Plan inputs for fitness apps | Snowplow Signals

Shorten the next session for a user who keeps stopping early

A user has started three sessions this week and finished none of them, each one stopped around the middle. The plan has a fourth session queued at the same length, and the report that would show the pattern arrives after the week is over. By then the fourth session has been served and stopped too. The right move is a shorter session, ready the next time this user opens the app.

What Signals knows

sessions_started_7d: Sessions started in the last seven days
sessions_finished_7d: Sessions finished in the last seven days
days_since_finish: Days since they last finished a session

What your plan builder does

Shortens the next session when three of this week's starts have no finish, and returns to the planned length once a finish comes through. Which exercises make up the shorter session stays with the plan builder.

plan-builder.ts · 13 lines

from datetime import timedelta
from snowplow_signals import Attribute, Event, StreamAttributeGroup, user_id

workout_started = Event(name="workout_started", vendor="com.example", version="1-0-0")
workout_completed = Event(name="workout_completed", vendor="com.example", version="1-0-0")

sessions_started_7d = Attribute(
    name="sessions_started_7d",
    type="int32",
    events=[workout_started],
    aggregation="counter",
    period=timedelta(days=7),
    default_value=0,
)

sessions_finished_7d = Attribute(
    name="sessions_finished_7d",
    type="int32",
    events=[workout_completed],
    aggregation="counter",
    period=timedelta(days=7),
    default_value=0,
)

days_since_finish = Attribute(
    name="days_since_finish",
    type="double",
    events=[workout_completed],
    aggregation="time_since_last",
    time_unit="d",
)

# Users are logged in to train, so the group keys on the user id.
fit_plans_abandoned = StreamAttributeGroup(
    name="fit_plans_abandoned",
    version=1,
    attribute_key=user_id,
    owner="you@example.com",
    attributes=[sessions_started_7d, sessions_finished_7d, days_since_finish],
)

// In the service that builds the user's next session
const user = await signals.getServiceAttributes({
  name: "fit_plans_abandoned",
  attribute_key: "user_id",
  identifier: userId,
});
// { sessions_started_7d: 3, sessions_finished_7d: 0, days_since_finish: 9.4 }

const unfinished = user.sessions_started_7d - user.sessions_finished_7d;

if (unfinished >= 3) {
  planBuilder.next({ userId, length: "short" });
}

Offer the next level to a user who finishes every session early

A user has finished every session they started this month, and each one took less time than the plan allowed for it. The program moves users up a level on a fixed week, so a user who is ready in week two waits until week four. Two more weeks of sessions that feel easy give them less reason to open the app. The time to offer the next level is while they are finishing early.

What Signals knows

sessions_started_30d: Sessions started in the last 30 days
sessions_finished_30d: Sessions finished in the last 30 days
finish_seconds_mean_30d: Average length of a finished session in the last 30 days, in seconds

What your plan builder does

Offers the next level once every start this month has a finish and the average finished session runs under the early-finish mark the plan builder sets for that level. The user takes it or stays where they are.

plan-builder.ts · 14 lines

from datetime import timedelta
from snowplow_signals import Attribute, Event, EventProperty, StreamAttributeGroup, user_id

workout_started = Event(name="workout_started", vendor="com.example", version="1-0-0")
workout_completed = Event(name="workout_completed", vendor="com.example", version="1-0-0")

sessions_started_30d = Attribute(
    name="sessions_started_30d",
    type="int32",
    events=[workout_started],
    aggregation="counter",
    period=timedelta(days=30),
    default_value=0,
)

sessions_finished_30d = Attribute(
    name="sessions_finished_30d",
    type="int32",
    events=[workout_completed],
    aggregation="counter",
    period=timedelta(days=30),
    default_value=0,
)

finish_seconds_mean_30d = Attribute(
    name="finish_seconds_mean_30d",
    type="double",
    events=[workout_completed],
    aggregation="mean",
    property=EventProperty(vendor="com.example", name="workout_completed", major_version=1, path="duration_seconds"),
    period=timedelta(days=30),
)

fit_plans_ahead = StreamAttributeGroup(
    name="fit_plans_ahead",
    version=1,
    attribute_key=user_id,
    owner="you@example.com",
    attributes=[sessions_started_30d, sessions_finished_30d, finish_seconds_mean_30d],
)

// In the service that builds the user's next session
const user = await signals.getServiceAttributes({
  name: "fit_plans_ahead",
  attribute_key: "user_id",
  identifier: userId,
});
// { sessions_started_30d: 11, sessions_finished_30d: 11, finish_seconds_mean_30d: 1420 }

const allFinished = user.sessions_started_30d > 0 && user.sessions_finished_30d >= user.sessions_started_30d;
const early = user.finish_seconds_mean_30d <= plan.earlyFinishSeconds;

if (allFinished && early) {
  planBuilder.offer({ userId, level: plan.nextLevel });
}

Fit the plan to the days they train

A user has trained on Monday, Wednesday and Friday every week this month, and the plan they are on schedules five sessions a week. Two of those are missed every week, and each one shows up as a skip. The schedule was set at signup from the days the user said they would train, and it changes when the user goes back into settings, which lags behind the habit by weeks. The days they train are in the sessions they start.

What Signals knows

top_training_day_30d: The weekday they started the most sessions on in the last 30 days
sessions_by_day_30d: Sessions started on each weekday in the last 30 days

What your plan builder does

Rebuilds the week around the days with a session started in most weeks this month, and moves the sessions scheduled on the other days onto them or drops them. The user sees a plan that matches the days they train, and the two skips a week stop appearing.

plan-builder.ts · 15 lines

from datetime import timedelta
from snowplow_signals import Attribute, Event, EventProperty, StreamAttributeGroup, user_id

workout_started = Event(name="workout_started", vendor="com.example", version="1-0-0")
day_of_week = EventProperty(vendor="com.example", name="workout_started", major_version=1, path="day_of_week")

top_training_day_30d = Attribute(
    name="top_training_day_30d",
    type="string",
    events=[workout_started],
    aggregation="most_frequent",
    property=day_of_week,
    period=timedelta(days=30),
)

sessions_by_day_30d = Attribute(
    name="sessions_by_day_30d",
    type="dict",
    events=[workout_started],
    aggregation="category_count",
    property=day_of_week,
    period=timedelta(days=30),
)

fit_plans_days = StreamAttributeGroup(
    name="fit_plans_days",
    version=1,
    attribute_key=user_id,
    owner="you@example.com",
    attributes=[top_training_day_30d, sessions_by_day_30d],
)

// In the service that lays out the user's week
const user = await signals.getServiceAttributes({
  name: "fit_plans_days",
  attribute_key: "user_id",
  identifier: userId,
});
// { top_training_day_30d: "monday", sessions_by_day_30d: { monday: 4, wednesday: 4, friday: 3 } }

const trainingDays = Object.entries(user.sessions_by_day_30d)
  .filter(([, starts]) => starts >= 3)
  .map(([day]) => day);

if (trainingDays.length > 0 && trainingDays.length < plan.daysPerWeek) {
  planBuilder.rebuild({ userId, days: trainingDays });
}

Swap out the type of session a user keeps skipping

A user's plan mixes runs, strength and mobility, and this month they have started every run and no strength session. The plan keeps scheduling strength, each one is logged as a skip, and the review that would change the mix comes at the end of the program block. Until then the user opens the app to a session they are going to skip. The sessions they do start say which type to keep.

What Signals knows

types_started_30d: Sessions started of each type in the last 30 days
top_type_finished_30d: The type of session they finished most in the last 30 days

What your plan builder does

Replaces a type with no start this month with more of the type the user finishes most, and keeps the swap until a session of the dropped type is started. Whether and when the type returns is the plan builder's rule.

plan-builder.ts · 13 lines

from datetime import timedelta
from snowplow_signals import Attribute, Event, EventProperty, StreamAttributeGroup, user_id

workout_started = Event(name="workout_started", vendor="com.example", version="1-0-0")
workout_completed = Event(name="workout_completed", vendor="com.example", version="1-0-0")

types_started_30d = Attribute(
    name="types_started_30d",
    type="dict",
    events=[workout_started],
    aggregation="category_count",
    property=EventProperty(vendor="com.example", name="workout_started", major_version=1, path="workout_type"),
    period=timedelta(days=30),
)

top_type_finished_30d = Attribute(
    name="top_type_finished_30d",
    type="string",
    events=[workout_completed],
    aggregation="most_frequent",
    property=EventProperty(vendor="com.example", name="workout_completed", major_version=1, path="workout_type"),
    period=timedelta(days=30),
)

fit_plans_types = StreamAttributeGroup(
    name="fit_plans_types",
    version=1,
    attribute_key=user_id,
    owner="you@example.com",
    attributes=[types_started_30d, top_type_finished_30d],
)

// In the service that builds the user's next session
const user = await signals.getServiceAttributes({
  name: "fit_plans_types",
  attribute_key: "user_id",
  identifier: userId,
});
// { types_started_30d: { run: 9, mobility: 4 }, top_type_finished_30d: "run" }

const skipped = plan.types.filter((type) => !(type in user.types_started_30d));

if (skipped.length > 0 && user.top_type_finished_30d) {
  planBuilder.swap({ userId, from: skipped, to: user.top_type_finished_30d });
}

A plan that fits the week the user just had.

Get more workouts finished by building the next session from this week.

Shorten the next session for a user who keeps stopping early

Offer the next level to a user who finishes every session early

Fit the plan to the days they train

Swap out the type of session a user keeps skipping

Try it on your own site.