A predictive content calendar does not predict single posts. It uses your last 90 days of results to decide what to schedule next, and it is right more often than a calendar built on hunches. Four weeks fixed, the rest held as themes, and every slot justified by a number from your own account.
TLDR: Export 90 days of posts, tag each by format and topic, find the tendencies that beat your median, and fill next month's calendar to match. Re-run it every four weeks.
What a history-based forecast can and cannot tell you
It cannot tell you that Tuesday's post will get 4,000 views. It can tell you that carousels about customer questions beat your median in 7 of 9 attempts.
That second statement is a tendency, and tendencies are enough to plan with. Treat each one as a bet you have evidence for, not a guarantee.
Public benchmarks help set expectations. In Mydrop's data, carousels ran 10% above their account's typical post, and a link in the caption cost 7%. Check whether the same holds for you before you build a plan on it.
Step 1: Pull 90 days of posts into one table
You need five columns per post: date and time, platform, format, topic and engagement rate. Anything under 30 posts is too thin to read, so extend the window to 180 days if you post rarely.
In Mydrop, the analytics view shows per-post metrics you can sort by platform, so the export takes minutes, not an afternoon.
Step 2: Tag every post by format and topic
Use at most five topics, for example "how-to", "customer story", "behind the scenes", "offer" and "opinion". More than five and the groups get too small to compare.
Step 3: Find the tendencies that beat your median
For each format and each topic, calculate the median engagement rate. Compare it with your overall median.
| Result | What to do with it |
|---|---|
| 20% or more above median, 5 or more posts | Schedule more of it next month |
| Within 10% of median | Keep it at current volume |
| 20% or more below median, 5 or more posts | Cut it or rework the format |
| Fewer than 5 posts | Run it two more times before judging |
The five-post floor matters. Two good posts in a row can be luck.
Step 4: Place slots by timing data, not by folklore
Weekday matters less than most guides say. In Mydrop's data, every weekday sat within 2% of the account's median, and the clear loser was posting between midnight and 6 am local time, 17% below the median across platforms. So avoid that window, then test two or three slots of your own for a month.
For the longer argument on this, see why best time to post is a vanity metric.
Step 5: Fill four weeks, hold the rest as themes
Lay out the next four weeks on the content calendar. Give each week one theme and put your winning formats in your tested slots. Leave one slot per week open for something reactive.
For the weeks after, write only the theme and the target format. Detail goes in when the previous month's results are in.
Keep a record of why each slot exists
Add a note to every calendar entry: "carousel, customer question, beat median in 7 of 9". When a result comes in weak, you can see which assumption broke.
For teams planning across several brands, the multi-brand planning approach shows how to keep one set of rules and give each brand its own numbers.
Three signs your forecast is overfitting
One viral post drives the tendency. Drop the best and worst post from each group and recheck the median. If the tendency vanishes, it was luck.
The sample is under five posts. Run it again before you plan around it.
The result depends on one launch week. Exclude launch weeks from the comparison and see whether it still holds.
Treat any tendency that survives all three checks as real enough to schedule, and any that fails as a question for next month.
Run the first pass this week
Export your last 90 days, tag them, and fill in the table from step 3. Then draft week one of next month using only the tendencies that passed. To keep the data, the calendar and the scheduling in one place, start free with Mydrop.
Google review
Trustpilot review
Product Hunt comment
