Every completed run produces a Decision Brief and a results deck you can present. These are the decks for all 12 decision packs, generated from bundled sample data and left exactly as the generator wrote them.
Marketing measurement
How much revenue does each paid channel drive, and what is its return on ad spend?
Slide 1 of 18 · Cover
Marketing Mix Modeling
Slides are shown here as pictures. In the .pptx, text, tiles and tables are editable; charts are images. Open this slide full size (opens in a new tab)
This is the cover of the Marketing Mix Modeling deck. Every deck can be downloaded as a .pptx from the cards below.
Each card shows the deck's cover and the headline its run produced. Open it in the viewer above, or download the PowerPoint file.
Paid media drove 52.7% of revenue, and TikTok returns the most per dollar (4.28x)
Key findings
Search and Email share the most conversion credit: 19.9% and 19.8%
Key findings
The intervention in US-W raised the outcome by about $198 per period (95% range $190 to $206)
Key findings
A geo test in US-SW, against 5 matched controls, can detect a 2% lift
Key findings
Your 796 customers are worth $4.82M in expected revenue (95% range $4.72M to $4.91M)
Key findings
64 of 513 active customers are at risk of churning in the next 6 months; SMB churns fastest
Key findings
A +10% price change would move units sold by -10.5% and revenue by -1.5%
Key findings
160 units expected next period (95% range 137 to 186); SKU-1001 is 69% of it
Key findings
In 30 days the series is forecast at about $66.8K (95% range $55.3K to $78.2K)
Key findings
3 days were flagged as unusual, but the 3 methods disagree (3, 0 and 19 flags)
Key findings
No clear effect: the change after 2026-02-01 can't be told apart from zero
Key findings
New button converts 10.5% of visitors against 7.6% for control: a +37.9% lift
Key findings
Every pack's deck follows the same order, so whoever reads it knows where to find the answer, the reasoning and the doubts.
The pack, the question it answers and the headline number.
The headline, the key numbers and findings, and what to do next, on one slide.
The question, the data and the method behind the answer.
A chart, and a column saying what it means and what to do, with a confidence chip where the numbers support one. Geo tests show the design on a US map.
The moves the findings support, from what to do now to what to test first and what to watch.
Agreement between the model specifications (for a geo test, how closely the controls match), the five verification checks, and every caveat the run raised.
The parameter estimates, the tables behind the charts where the pack has them, and how to verify a real run's signed model card.
Opens in PowerPoint and Google Slides. Text, tiles and tables are editable shapes; charts are images.
Nothing on these slides was laid out or worded by hand for this page. They are what the platform's deck generator produced, and the same generator builds the deck for every completed run.
Each deck was produced by the code that builds the results deck for every completed run, pointed at TriaCaf's bundled sample data.
Every number comes from the bundled sample data, and every slide says so.
Every sentence is written by fixed rules from the run's stored results. No AI model writes any of it, and nobody edited the files after they were generated.
These samples carry no model-card signature. Their last slide, How to reproduce and verify this run, still explains how a real run's signature is checked.
The slide images on this page are drawn from each .pptx by TriaCaf's slide previewer; PowerPoint and Google Slides may differ slightly. Download the .pptx for the file itself.
Google Slides has the deck's fonts built in. PowerPoint uses Space Grotesk and Manrope when they're installed (both are free on Google Fonts).
Every completed run produces a Decision Brief and a results deck, ready to download from the portal.