Analysis to action
Archeus turns business questions into sandboxed queries, confirms anomalies against cohorts, and drafts the decision memo — with the methodology riding along, never cropped out.
Retention — Q2→Q3 self-serve
Question → Query → Readout
Generated query
“Why did self-serve churn spike at week 6?”
-- confidence 0.94 · owner review
SELECT cohort, week, retention
FROM retention_weekly
WHERE segment ='self_serve'
Monday briefing drafted
Ready for Slack · data-owner gated
Real workspace preview — query, cohort evidence, and the briefing gate in one view.
Built for teams that review before they ship
In action
A real working view — reviewable before anything ships.
Plain English becomes a sandboxed query — confidence, caveats, and the data owner's gate travel with it.
How it works
A practical, reviewable workflow—not a black box.
The business question, the audience, and the threshold for action are pinned before any data is touched — dashboards show what changed; framing decides what matters.
Prompt
“Frame the decision: why did self-serve churn spike at week 6 in Q2, and what would change our Q4 plan?”
Pipeline
Define the decision
Re-baseline or intervene
Set audience + threshold
VP Growth · −5pp = act
Name the data owner
Analytics guild validates
Verifiable output
Plain English becomes a sandboxed query — confidence, caveats, and the data owner's validation attached for whoever checks the work.
Prompt
“Turn the question into a sandboxed query with confidence and caveats stated.”
Pipeline
Generate SQL
Confidence 0.94
Attach caveats
Small-n · tier drift
Run sandboxed
1.8s · read-only
Verifiable output
SELECT start_month, week_n, retention_pct
FROM analytics.retention_weekly
-- confidence 0.94 · owner review pending
Anomalies are confirmed against cohorts and root causes are ranked by evidence weight — noise explicitly dropped, not summarized.
Prompt
“Confirm the anomaly across cohorts and rank root causes by evidence.”
Pipeline
Compare cohorts
Self-serve −14pp · sales-led +1pp
Isolate the variable
Onboarding checklist removal
Weight the causes
71% · 22% · noise
Verifiable output
Facts, interpretation, and next steps are separated into a decision memo — footnotes carry the assumptions projections make.
Prompt
“Turn the numbers into a decision memo — facts, interpretation, and next steps separated.”
Pipeline
Separate fact vs read
Two-column memo
Footnote assumptions
No pricing change assumed
Name the next step
Hold fix · re-baseline Dec
Verifiable output
The Monday briefing assembles itself from the week's runs — and ships to Slack or email only through a human approval gate.
Prompt
“Draft Monday's briefing from the week's runs and hold it for human approval.”
Pipeline
Assemble from run IDs
W42-01 · W42-03
Human approval gate
Held until signed off
Ship on cadence
Slack + Email · Mon 8 AM
Verifiable output
Capabilities
Analysts and executives ask in plain English — the query, confidence, and caveats stay attached for whoever needs to check the work.
Every run is sandboxed, read-only, and logged — analysts keep their governed layer.
Caveats ride along with every chart — confidence, sample size, and assumptions never get cropped out of the story.
Confidence score on every generated query
Small-sample flags surface automatically
One analysis becomes the query, the chart, and the Monday briefing — assembled from run IDs and shipped through a human approval gate.
The briefing cites its run IDs — anyone can open the query behind any sentence.
Monday briefing — anatomy
Human checkHeadline
Churn anomaly resolved — 61% recovered.
Evidence
W42-03 · cohort comparison, validated.
Next step
Hold the fix; re-baseline in December.
Where it helps
Start with the workflow you already repeat — the pipeline maps to it, not the other way around.
A focused readout for the recurring operating meeting — findings with run IDs, not screenshots.
Why the metric moved, isolated against cohorts and baselines — before anyone assigns a conclusion.
The weekly decision memo, drafted from validated analysis and gated by a human send.
QBR and board charts with footnoted projections — every assumption stated on the slide.
Why teams switch
A dashboard shows what changed, not what to check next.
The task connects a business question, the evidence, and a drafted next step.
Confidence is implied by a polished chart.
Confidence 0.94, small-n flags, and tier-drift caveats ride along with the answer.
The analyst rewrites the same findings for three audiences.
One analysis produces the query, the memo, and the Monday briefing — cited to run IDs.
The weekly update is assembled by hand, late Sunday.
The briefing drafts itself from the week's runs and ships through your sign-off.
Enterprise Standards
Customer prompts, uploaded business records, connected databases, and generated outputs are strictly ephemeral and never used to train base or commercial foundation models.
Automations, code evaluation, and browser executions occur in isolated, scoped environments with strict egress policies and encrypted secrets management.
Consequential actions—such as customer communications, database writes, or live deployments—require deliberate human confirmation with comprehensive audit logs.
Questions
Start with one real workflow, keep people in control, and build from what works.