Experiments
Test tone and pacing changes on real users, with a proper significance test.
Define an experiment with two variants, each a set of engine settings, and Humane assigns users and applies the right overlay on every call. Results compare the variants with lift and a two-proportion z-test.
Experiment · warmer_check_ins
empathy_weighting: 70
7-day return
38%
empathy_weighting: 85
7-day return
41%
Find out whether warmer is better, or just slower
Experiments let you change how the AI behaves for half your users without branching your code. Each response says which variant the user is in, and the experiment's start and end fire webhooks.
- Variants A and B, each a config overlay such as empathy or pacing
- The assignment comes back on every response under experiments
- Comparison with lift and a two-proportion z-test p-value
- experiment.started and experiment.concluded events
Behavior Score · last 7 days
- retention_deltaweight 30%
- conversation_depthweight 25%
- proactive_response_rateweight 20%
- abandonment_inverseweight 15%
- sentiment_trendweight 10%
Know which variant a user saw
r = client.process(user_id="user_42", message=text)
for e in r.experiments:
print(e.experiment_name, e.variant, e.config_overlay)
# warmer_check_ins b {'empathy_weighting': 85}Experiment questions
Something else? Ask us.
Works well with
Tone and context
Before your model runs, Humane assembles its prompt from the user's state, relationship, recent conversation, relevant memories and open tasks, scoring and trimming each section to fit a budget. It also sets the tone, empathy, formality and length the reply should take.
ExploreAnalytics and cohorts
The Behavior Score is a 0–100 composite of retention, conversation depth, response to proactive outreach, abandonment and sentiment trend. Cohort views show retention and average mood by month of first contact.
ExploreTiming engine
Humane recommends how long to wait before each reply, based on the user's mood, energy, the intensity of their message and the pace of the conversation. Every delay is broken into labelled contributors, so you can see and chart why.
ExploreEvery AI should know who it's talking to.
The safety and human-oversight layer for emotional AI: user state across sessions, risk alerts, safety gates and human takeover, with every decision explained.
