Next best action for every member, across every location and banner
What should we say to this member, at this tier, who just did this thing, at this location? The answer changes constantly. We compute it continuously.
The problem
A portfolio operator with 4 banners, 16 lifecycle states, 24 loyalty tiers, and 200 locations has roughly 77,000 decision states to manage. For any given member touching any banner, what is the best next thing to send them — what offer, what message, what timing, with what content? In practice, the lifecycle team writes heuristic segment rules, runs them quarterly, and accepts that about half of recommendations are stale at any given time. Enterprise decision engines (Pega, SAS, FICO, Amperity, Salesforce Einstein, Adobe Sensei) ship powerful decisioning. Real-time personalization platforms ship variable swapping. Loyalty platforms ship platform-native decisioning. None of them maintain a unified member view across multiple banners, apply your state-by-state rules to the decisioning, and propose the right next step that respects each location's actual offer mix.
What success looks like
Every member, at any moment, gets the right next message at the right location across the right banner — based on what they have actually done, what tier they are at, what offers exist where they live, and what state rules apply. A high-value member who just lapsed at the gym gets a reactivation tied to that location's current offer. A new member at the cafe in their first week gets onboarding. A member who just hit a tier threshold gets a milestone message. Decisioning runs continuously rather than as a quarterly recalibration. The 51% stale rate falls below 10%. The lifecycle team reviews and approves decision logic; they stop hand-tuning segment rules in spreadsheets.
How most operators solve this today
Five categories already offer decisioning. None of them keep 77,000 decision states current across multiple banners with state-by-state rules:
Decision engines (Pega Customer Decision Hub, SAS Real-Time Decision Manager, FICO, Amperity, Salesforce Einstein Next Best Action, Adobe Sensei, Optimove, Bloomreach)
$25/user/month to $500,000+/year
Powerful platforms. They expect you to bring the unified member view and the rule set. That work is the unsolved part.
Real-time personalization (Dynamic Yield, Optimizely, Mutiny, Personyze, Adobe Target, Bloomreach Discovery)
$200 to $500,000+/year
Swaps variables at render time. They are useful at the surface; the decisioning behind them still has to be defined and maintained.
Loyalty platform decisioning (Yotpo, Annex Cloud, Antavo, Talon.One, LoyaltyLion)
$199 to $500,000+/year
Decides within their own platform. Cross-banner decisioning is not what they do.
In-house decisioning engineer + custom pipeline
$100-180k/year decisioning engineer + $130-220k/year engineer + ongoing maintenance
Custom Snowflake plus Python plus Airflow plus an ML model. Falls behind as the number of decision states grows.
Lifecycle manager running heuristic segment rules quarterly
$70-130k/year manager + platform fees
Quarterly batches at 77,000 decision states is why half of recommendations are stale at any given time.
What changes when this is an agent skill
Decisioning runs continuously against a unified member view. Every member is identified across banners (so the gym member and the cafe member are the same person if they are). For any given member at any moment, the system computes the right next step based on tier, recent behavior, time since last visit, available offers at the relevant location, and the rules of the state that location is in. The decision is made before the message ships, not after a quarterly batch refresh. When tier benefits change, decisions update. When a location adds an offer, decisions update. When state rules change, decisions update. The lifecycle team reviews decision logic at the rule level — what does a high-value lapsed member get? — rather than hand-tuning 77,000 segment combinations. Every decision is logged with the inputs that drove it and the recommendation that came out, so any specific message can be audited later.
Agents that include this skill
Skills live inside agent rentals. To get this skill in production, hire any of the agents below — context-tuning at onboarding is included in the first month.
Loyalty Member Journey Orchestration Agent
Decides when each member gets which offer — tier-transition timing, cross-location earn/redeem, offer dedup.
FAQ
- How is this different from Pega Customer Decision Hub or Salesforce Einstein Next Best Action?
- Those are powerful enterprise decision engines. They expect you to provide the unified member view and the rule set. That setup is the unsolved part for multi-banner operators. We do that work.
- How is this different from personalization platforms like Dynamic Yield or Mutiny?
- Personalization platforms swap variables at the surface. The decisioning behind them — what to show, why, to whom, when — is still your team's responsibility. We make those decisions continuously.
- How is this different from the decisioning that loyalty platforms like Antavo or Talon.One provide?
- Those decision within their own platform. Cross-banner decisioning, where a member action at the gym informs a message from the cafe, is what we add.
- Does this replace our existing platforms?
- No. It feeds them. Decisions land back in your loyalty platform, your journey orchestration platform, and your personalization layer through their existing APIs.
- What decision types are supported?
- Welcome, onboarding, recall, reactivation, win-back, milestone, tier progression, tier downgrade warning, upsell, cross-sell, referral request, churn prevention. Anything tied to a lifecycle state at a tier in a location.
- How are state rules applied?
- State rules and compliance constraints are encoded once and applied per decision. If a state restricts certain claims or requires certain disclosures, decisions that would violate those rules are filtered out before they reach a member.
- How is the decision logic reviewed?
- Your team reviews and approves logic at the rule level — what should a high-value lapsed member get? — rather than tuning 77,000 segment combinations by hand. Changes are versioned so you can roll back if a rule produces unintended results.
- How is history captured?
- Every decision is logged with the inputs (tier, behavior, location, offers available, rules applied) and the output (recommended action). You can audit any specific message that went out back to the decision that produced it.