ADOPTION
RATE.

Weekly Adoption Rate =

weekly_active_eligible_users
--------------------------------
eligible_users
× 100%

Example:

eligible users = 1,000
weekly active eligible users = 420

Weekly Adoption Rate = 42%

active_user =
  has ≥ 1 meaningful_ai_task_completed
  within measurement window

OPTIONAL stronger rule:
  has ≥ N meaningful tasks
  OR ≥ X active days
  depending on product usage cadence

Possible habitual user rule:

user is habitual if:

  active in ≥ 3 of last 4 weeks
  AND
  completed ≥ 5 meaningful tasks
  AND
  at least 1 task accepted/successful

Then:

Habitual Adoption Rate =

habitual eligible users
------------------------
eligible users

W4 retention =
  cohort users active in week 4
  / users activated in cohort week

Activation cohort   W0     W1     W2     W3     W4
----------------------------------------------------
2026-07-06         100%    68%    57%    51%    48%
2026-07-13         100%    73%    64%    58%    55%
2026-07-20         100%    79%    72%    67%    63%
2026-07-27         100%    81%    75%    70%    ...

Interpretation:
  newer onboarding / product version
  may be producing better retention.

Need:
  compare cohort composition,
  not only raw percentages.

eligible_user if:

  account_active = true
  AND AI_access = true
  AND role in supported_roles
  AND at least one relevant use case
  AND not test/service/bot account
  AND policy allows usage
  AND user is within rollout cohort

optional:
  minimum work volume
  expected task frequency

Company:
  10,000 employees

AI licenses:
  7,500

Eligible users:
  3,200

WAU:
  1,600

If denominator = employees:
  Adoption = 16%

If denominator = licenses:
  Adoption = 21.3%

If denominator = eligible users:
  Adoption = 50%

All three are mathematically correct.

Only one answers:
  “What share of people who can reasonably use this AI
   are actually using it weekly?”

time_to_activation =
  first_successful_ai_task_at
  − eligibility_start_at

Track:
  median
  p75
  p90

Example:
  median = 1.2 days
  p90 = 12 days

This suggests:
  most users activate quickly,
  but a significant tail struggles.

100 weekly active users

1 power user:
  800 tasks

99 users:
  2 tasks each

Average:
  9.98 tasks/user

Median:
  2 tasks/user

Conclusion:
  average suggests heavy adoption,
  median shows most users barely use it.

Always inspect:
  median
  p25/p75
  power-user concentration

Potential company time saved =

eligible users
× adoption rate
× average eligible task frequency
× average human active time saved per task

Example:

1,000 eligible users
× 40% weekly adoption
× 8 tasks/week
× 6 min saved/task
= 19,200 human minutes/week
= 320 hours/week

But only if:
  task frequency,
  time savings,
  quality
  are empirically measured.

Metric:
  Weekly Adoption Rate

Purpose:
  measure broad weekly usage
  among users for whom AI is relevant and available.

Numerator:
  unique eligible users with
  ≥1 meaningful AI task completed in week

Denominator:
  eligible users active in organization
  during that week

Exclusions:
  test accounts
  bots/service users
  disabled users
  roles without relevant AI use cases

Window:
  ISO week / rolling 7d
  (choose one and version it)

Mandatory slices:
  role/team
  use_case
  rollout cohort
  region/language
  product version

Guardrails:
  acceptance_rate
  time_savings
  support/error rate

Owner:
  AI Product Analytics

Version:
  adoption_rate_v1

{
  "event_id":"EVT-...",
  "event_type":"meaningful_ai_task_completed",
  "occurred_at":"...",
  "user_id":"USR-...",
  "task_id":"TASK-...",
  "use_case":"draft_marketing_post",
  "tenant_id":"...",
  "eligibility":{
    "ai_eligible":true,
    "eligible_reason":"role_use_case"
  },
  "outcome":{
    "success":true
  },
  "product_version":"2026.09.1",
  "traffic_type":"human"
}

eligible_user_daily(
  date
  tenant_id
  user_id
  account_active
  ai_access
  supported_role
  rollout_enabled
  policy_allowed
  relevant_use_case
  eligible
  eligibility_version
)

WITH eligible AS (
  SELECT DISTINCT user_id
  FROM eligible_user_daily
  WHERE date BETWEEN :start AND :end
    AND eligible = true
),
active AS (
  SELECT DISTINCT user_id
  FROM ai_events
  WHERE occurred_at >= :start
    AND occurred_at < :end
    AND event_type = 'meaningful_ai_task_completed'
    AND traffic_type = 'human'
)
SELECT
  COUNT(active.user_id) AS active_users,
  COUNT(eligible.user_id) AS eligible_users,
  COUNT(active.user_id)::float
    / NULLIF(COUNT(eligible.user_id), 0)
    AS adoption_rate
FROM eligible
LEFT JOIN active USING (user_id);

BAD OBJECTIVE:
  “Get adoption to 80%.”

BETTER:
  “Increase weekly adoption
   among eligible users
   from 42% → 55%

   while:
     acceptance_rate ≥ baseline
     time_savings > 0
     support incidents do not increase
     no risk-policy regression.”

Usage is a means.
Value is the outcome.

LOW ADOPTION
  ↓
Do users have access?
  no → rollout/access problem
  yes
  ↓
Are they eligible/relevant?
  no → denominator problem
  yes
  ↓
Do they know the use case?
  no → discovery/training
  yes
  ↓
Do they attempt first task?
  no → UX/integration/friction
  yes
  ↓
Do they reach first success?
  no → quality/tool failure
  yes
  ↓
Do they return?
  no → weak recurring value / habit / workflow fit
  yes
  ↓
Is usage concentrated in few users?
  yes → breadth problem
  no → adoption broadly healthy

Then inspect:
  Acceptance
  Coverage
  Time Savings
  Support/Error
  Cohorts

DATA

1. user_id
2. eligibility flag
3. eligibility version
4. meaningful_ai_task_completed
5. use_case
6. occurred_at
7. traffic_type
8. product version

METRICS

1. eligible users
2. activated users
3. activation rate
4. weekly active eligible users
5. weekly adoption rate
6. W4 retention
7. median meaningful tasks / WAU
8. top-10% usage share

SLICES

1. role/team
2. use_case
3. rollout cohort
4. product version

GUARDRAILS

1. acceptance rate
2. time savings
3. support/error rate

You are designing Adoption Rate analytics for an AI product.

1. Define the target user population.
2. Separate:
   access,
   eligibility,
   activation,
   active usage,
   habitual adoption,
   retention.
3. Do not use licenses or logins as adoption.
4. Define one meaningful usage event.
5. Choose DAU/WAU/MAU based on natural task cadence.
6. Specify:
   numerator,
   denominator,
   exclusions,
   measurement window,
   cohort dimensions,
   definition version.
7. Exclude:
   bots,
   service accounts,
   test traffic,
   disabled users,
   irrelevant roles.
8. Show:
   activation rate,
   adoption rate,
   retention,
   median task frequency,
   usage concentration.
9. Segment by:
   role/team,
   use case,
   rollout cohort,
   product version,
   region/language where relevant.
10. Pair adoption with:
    Acceptance Rate,
    Coverage,
    Time Savings
    and operational guardrails.
11. Treat survey data as complementary,
    not as authoritative behavioral measurement.
12. Compute authoritative rates deterministically
    in analytics SQL/code, not with an LLM.

ALL ACCOUNTS
  ↓
AI ACCESS?
  ↓
LICENSED / ENABLED
  ↓
RELEVANT ROLE / USE CASE?
POLICY ALLOWED?
REAL WORK EXISTS?
  ↓
ELIGIBLE USERS
  ↓
FIRST MEANINGFUL TASK ATTEMPT
  ↓
FIRST SUCCESS
  ↓
ACTIVATED USERS
  ↓
RETURN IN PERIOD?
  ↓
DAU / WAU / MAU
  ↓
ADOPTION RATE

active eligible users
---------------------
eligible users

  ↓
RETURN ACROSS PERIODS?
  ↓
RETENTION
  W1
  W4
  W8

  ↓
REPEATED MEANINGFUL USAGE?
  ↓
HABITUAL ADOPTION

  ↓
HOW BROAD?

cohorts
teams
roles
use cases
regions

  ↓
HOW DEEP?

median tasks / active user
use-case breadth
coverage

  ↓
IS IT ACTUALLY GOOD?

Acceptance Rate
Time Savings
Coverage
Support/Error
Quality Evals

═══════════════════════════════════

DO NOT COUNT:

LICENSE
as ADOPTION

LOGIN
as ACTIVE USE

PROMPTS
as USERS

BOT TRAFFIC
as HUMAN ADOPTION

ONE-TIME TRY
as HABIT

FORCED WORKFLOW
as VALUE

═══════════════════════════════════

THE CENTRAL FORMULA:

WEEKLY ADOPTION RATE =

WEEKLY ACTIVE
ELIGIBLE USERS
──────────────
ELIGIBLE USERS

BUT THE CENTRAL QUESTION IS:

“ARE THE PEOPLE
WHO COULD REALISTICALLY
BENEFIT FROM THIS AI
RETURNING TO USE IT
IN REAL WORK —

BECAUSE IT IS
ACTUALLY USEFUL?”

THAT IS ADOPTION.