Agentic CEP vs Traditional CEP: What's Actually Different?
Agentic CEPs receive goals and execute autonomously. Traditional CEPs wait for instructions. Here's what that gap means for your lifecycle team.
Abhimanyu
·
Blogs
·
"Agentic" is now on every CEP vendor's homepage. MoEngage put it in their tagline. Braze built an Agent Console. Iterable shipped Nova Agent. If every platform is calling itself agentic, the reasonable question is whether the word means anything at all, or whether it is just a 2026 way of saying "we have AI."
The distinction is real. But it is architectural, not cosmetic, and it does not show up in a feature list. It shows up in the workflow your lifecycle team runs every day: who decides what campaigns to run, who builds the audience, who monitors the results, and what happens when something changes. For a full grounding in what an agentic customer engagement platform actually is, that article covers the definition in depth. This one focuses on the practical difference from traditional CEPs, with a concrete scenario that makes the gap visible.
By Q2 2026, approximately 65% of B2B marketing organizations had deployed at least one AI agent in their automation stack (Source: Blueshift, 2026). Most of those deployments are additions to existing rule-based platforms, not replacements. Which means most marketing teams are running a hybrid that preserves the fundamental constraint of the traditional model, even with AI features layered on top.
How a Traditional CEP Works (and Where It Stops Working)
A traditional CEP is built around a rule-based execution model. A marketer defines a trigger — a user event, a time elapsed, a segment condition — and connects it to a message. The platform fires the message when the trigger condition is met. That is the core loop, and it works well for predictable, stable lifecycle moments: onboarding sequences, weekly digests, anniversary campaigns.
The ceiling appears at scale. As the customer base grows and the number of behaviorally distinct segments increases, the number of journeys a team needs to configure grows with it. There is no upper bound on the combinations of user state, channel, message, and timing that matter. There is an upper bound on how many journeys one lifecycle team can build, test, and maintain.
Three structural limits define where traditional CEPs stop working:
The configuration bottleneck. Every journey requires human setup. Changing a segment means changing the rule. Adding a channel means adding a branch. Teams end up with a backlog of lifecycle moments they know matter but have not had time to build.
The reactive posture. Traditional platforms respond to what you tell them to watch. A churn signal your rules don't cover does not generate an alert. A campaign underperforming in one cohort does not trigger a reallocation unless someone is watching the dashboard and acts on it.
The data translation gap. Most best customer engagement platforms connect to event data, but turning that data into an actionable segment still requires SQL knowledge or an engineering ticket. The data and the execution layer are separate, with a human in the middle.
None of this is a flaw in traditional CEPs. It is an accurate description of what they were designed to do: execute what a human has specified, reliably, at scale. The problem is that the marketing work that matters most, identifying the right moment to reach the right person with the right message, was never the platform's job. It was always the team's job.
How an Agentic CEP Works Differently
An agentic CEP starts with an outcome, not an instruction. Instead of "send this message when this trigger fires," the input is "reduce activation drop-off for users who completed signup but haven't used the core feature in seven days." The platform determines what segments to build, what content to send, which channels to use, and when.
The architectural difference is in where the decision-making lives. In a traditional CEP, decisions live with the marketer. In an agentic CEP, decisions about execution live with the platform. The marketer sets goals and reviews output. The platform configures and monitors.
This produces four practical differences in how work happens:
Audience building without a ticket. Agentic platforms connect directly to behavioral data and build segments in natural language. "Users who completed onboarding but haven't triggered the core feature in seven days, excluding users already in an active campaign" is a query the platform resolves without an engineer.
Content generation as part of execution. The platform drafts message variants based on the goal and the audience's behavioral profile. The team reviews and approves. Writing copy from scratch is no longer the first step.
Behavioral anomaly detection. The platform runs continuous analysis in the background and surfaces shifts before they appear in a weekly review. A cohort that was converting at 18% last week and dropped to 11% this week shows up as an alert with a recommended response, not as a number waiting to be noticed.
Outcome-based adjustment. After a campaign runs, the platform incorporates the result into its model of what works for that audience. It does not wait for a human to read the report and decide what to change.
HubSpot's 2026 State of Marketing report found that 32.8% of marketers using AI tools save between 10 and 14 hours per week (Source: HubSpot State of Marketing, 2026). Agentic platforms produce that kind of time recovery specifically because the operational work, the configuration, the audience building, the monitoring, moves to the platform. The human time that remains is higher-order: goal-setting, creative review, strategic decisions.
“Right from insight to idea to actually seeing a metric being converted, happening in less than a week, was quite surreal to experience. ”
Arnav Grover, Co-founder, Rovia
“Right from insight to idea to actually seeing a metric being converted, happening in less than a week, was quite surreal to experience. ”
Arnav Grover, Co-founder, Rovia
“Right from insight to idea to actually seeing a metric being converted, happening in less than a week, was quite surreal to experience. ”
Arnav Grover, Co-founder, Rovia
Churn Prevention, Two Ways
The clearest way to see the difference is to walk through the same lifecycle problem on both architectures. Churn prevention is the right example: it is high-stakes, time-sensitive, and requires behavioral precision to work.
On a traditional CEP
Your data team flags that 30-day churn has increased two points over the last month. You decide to build a re-engagement campaign. The steps:
Submit a data request: pull all users who have been inactive for 14 or more days, exclude those already in an active journey, segment by acquisition channel. Wait three to five days for the query.
Build the campaign: write two message variants, decide on push versus email, set a send time based on past campaign performance.
Launch the campaign, monitor open and click rates manually over the next week.
Run a second campaign for non-openers. Repeat the configuration process.
After 30 days, pull a report. Attribute the churn change to the campaign or to other variables. Decide whether to iterate.
Total time from signal to campaign live: one to two weeks. Total human steps: eight to twelve, depending on the team's tooling.
On an agentic CEP
The platform surfaces the churn signal before you see it in a report. An alert: "30-day activation rate declining in the cohort that signed up in the last 45 days, specifically among users who haven't triggered Feature X. Recommended action: re-engagement sequence, push primary channel." The audience is already built. A draft message is ready for review.
Review the platform's audience definition and proposed message. Approve or edit.
Campaign runs. Platform monitors and detects that push is outperforming email for this cohort and adjusts channel weighting.
Thirty days later, the platform surfaces the outcome: conversion rate, cohort breakdown, and the recommended follow-up for users who did not convert.
Total time from signal to campaign live: under an hour. Total human steps: two to three.
The difference is not speed for its own sake. It is that the second approach covers lifecycle moments that the first approach would have missed, either because the signal was not on anyone's radar or because the team did not have the bandwidth to build the campaign before the window closed.
Agentic CEP vs Traditional CEP: Head-to-Head
Dimension | Traditional CEP | Agentic CEP |
|---|---|---|
Input model | Rules and triggers defined by the marketer | Business goals set by the marketer |
Audience building | Manual configuration or engineering request | Autonomous from behavioral data |
Content creation | Template + human copywriting | AI-generated drafts, human review |
Monitoring | Dashboard-based; human checks | Continuous background analysis with proactive alerts |
Response to change | Marketer updates the rule or journey | Platform adjusts based on outcome data |
Engineering dependency | Required for complex segments and integrations | Self-serve for lifecycle marketers |
Learning | A/B test results reported; human acts | Results incorporated into platform model |
Time from signal to campaign | Days to weeks | Hours |
Best for | Stable, predictable lifecycle moments | Dynamic, high-variation customer bases |
Which Teams Are Ready for an Agentic CEP?
Not every team is at the right stage for an agentic platform. The benefits compound with scale, data volume, and team size constraints. Smaller, earlier-stage programs get less out of agentic automation because the problem it solves, too much behavioral variation to cover manually, has not fully arrived yet.
Teams that are ready:
Your team is likely ready for an agentic CEP if lifecycle moments are being missed because there is no bandwidth to build the journeys, if campaign backlog is growing faster than the team can clear it, or if the gap between what the data could tell you and what the team can actually act on is widening. The signal is not team size. The signal is the gap between what you know matters and what you can actually do about it.
Teams that are not ready yet:
If your monthly active user base is small enough that one lifecycle marketer can realistically cover every high-value moment, an agentic platform adds capability you do not need yet. The other limiting factor is data infrastructure. Agentic systems connect directly to behavioral data in a warehouse. Without that connection, the platform cannot build the audiences or detect the anomalies that define agentic operation. Data readiness comes before platform capability.
The platform architecture question:
When evaluating whether a platform is genuinely agentic or agentic by label, ask where the decision-making lives. If a human has to configure every journey and the "AI" generates content inside that journey, the architecture is still rule-based with an AI assistant. If the platform takes a goal and determines the audience, content, and channel without human configuration, the architecture is agentic.
Sortment was built as an agentic system from the start. It connects to your data warehouse, reads behavioral signals continuously, and runs a multi-agent architecture where specialized agents own specific lifecycle goals: audience identification, content generation, anomaly detection, and outcome learning. The lifecycle team at a nonprofit engagement platform using Sortment runs over 2,000 campaigns a year. One person covers what would normally require three to five full-time team members, because the platform handles the operational work and the human reviews the decisions.
The entry model reflects the architecture. A 30-day proof of concept tied to one outcome you define — activation, retention, or conversion — runs on your live data before any long-term commitment. For a deeper look at the broader customer lifecycle marketing decisions that agentic platforms support, that guide covers the full strategy layer.
See what Sortment can do in 30 days
Pick one lifecycle goal, like activation, retention, or monetization and see how Sortment can achieves your business goals.
See what Sortment can do in 30 days
Pick one lifecycle goal, like activation, retention, or monetization and see how Sortment can achieves your business goals.
See what Sortment can do in 30 days
Pick one lifecycle goal, like activation, retention, or monetization and see how Sortment can achieves your business goals.
Frequently Asked Questions About Agentic CEPs vs Traditional CEPs
What is the main difference between an agentic CEP and a traditional CEP?
The main difference is where decision-making lives. A traditional CEP executes the rules and journeys a marketer configures. An agentic CEP takes a goal and determines the execution path, including audience, content, channel, and timing, without requiring a human to configure each step. The marketer sets objectives and reviews output rather than building every campaign from scratch.
Is Braze an agentic CEP?
Braze launched agentic features in April 2026, including BrazeAI Operator and Agent Console, which allow marketers to build campaigns through natural language and deploy per-user decisioning agents. These are meaningful additions to a platform originally built on rule-based Canvas orchestration. Whether Braze qualifies as a genuinely agentic CEP depends on whether those features are integrated into the core execution model or layered on top of it.
Can a traditional CEP be upgraded to become agentic?
Traditional platforms can add agentic capabilities through acquisitions or feature development. MoEngage acquired Aampe in 2026 to add per-user decisioning agents to its platform. The practical question is whether those agentic features are deeply integrated into the platform's execution model or bolted onto a rule-based architecture. Integration depth determines how effectively agentic and rule-based workflows work together.
Do I need a data warehouse to use an agentic CEP?
Most agentic CEPs connect directly to a data warehouse — Snowflake, BigQuery, or Redshift — to access the behavioral data required for autonomous audience building and anomaly detection. Without that connection, the platform cannot make the per-user decisions that define agentic operation. Data warehouse readiness is the most common prerequisite that determines how quickly a team sees results.
How long does the switch from a traditional CEP to an agentic one take?
Most platform migrations take six to eight weeks for the core transition and longer for full coverage across all lifecycle journeys. The most common approach is to connect the agentic platform for audience building and analysis while keeping delivery on the existing platform, then migrate delivery once the new platform is validated against live data. Running both in parallel reduces the risk of disrupting campaigns already in flight.
What happens to existing campaigns when switching to an agentic CEP?
Existing campaigns can continue running on the traditional platform during the transition. The standard migration approach is to start the agentic platform with new lifecycle moments, validate performance against your behavioral data, and then replicate high-priority existing journeys before switching delivery. Most teams do not migrate all campaigns at once; they prioritize by lifecycle stage and business impact.
Is agentic CEP technology mature enough to rely on in 2026?
Agentic CEP technology is in active production at companies with large consumer user bases. The category is maturing rapidly: MoEngage, Braze, and Iterable all shipped agentic features in 2026, and platforms built natively on agentic architecture have been in production for longer. The practical readiness question is not the technology. It is whether your data infrastructure and behavioral data are clean enough for the platform to make good decisions.
See also
5 Best Customer Engagement Platforms for 2026 (Reviewed)
5 Best Customer Engagement Platforms for 2026 (Reviewed)
5 Best Customer Engagement Platforms for 2026 (Reviewed)
The 5 best customer engagement platforms for US businesses — Sortment, Braze, HubSpot, Intercom & Salesforce — compared on features, channels, and pricing.
The 5 best customer engagement platforms for US businesses — Sortment, Braze, HubSpot, Intercom & Salesforce — compared on features, channels, and pricing.
See what Sortment can do for your goals.
See what Sortment can do for your goals.
Book a 30-minute call. We'll show you how the pilot works with your data and your stack.
Book a 30-minute call. We'll show you how the pilot works with your data and your stack.
AGENTS
CASE STUDIES
RESOURCES
AGENTS
CASE STUDIES
RESOURCES
"Agentic" is now on every CEP vendor's homepage. MoEngage put it in their tagline. Braze built an Agent Console. Iterable shipped Nova Agent. If every platform is calling itself agentic, the reasonable question is whether the word means anything at all, or whether it is just a 2026 way of saying "we have AI."
The distinction is real. But it is architectural, not cosmetic, and it does not show up in a feature list. It shows up in the workflow your lifecycle team runs every day: who decides what campaigns to run, who builds the audience, who monitors the results, and what happens when something changes. For a full grounding in what an agentic customer engagement platform actually is, that article covers the definition in depth. This one focuses on the practical difference from traditional CEPs, with a concrete scenario that makes the gap visible.
By Q2 2026, approximately 65% of B2B marketing organizations had deployed at least one AI agent in their automation stack (Source: Blueshift, 2026). Most of those deployments are additions to existing rule-based platforms, not replacements. Which means most marketing teams are running a hybrid that preserves the fundamental constraint of the traditional model, even with AI features layered on top.
How a Traditional CEP Works (and Where It Stops Working)
A traditional CEP is built around a rule-based execution model. A marketer defines a trigger — a user event, a time elapsed, a segment condition — and connects it to a message. The platform fires the message when the trigger condition is met. That is the core loop, and it works well for predictable, stable lifecycle moments: onboarding sequences, weekly digests, anniversary campaigns.
The ceiling appears at scale. As the customer base grows and the number of behaviorally distinct segments increases, the number of journeys a team needs to configure grows with it. There is no upper bound on the combinations of user state, channel, message, and timing that matter. There is an upper bound on how many journeys one lifecycle team can build, test, and maintain.
Three structural limits define where traditional CEPs stop working:
The configuration bottleneck. Every journey requires human setup. Changing a segment means changing the rule. Adding a channel means adding a branch. Teams end up with a backlog of lifecycle moments they know matter but have not had time to build.
The reactive posture. Traditional platforms respond to what you tell them to watch. A churn signal your rules don't cover does not generate an alert. A campaign underperforming in one cohort does not trigger a reallocation unless someone is watching the dashboard and acts on it.
The data translation gap. Most best customer engagement platforms connect to event data, but turning that data into an actionable segment still requires SQL knowledge or an engineering ticket. The data and the execution layer are separate, with a human in the middle.
None of this is a flaw in traditional CEPs. It is an accurate description of what they were designed to do: execute what a human has specified, reliably, at scale. The problem is that the marketing work that matters most, identifying the right moment to reach the right person with the right message, was never the platform's job. It was always the team's job.
How an Agentic CEP Works Differently
An agentic CEP starts with an outcome, not an instruction. Instead of "send this message when this trigger fires," the input is "reduce activation drop-off for users who completed signup but haven't used the core feature in seven days." The platform determines what segments to build, what content to send, which channels to use, and when.
The architectural difference is in where the decision-making lives. In a traditional CEP, decisions live with the marketer. In an agentic CEP, decisions about execution live with the platform. The marketer sets goals and reviews output. The platform configures and monitors.
This produces four practical differences in how work happens:
Audience building without a ticket. Agentic platforms connect directly to behavioral data and build segments in natural language. "Users who completed onboarding but haven't triggered the core feature in seven days, excluding users already in an active campaign" is a query the platform resolves without an engineer.
Content generation as part of execution. The platform drafts message variants based on the goal and the audience's behavioral profile. The team reviews and approves. Writing copy from scratch is no longer the first step.
Behavioral anomaly detection. The platform runs continuous analysis in the background and surfaces shifts before they appear in a weekly review. A cohort that was converting at 18% last week and dropped to 11% this week shows up as an alert with a recommended response, not as a number waiting to be noticed.
Outcome-based adjustment. After a campaign runs, the platform incorporates the result into its model of what works for that audience. It does not wait for a human to read the report and decide what to change.
HubSpot's 2026 State of Marketing report found that 32.8% of marketers using AI tools save between 10 and 14 hours per week (Source: HubSpot State of Marketing, 2026). Agentic platforms produce that kind of time recovery specifically because the operational work, the configuration, the audience building, the monitoring, moves to the platform. The human time that remains is higher-order: goal-setting, creative review, strategic decisions.
“Right from insight to idea to actually seeing a metric being converted, happening in less than a week, was quite surreal to experience. ”
Arnav Grover, Co-founder, Rovia
“Right from insight to idea to actually seeing a metric being converted, happening in less than a week, was quite surreal to experience. ”
Arnav Grover, Co-founder, Rovia
“Right from insight to idea to actually seeing a metric being converted, happening in less than a week, was quite surreal to experience. ”
Arnav Grover, Co-founder, Rovia
Churn Prevention, Two Ways
The clearest way to see the difference is to walk through the same lifecycle problem on both architectures. Churn prevention is the right example: it is high-stakes, time-sensitive, and requires behavioral precision to work.
On a traditional CEP
Your data team flags that 30-day churn has increased two points over the last month. You decide to build a re-engagement campaign. The steps:
Submit a data request: pull all users who have been inactive for 14 or more days, exclude those already in an active journey, segment by acquisition channel. Wait three to five days for the query.
Build the campaign: write two message variants, decide on push versus email, set a send time based on past campaign performance.
Launch the campaign, monitor open and click rates manually over the next week.
Run a second campaign for non-openers. Repeat the configuration process.
After 30 days, pull a report. Attribute the churn change to the campaign or to other variables. Decide whether to iterate.
Total time from signal to campaign live: one to two weeks. Total human steps: eight to twelve, depending on the team's tooling.
On an agentic CEP
The platform surfaces the churn signal before you see it in a report. An alert: "30-day activation rate declining in the cohort that signed up in the last 45 days, specifically among users who haven't triggered Feature X. Recommended action: re-engagement sequence, push primary channel." The audience is already built. A draft message is ready for review.
Review the platform's audience definition and proposed message. Approve or edit.
Campaign runs. Platform monitors and detects that push is outperforming email for this cohort and adjusts channel weighting.
Thirty days later, the platform surfaces the outcome: conversion rate, cohort breakdown, and the recommended follow-up for users who did not convert.
Total time from signal to campaign live: under an hour. Total human steps: two to three.
The difference is not speed for its own sake. It is that the second approach covers lifecycle moments that the first approach would have missed, either because the signal was not on anyone's radar or because the team did not have the bandwidth to build the campaign before the window closed.
Agentic CEP vs Traditional CEP: Head-to-Head
Dimension | Traditional CEP | Agentic CEP |
|---|---|---|
Input model | Rules and triggers defined by the marketer | Business goals set by the marketer |
Audience building | Manual configuration or engineering request | Autonomous from behavioral data |
Content creation | Template + human copywriting | AI-generated drafts, human review |
Monitoring | Dashboard-based; human checks | Continuous background analysis with proactive alerts |
Response to change | Marketer updates the rule or journey | Platform adjusts based on outcome data |
Engineering dependency | Required for complex segments and integrations | Self-serve for lifecycle marketers |
Learning | A/B test results reported; human acts | Results incorporated into platform model |
Time from signal to campaign | Days to weeks | Hours |
Best for | Stable, predictable lifecycle moments | Dynamic, high-variation customer bases |
Which Teams Are Ready for an Agentic CEP?
Not every team is at the right stage for an agentic platform. The benefits compound with scale, data volume, and team size constraints. Smaller, earlier-stage programs get less out of agentic automation because the problem it solves, too much behavioral variation to cover manually, has not fully arrived yet.
Teams that are ready:
Your team is likely ready for an agentic CEP if lifecycle moments are being missed because there is no bandwidth to build the journeys, if campaign backlog is growing faster than the team can clear it, or if the gap between what the data could tell you and what the team can actually act on is widening. The signal is not team size. The signal is the gap between what you know matters and what you can actually do about it.
Teams that are not ready yet:
If your monthly active user base is small enough that one lifecycle marketer can realistically cover every high-value moment, an agentic platform adds capability you do not need yet. The other limiting factor is data infrastructure. Agentic systems connect directly to behavioral data in a warehouse. Without that connection, the platform cannot build the audiences or detect the anomalies that define agentic operation. Data readiness comes before platform capability.
The platform architecture question:
When evaluating whether a platform is genuinely agentic or agentic by label, ask where the decision-making lives. If a human has to configure every journey and the "AI" generates content inside that journey, the architecture is still rule-based with an AI assistant. If the platform takes a goal and determines the audience, content, and channel without human configuration, the architecture is agentic.
Sortment was built as an agentic system from the start. It connects to your data warehouse, reads behavioral signals continuously, and runs a multi-agent architecture where specialized agents own specific lifecycle goals: audience identification, content generation, anomaly detection, and outcome learning. The lifecycle team at a nonprofit engagement platform using Sortment runs over 2,000 campaigns a year. One person covers what would normally require three to five full-time team members, because the platform handles the operational work and the human reviews the decisions.
The entry model reflects the architecture. A 30-day proof of concept tied to one outcome you define — activation, retention, or conversion — runs on your live data before any long-term commitment. For a deeper look at the broader customer lifecycle marketing decisions that agentic platforms support, that guide covers the full strategy layer.
See what Sortment can do in 30 days
Pick one lifecycle goal, like activation, retention, or monetization and see how Sortment can achieves your business goals.
See what Sortment can do in 30 days
Pick one lifecycle goal, like activation, retention, or monetization and see how Sortment can achieves your business goals.
See what Sortment can do in 30 days
Pick one lifecycle goal, like activation, retention, or monetization and see how Sortment can achieves your business goals.
Frequently Asked Questions About Agentic CEPs vs Traditional CEPs
What is the main difference between an agentic CEP and a traditional CEP?
The main difference is where decision-making lives. A traditional CEP executes the rules and journeys a marketer configures. An agentic CEP takes a goal and determines the execution path, including audience, content, channel, and timing, without requiring a human to configure each step. The marketer sets objectives and reviews output rather than building every campaign from scratch.
Is Braze an agentic CEP?
Braze launched agentic features in April 2026, including BrazeAI Operator and Agent Console, which allow marketers to build campaigns through natural language and deploy per-user decisioning agents. These are meaningful additions to a platform originally built on rule-based Canvas orchestration. Whether Braze qualifies as a genuinely agentic CEP depends on whether those features are integrated into the core execution model or layered on top of it.
Can a traditional CEP be upgraded to become agentic?
Traditional platforms can add agentic capabilities through acquisitions or feature development. MoEngage acquired Aampe in 2026 to add per-user decisioning agents to its platform. The practical question is whether those agentic features are deeply integrated into the platform's execution model or bolted onto a rule-based architecture. Integration depth determines how effectively agentic and rule-based workflows work together.
Do I need a data warehouse to use an agentic CEP?
Most agentic CEPs connect directly to a data warehouse — Snowflake, BigQuery, or Redshift — to access the behavioral data required for autonomous audience building and anomaly detection. Without that connection, the platform cannot make the per-user decisions that define agentic operation. Data warehouse readiness is the most common prerequisite that determines how quickly a team sees results.
How long does the switch from a traditional CEP to an agentic one take?
Most platform migrations take six to eight weeks for the core transition and longer for full coverage across all lifecycle journeys. The most common approach is to connect the agentic platform for audience building and analysis while keeping delivery on the existing platform, then migrate delivery once the new platform is validated against live data. Running both in parallel reduces the risk of disrupting campaigns already in flight.
What happens to existing campaigns when switching to an agentic CEP?
Existing campaigns can continue running on the traditional platform during the transition. The standard migration approach is to start the agentic platform with new lifecycle moments, validate performance against your behavioral data, and then replicate high-priority existing journeys before switching delivery. Most teams do not migrate all campaigns at once; they prioritize by lifecycle stage and business impact.
Is agentic CEP technology mature enough to rely on in 2026?
Agentic CEP technology is in active production at companies with large consumer user bases. The category is maturing rapidly: MoEngage, Braze, and Iterable all shipped agentic features in 2026, and platforms built natively on agentic architecture have been in production for longer. The practical readiness question is not the technology. It is whether your data infrastructure and behavioral data are clean enough for the platform to make good decisions.