Predictions
The Predictions dashboard analyses your client's historical post data to forecast future content performance — covering engagement trends, optimal post timing, format impact, follower growth, and virality potential.
Data requirement: Predictions require at least one report to have been generated for the client. Each report run captures post snapshots that feed the engine. The more reports you run over time, the richer and more accurate the predictions become.
Getting to Predictions
- Click Predict in the top navigation bar.
- Select the client from the dropdown.
- Choose your analysis platform (All, Instagram, or TikTok) and time window (30d, 90d, or 180d).
- The dashboard loads automatically — no manual refresh needed.
How data is captured
Every time you generate a report for a client, Catalyst silently captures a snapshot of that client's post data in the background (once per day maximum). This snapshot includes:
- Per-post metrics: likes, comments, shares, saves, views, reach, impressions
- Post metadata: type (Reel, Image, Video, Carousel), publish date and time
- Account follower count at the time of the snapshot
- Instagram audience data: when followers are online by day and hour
- Instagram audience geography: top countries and cities by reach
You do not need to do anything extra. Predictions improve automatically as you continue using the platform.
Dashboard sections
1. Engagement Health
Shows rolling average engagement rates for the last 30, 60, and 90 days, a trend direction indicator, and a predicted engagement rate range for the client's next post.
| Metric | What it means |
|---|---|
| 30/60/90-Day Avg ER | Mean engagement rate across all posts in each window. Use the 90-day average as the stable baseline; the 30-day average reflects recent momentum. |
| Trend Direction | IMPROVING = 30d avg is more than 3% above the 90d average. DECLINING = more than 3% below. STABLE = within ±3%. |
| Predicted Next Post | A range based on the last 14 days of posts ±20%. This is not a guarantee — it reflects the trajectory if recent patterns continue. |
2. Content Type Performance
Compares average and median engagement rates across post formats: Images, Videos, Reels, Carousels. Ranked from highest to lowest performing.
Use this section to understand which format is driving the most engagement for this specific account — not what works generally, but what works for them.
3. Content Mix Forecast
Shows the client's current format mix (e.g. 40% Images, 35% Reels, 25% Carousels) and projects what their average engagement rate will be over the next 30, 60, and 90 days if they maintain that mix.
A second scenario is shown: what happens if they shift 20% of their output toward their highest-performing format. The delta percentage shows how much improvement this could unlock.
This section directly answers: "If our client posts more Reels and fewer Images, how does that affect their performance over the next quarter?"
4. Best Time to Post
Two data sources are combined to identify optimal posting windows:
- Engagement heatmap — a 7×24 grid showing which day/hour combinations have historically generated the highest engagement for posts published at those times.
- Audience activity (Instagram only) — when the client's actual followers are online, pulled directly from Instagram's API.
The Timezone Overlay converts the top UTC posting windows into local times for the client's top audience countries — so if a peak window is 8pm UTC and 40% of the audience is in Australia, it shows that as 6am AEST the following morning.
Note on algorithm timing: Social media algorithms distribute content based on individual user behaviour, not just posting time. Timing is one of several factors — not a guarantee. Use these windows as a starting point, not a hard rule.
5. Follower Growth Forecast
Uses the historical follower count data captured across report snapshots to calculate a daily growth rate, then projects that rate forward 30, 60, and 90 days.
This is a linear projection — it assumes the current growth rate continues unchanged. Viral posts or major campaigns will cause the actual number to diverge from the forecast.
6. Virality Analysis
This section classifies the client's past posts into three virality tiers based on how they performed relative to the account's own baseline — not against industry averages.
Virality tiers defined
| Tier | ER Multiplier | Reach Multiplier | Plain English |
|---|---|---|---|
| NANO | ≥ 2× account average ER | ≥ 1.5× follower count | Content resonating beyond the core audience. Not viral to the world, but meaningfully viral within their own community. |
| MICRO | ≥ 4× account average ER | ≥ 4× follower count | Breaking into broader discovery. The algorithm is picking this up and showing it to non-followers at scale. |
| MACRO | ≥ 8× account average ER | ≥ 8× follower count | Genuinely breaking the algorithm. Exceptionally rare. These posts set a new benchmark for the account. |
How the baseline is calculated
The baseline engagement rate is the rolling 90-day average ER across all the account's posts. This updates as new data comes in, so the threshold adjusts with the account's growth.
Noise filter
A post must reach a minimum absolute audience to qualify for any tier. This is calculated as max(500, follower_count × 10%). This prevents a post accidentally qualifying as "viral" simply because it was shared a few times on a very small account.
Both factors must be met
A post must meet both the ER multiplier and the reach multiplier to achieve a tier. High engagement alone without reach is not virality — it needs distribution. High reach without engagement is not virality — it's just impressions.
7. Competitor Benchmark
If competitors have been added and scraped in the Competitor Intelligence section, this panel shows how the client's average engagement rate compares to their competitors' top post engagement rates.
A positive delta means the client is outperforming their competitors. A negative delta means there is room to improve.
Engagement rate — how it's calculated
| Platform | Formula |
|---|---|
| (Likes + Comments + Shares + Saves) ÷ Follower Count | |
| TikTok | (Likes + Comments + Shares + Saves) ÷ Views (Plays) |
Using follower count as the denominator for Instagram gives a consistent measure of how well content reaches and resonates with the existing audience. Using plays for TikTok reflects the platform's view-first distribution model.
Frequently asked questions
Why are my predictions showing all zeros?
This usually means the platform is still in sandbox/development mode and returning mock data. The prediction engine only captures real data — it will not store mock data. Once both Instagram and TikTok apps are approved for production, real data will begin flowing in after the next report generation.
How long until predictions are meaningful?
A minimum of 30 days of post data across at least 10–15 posts gives the engine enough signal to produce useful results. With 90 days of data the predictions become significantly more reliable, particularly the timing heatmap and virality analysis.
Why do I see "Not enough data" for the heatmap?
The heatmap requires at least 2 posts published in the same day/hour slot to show a reliable ER for that window. Early on, most cells will be empty. This fills in naturally over time.
Can I run predictions for just Instagram or just TikTok?
Yes — use the platform selector at the top of the page to filter to a single platform. This is useful when the two platforms have very different posting strategies.
Does the system learn over time?
Phase 1 is a statistical engine — it recalculates on every page load from the stored post data. There is no machine learning in Phase 1. As more data accumulates, the statistics become more accurate. A future AI-powered phase will layer natural language insights and richer pattern recognition on top of this foundation.