The pipeline that runs Predictify's channels
Three social accounts, one person, and no hours in the day. It now runs without me.
The situation
Predictify has accounts on X, Instagram and TikTok. I am the entire marketing department, and I also build the app.
Football content has a short shelf life. A transfer story is interesting for about a day. A match preview is worth nothing on Monday.
What got built
A scheduled job runs every day and gathers what happened in football, from a fixed set of sources chosen for being both reliable and actually readable by a machine. From that it drafts posts across four content types: news, transfers, football history, and previews of matches worth previewing.
A second job watches for breaking news through the day, reading the front pages every hour or two. That one only publishes if a story clears an importance threshold, because an account that posts every minor story is worse than one that posts nothing.
The judgement built into it
This is the part that matters, and the part that takes the time.
A system like this is only as good as the rules about what it refuses to do.
None of that is technically difficult. All of it is the difference between an account that reads as a person and one that reads as a bot.
- Nothing older than 24 hours.
- Never friendly matches.
- Two sources before publishing.
The boring problem that ate a day
What broke
The images are generated by one service and published by another. The publishing service only stores a link to the image, and the generating service's links expire, sometimes after three hours, sometimes after a day, with no way to predict which.
So posts scheduled for the afternoon would go out with a broken image.
The fix
A permanent image host in the middle of the chain. Half a day of work, entirely invisible in the finished product, and without it the whole thing quietly does not work.
Every automation has one of these.
Budgeting for it is most of what experience buys you.
It started in draft mode. It does not need to any more.
- For the first week, nothing published. Everything went into a queue for me to approve or reject.
- Headlines that were technically accurate and completely lifeless. Image text too short to say anything. Every one of those became a rule.
- Then the corrections got rarer, then rare, and at some point I was approving everything without changing it.
The system now runs on its own, across all three channels, and I do not review posts before they go out.
Tell me where your week goes.
Thirty minutes. If I cannot help, I will say so on the call.