Act · Margdarshak मार्गदर्शक Available now
The right nudge, to the right manager, with a holdout that proves it
A hypothesis in plain English resolves to a precise cohort from behavioural signals. Python decides who is eligible and who is held back; the model writes one short, grounded sentence. At the readout date the two arms are compared and the lift is reported with a confidence interval — or withheld, if fewer than three people would make it identifiable.
- Delivered where people already are — Teams, Slack, email — always refusable, frequency-capped
- Lawful basis recorded per purpose, per person; the consent gate cannot be bypassed
- Grounded in your own documents, not generic advice
- Randomised 20% holdout assigned before anything sends; pre-registered readout date
- Outcomes view: cohort → delivered → acknowledged → behaviour changed → follow-ups
Illustrative · synthetic demo tenant
Nudged · 80%
51%
gave recognition in 30 days
Held back · 20%
16%
gave recognition in 30 days
+35.5 pp 95% CI +18.9 … +47.0
GROUNDED IN · Recognition and Rewards policy · eligible cohort 190 · randomised before send
Part of the loop
Every door installs Sangam underneath
Connect
Sangam सङ्गम
Connectors, one person directory, consent ledger, lineage. Present in every edition.
Answer
Sutradhar सूत्रधार
Policy and helpdesk copilot for every employee. Quotes the clause.
Ask
Gargi गार्गी
Text-to-insight for HR and finance leaders: governed metrics, detectors, signals.
Act
Margdarshak मार्गदर्शक
Signals become the right nudge to the right manager, with a holdout that proves it.
Measure
Phala फल
Outcomes ledger: what it was worth, with the evidence class on every line.