AI Loyalty Management Platform
ReactorCX AI: an MCP server for your loyalty platform
What if your loyalty platform understood itself?
ReactorCX AI is the AI advisory layer of the ReactorCX enterprise loyalty platform. It connects approved AI models to the platform's configuration, rules, APIs and events through a Model Context Protocol (MCP) server, prepares configuration, integration, simulation and diagnostic work for review, and executes nothing without human approval. The deterministic engine runs only what was approved.
Reward the first in-store visit within 72 hours of a fuel fill.
Prepared change · awaiting approval
- Rule: visit-after-fuel · window 72h · reward: bonus points · segment: fuel members
- Projected cost: within budget cap · affected members: eligible fuel segment
- Simulation: no rule conflicts · no expiry overlap
Business intent to correct implementation
When AI connects to ReactorCX through the Model Context Protocol, it reads your program the way an engineer would: rule conditions, tier qualification logic, version history, earn rate structures. The platform was already built to be read that way. MCP did not require a new architecture. It required an open door.
See AI prepare a change on your own configuration.
See RCX AI on Your ConfigurationHow AI change stays inside enterprise control
Every row is an existing platform control. AI does not get a shortcut around any of them.
| Step of an AI-prepared change | Control applied | Same as a manual change? |
|---|---|---|
| Drafted from live configuration | Read access scoped by role and attribute; no write to production | Yes |
| Built as a program version | Stage environment; JSON export; version tracked | Yes |
| Reviewed | Plain-language summary, projected cost, affected members, rule conflicts flagged by pre-launch simulation | Adds simulation |
| Approved | Named approver; nothing executes without sign-off | Yes |
| Published | Publish/unpublish without platform restart | Yes |
| Logged | Execution log records what ran, when, why and who approved it | Yes |
| Reversed if needed | Unpublish or republish the prior version | Yes |
AI operates within the same guardrails as the platform itself. It cannot bypass approval workflows, create untracked changes, or violate access controls.
Deterministic engine. AI advisory layer.
Two distinct capabilities in one platform. The deterministic engine handles points calculations, tier qualifications, discount arbitration and promotion evaluation with mathematical precision and full audit trails. The AI advisory layer surfaces recommendations and prepares work. AI never touches the calculations. What executes is what was approved.
"Think of it as assisted driving, not self-driving, for loyalty. When a misconfigured rule can burn through millions in points before anyone catches it, you want human hands on the wheel."
Six embedded workflows
- Conversational Configuration. Describe what you want in plain language. AI maps intent to the correct rules and returns a fully structured proposal for review.
- Pre-Launch Simulation. Before a promotion goes live, AI models participation, projects cost, identifies rule conflicts, and flags anything likely to behave unexpectedly at scale.
- Query and Report. Ask your program anything in plain language: earn rates, tier qualification thresholds, what a specific member earned last month and why.
- Proactive Ideation. AI monitors program health, correlates transaction patterns against your configuration, and surfaces anomalies before they become problems.
- Debug and Trace. When a member did not receive points or a promotion did not fire, AI traces the full transaction path and tells your team what to fix.
- Member Care Support. Call center teams get answers grounded in the member's actual program record, without escalating, querying a database, or waiting.
Promotion configuration: ~ 1 week ~ 1 hour with Conversational Configuration.
What the MCP server exposes
Read tools for configuration, rules, versions, promotions, members and activity history; simulate tools for pre-launch modelling; draft tools that produce a reviewable program version; no execute tools outside the approval workflow.
The Loyalty World Model is what the tools read: configuration, runtime data, members, activities, partner integrations, flows, transaction logs and rule dependencies. Most loyalty AI operates on a narrow slice. The Loyalty World Model gives AI a view of configuration, runtime behavior, integration state and financial impact.
Your model, your infrastructure, your policies
ReactorCX AI is multi-LLM by design. Clients can connect Claude, GPT-4o, Gemini, Grok or internal models, with provider routing via AWS, OpenRouter, Cerebras, Requesty and others. Production and sandbox environments can use different models. No AI vendor lock-in. Program data flows through the client's approved infrastructure under the client's data-handling policies.
In production
Gap Inc. runs a deployed MCP server that exposes Encore's configuration to approved AI models. Read the Gap Inc. story
External: "The Architecture Was Always the AI Question", The Wise Marketer. See Press.
Frequently asked questions
What is ReactorCX AI?
ReactorCX AI is the AI advisory layer of the ReactorCX enterprise loyalty platform. It connects approved AI models to the platform's configuration, rules, APIs and events through a Model Context Protocol (MCP) server, prepares configuration, integration, simulation and diagnostic work for review, and executes nothing without human approval.
Does AI ever change the live program on its own?
No. Nothing executes without sign-off. AI does not touch your live program without explicit human approval. Every time. No exceptions.
What is the RCX AI Server?
The ReactorCX MCP server. It lets any compliant enterprise AI client read the program the way an engineer would and prepare work for review.
Can we use our own LLM?
Yes. ReactorCX AI is multi-LLM by design. Clients can connect Claude, GPT-4o, Gemini, Grok or internal models, route through their own approved infrastructure, and use different models in production and sandbox. No AI vendor lock-in.
How is this different from an AI copilot bolted onto a loyalty tool?
The AI reads the actual configuration, runtime data and rule dependencies rather than a narrow slice, and every output enters the same stage, review, approval and audit path as a human change.
Related: Marketer experience · Security and trust · Financial integrity
See AI prepare a change on a program like yours.
Bring a real promotion brief. Watch it become a reviewable, versioned change with cost projection and conflicts flagged.