. Even when all three hold, page loads count every load, repeat visits and link-preview bots included, so call the figure "page loads per link sent", which can pass 100%, and never a conversion rate or a number of people.

Clicks are taps on a page's links, not people: one person can tap twice, and a few taps are missed. Saypage counts them since November 2026, so a funnel that ran before then shows fewer than it got, and it sees nothing after the click.

3. Read the path

Wait for at least 48 hours of activity. Below about 30 keyword comments, show the counts and skip the percentages: a few people more or less would swing them.

• Keyword comments per 1,000 views: triggered ÷ views × 1,000.
• Tap rate: tapped ÷ first_dm_sent.
• Gate rate: gave_email ÷ asked_for_email.
• Delivery: links_sent ÷ triggered.
• Click rate, when step 2 holds: instagram_dm in click_channels ÷ instagram_dm in channels, labelled as an estimate.
• Sale rate, when the user gave a number and step 2 holds: sales ÷ clicks from DMs, labelled as an estimate.

People still under way (waiting_for_tap, waiting_for_follow, waiting_for_email, in_progress) are not lost yet: say how many there are.

There is no universal good rate, so quote no benchmark. Compare a funnel with the user's earlier funnels, and with itself after a change.

4. Compare the funnels

With several automations, line them up before fixing anything, one row per funnel, using the measures of step 3. Then:

• Start with the funnel where a fix reaches the most people: new keyword comments keep coming on a live Reel, and a quiet one has little left to win, so note its fix for next time.
• Copy what the best funnel does. When one taps far better than the others, read both sets of messages side by side: its wording, its button, its gate, its keyword. Different Reels bring different people, so treat the gap as a lead to test, not as proof.

5. Find the weakest step, and fix only that

When the page does not send people on, reread the page and go through "A page made for one Reel" in get_page_guide: does the first screen match the Reel, is the main button visible at once, does it say what they get and the price, does the page answer the doubts from the comments? If the page holds up, the question is the offer (offer-finder), and the user may know more than the numbers do: ask what people say when they do not buy.

Propose one change, written out in full. Two changes at once cannot be told apart.

6. Apply it, with the user's go-ahead

• DMs and keyword: update_automation. A live automation stays live and keeps its history. Each message stays in the user's language, and button labels stay within 20 characters. The change applies to new comments only.
• Page: update_page, preview_page, then publish_page once they agree.

7. Check it worked

Followers cannot be split into two groups here, so each test runs one after the other: note the numbers, change one thing, then compare what came after.

The activity summary keeps adding up, so keep today's numbers and the date in the output. At the next round, subtract them to get the period since the change, and wait for about 30 new keyword comments before judging it. A better result on a fresher, bigger Reel may come from the Reel, so say so. Keep what helped, undo what did not, then move to the next weakest step.

Output

FUNNEL  ·  PASTA  ·  Reel "15-minute lemon pasta"  ·  since 3 Oct

views               52,300
keyword comments       612   12 per 1,000 views
first DM sent          604
tapped                 391   65% of first DMs    <- biggest loss
links sent             388
page loads from DMs    402   about 1 per link sent (loads, not people)
clicks from DMs        118   29% of page loads (taps, not people)
sales                   31   from your shop, your number

STILL UNDER WAY   9 waiting for the tap

WEAKEST STEP   a third of people never tap the first DM
WHY            the message doesn't say what they get
CHANGE         "Here's the 15-minute pasta 🍋 Tap below and the recipe is yours."
               button: "Send me the recipe"
COMPARE WITH   tapped 391 of 604 on 10 Oct: review again in a week

Run every rewritten message through natural-voice before showing it.