Skip to content
Sightrz
Sightrz AI

Your marketing analyst, on duty daily

Most "AI analytics" is a chat box bolted onto a chart. Sightrz AI runs the statistics in the database, then uses a model only for the part models are good at: explaining what happened in a sentence a CMO can act on.

How it works

Detection is maths. Narration is the model.

  1. Baseline

    Every metric on every entity carries a rolling 30-day mean and standard deviation, refreshed at load.

  2. Detect

    A z-score above the threshold, sustained over the minimum window, becomes an anomaly. Free, deterministic, and identical every time you look.

  3. Rank

    Anomalies are scored by spend at risk, not by how unusual they are. A 40% swing on a $200 campaign does not outrank a 9% swing on a $40,000 one.

  4. Narrate

    Only the top-ranked items reach Claude, with the numbers already computed. The model writes the explanation and the recommended action.

This is also the cost story. Statistics cost nothing to run, so the AI layer is part of every build rather than an add-on — and whether the model calls land on your Anthropic account or ours is your choice.

0Health
Account health score

A single 0–100 score across efficiency, stability, coverage and freshness, so a glance tells you whether to dig in.

Daily executive summary06:00
Yesterday: $41,200 spent at 3.4× ROAS, 6% above the weekly mean. Meta carried the gain; TikTok fell 18% on creative fatigue. Action: refresh the top three TikTok creatives today.
Anomaly detection
CPA on "Ramadan — Conversions" is 3.2σ above its baseline for three consecutive days. Spend at risk: $8,900 per week.

What it does

Daily executive summary

Available now

One paragraph every morning: what moved, by how much, and what to do today. Delivered by email in your language.

Anomaly detection

Available now

Rolling z-score against each entity's own baseline, with the measurement window and confidence stated on every finding.

Ranked recommendations

Available now

Scale, pause, reallocate or refresh — ordered by expected impact, each carrying the number that justifies it.

Account health score

Available now

A single 0–100 score across efficiency, stability, coverage and freshness, so a glance tells you whether to dig in.

Ask anything

In development

A chat surface over your own normalised data, with the SQL it ran shown alongside the answer.

Forecasting

In development

Statistical spend and revenue projection with a confidence band, so budget conversations start from a number.

Guardrails

What it will not do

  • It never invents a number. Every figure in a summary comes from a query result passed into the prompt.
  • It never changes your campaigns. Recommendations are recommendations; the platform buttons stay yours.
  • It states its confidence and the window it measured. "High confidence, 34 days of data" beats a decisive-sounding guess.
  • Your data is not used to train anyone's model.
What the model receives
{
  "entity":     "campaign/23847_ramadan",
  "metric":     "cpa",
  "value":      182.40,
  "baseline":   118.90,
  "stddev":     19.85,
  "z_score":    3.20,
  "window":     "30d",
  "spend_at_risk": 8900,
  "currency":   "SAR"
}

The numbers are inputs, not outputs. The model writes the sentence; the query writes the figure.

Point the analyst at your own numbers

Tell us what you run. We will connect your channels in a demo workspace and walk you through your own data — not a canned dataset. Replies within one business day.