AI in customer loyalty is shifting programs from cost to growth

At a glance

What this is: This is a Comarch recap of a Loyalty 360 webinar arguing that AI is changing loyalty programs from retention tooling into growth infrastructure.

Why it matters: It matters because loyalty platforms increasingly expose first-party data, reward entitlements, and conversational access patterns that identity, fraud, and governance teams must secure.

By the numbers:

• Top-quartile loyalty programs can make consumers 50% more likely to increase purchase frequency.
• Over 35% of consumers already use AI tools like ChatGPT to research and evaluate loyalty programs.
• Highly rated loyalty programs make members 20% more likely to choose the brand over competitors.
• Top-quartile loyalty programs can produce up to a 14-point higher Net Promoter Score.

👉 Read Comarch's analysis of AI-driven customer loyalty and MCP

Context

AI-driven loyalty is moving beyond campaign optimisation into customer-facing workflows that change how brands discover, rank, and deliver offers. The governance gap is that these workflows increasingly depend on first-party data, API access, and conversational interfaces, which makes loyalty a security and identity problem as much as a marketing one.

Model Context Protocol adds a further layer of exposure because it connects AI tools to live data and actions. When loyalty balances, tier status, and redemption rights become callable through agent-facing interfaces, teams need to think about access control, consent, and abuse prevention in the same frame they use for identity and entitlements.

The article reflects a broad market shift rather than an unusual outlier. Many enterprise loyalty programmes still operate with static tiers and fragmented systems, but the next phase will reward organisations that can govern data, access, and AI-mediated decisioning together.

Key questions

Q: How should security teams govern AI-connected loyalty platforms?

A: Security teams should treat loyalty platforms as entitlement systems with identity-sensitive workflows, not just marketing tools. That means scoped API access, strong token governance, audit logs for every redemption or tier change, and policy gates for any AI action that affects customer value. The right control model combines IAM, fraud detection, and business-rule enforcement.

Q: Why do loyalty programs need identity controls when AI assistants are involved?

A: AI assistants can query data and trigger actions at machine speed, so loyalty access is no longer limited to human users clicking through a portal. Identity controls are needed to ensure the assistant only acts within approved context, for the right customer, and within narrow tool permissions. Without that, the AI layer becomes a powerful proxy for misuse.

Q: What breaks when loyalty entitlements are spread across multiple systems?

A: When rewards, partner rules, and redemption logic are fragmented, organisations lose a clear view of who can access what and under which conditions. That increases fraud risk, creates inconsistent customer experiences, and makes revocation or rollback difficult. It also weakens assurance when AI systems are deciding or executing offers across channels.

Q: How do organisations keep agentic loyalty automation under control?

A: Use hard policy boundaries, human approval for high-impact actions, and full logging of model decisions, overrides, and exceptions. Agentic automation should be limited to low-risk optimisation tasks until teams can prove it behaves predictably under edge cases. Governance should focus on containment, not just output quality.

Technical breakdown

Why loyalty data becomes an identity problem in AI workflows

Loyalty platforms increasingly hold first-party data, reward balances, tier entitlements, and behavioural signals that are valuable both to marketers and attackers. Once those systems are exposed through APIs or AI interfaces, the security model stops being simple application access and becomes entitlement governance. The key technical issue is not just whether the system authenticates a user, but whether the requested action is authorised for that identity, context, and session. That is where loyalty begins to intersect with IAM, fraud controls, and delegated access management.

Practical implication: treat loyalty APIs and AI connectors as governed access paths, not just marketing integrations.

How MCP changes the attack surface for customer-facing AI

Model Context Protocol standardises how AI tools connect to external systems, which is useful for interoperability but also expands the trust boundary. In loyalty use cases, an AI assistant may query balances, retrieve account data, or initiate redemptions on behalf of a user. That means the security model has to account for tool-level permissions, scoped tokens, and prevention of overreach by the model or connected agent. Without those controls, the AI layer can become a high-trust proxy with more capability than the user intended.

Practical implication: define per-tool scopes and session boundaries before exposing loyalty systems to AI assistants.

Why agentic loyalty execution needs more than rules engines

The article points to an agentic shift where AI assistants can simulate budgets, execute campaigns, and resolve offer conflicts automatically. That is a material change from rule-based personalisation because it introduces runtime discretion. If the system can decide which offer to apply or which audience segment to target, governance must cover model outputs, override paths, logging, and rollback. In practice, that pushes loyalty architecture closer to controlled automation than traditional marketing automation, especially where financial incentives and customer entitlements are involved.

Practical implication: require human approval or hard policy constraints for any AI action that changes reward value or customer eligibility.

Threat narrative

Attacker objective: The attacker aims to monetise loyalty access by stealing value, manipulating entitlements, or harvesting customer data for follow-on fraud.

1. Entry occurs through exposed loyalty APIs, conversational AI connectors, or weakly scoped tokens that allow access to customer reward systems.
2. Escalation happens when an AI assistant or attacker uses overly broad permissions to query balances, modify entitlements, or trigger redemptions beyond intended scope.
3. Impact is fraud, entitlement abuse, or unauthorised disclosure of first-party customer data and loyalty behaviour.

NHI Mgmt Group analysis

AI-native loyalty is becoming an access-governance problem, not just a marketing problem. Once balances, tiers, and redemption logic are callable through AI assistants, the control plane shifts from campaign design to authorisation design. That means identity, session, and entitlement controls now shape revenue outcomes as much as customer experience does. Practitioners should treat loyalty automation as governed access to business value, not merely personalisation.

Model Context Protocol creates a new trust boundary that loyalty teams cannot ignore. MCP is useful because it standardises how tools connect, but the same standardisation makes permission scoping and tool confinement more important. If loyalty agents can query or act on customer accounts, the programme needs least privilege at the tool layer and strict auditability for every action. Practitioners should design for constrained delegation, not broad conversational convenience.

Named concept: loyalty entitlement sprawl. This is the accumulation of reward balances, partner rules, API paths, and AI-mediated actions across multiple systems without a single governance view. It creates inconsistent access decisions and weakens fraud detection because the entitlement state is distributed. Practitioners should map reward entitlements as governed assets, not leave them scattered across marketing, CRM, and partner platforms.

Agentic execution changes the risk profile of promotional logic. When AI can optimise spend or resolve conflicts automatically, errors become operationally and financially material very quickly. That elevates the need for policy gates, logging, and fallback workflows. Practitioners should assume that any autonomous decisioning affecting customer value requires the same discipline as high-risk access change management.

Fraud prevention and identity governance are converging inside customer experience systems. The article’s mention of unauthorized card scanning and internal misuse is a reminder that loyalty platforms sit at the intersection of consumer trust and entitlement abuse. The next governance model will combine fraud signals, identity context, and API authorisation into one control surface. Practitioners should plan for cross-functional ownership rather than marketing-led control alone.

What this signals

Loyalty entitlement sprawl: AI-connected loyalty platforms will force teams to manage reward access with the same discipline they use for privileged business systems. Once customer value can be queried or redeemed through conversational interfaces, access boundaries matter more than campaign novelty.

The practical signal for security leaders is that loyalty data and reward logic should be reviewed in the same governance cycle as APIs, partner tokens, and third-party access. Where AI search or MCP-based assistants are involved, the control set should include scoped delegation, session logging, and revocation of stale connectors.

For practitioners

• Map loyalty entitlements to governed access paths Inventory balances, tiers, partner rewards, and redemption APIs as protected resources with explicit owners, approved consumers, and audit logging. Include conversational interfaces and partner integrations in the same access review process.
• Scope every AI tool connection to least privilege Assign narrow, task-specific permissions for MCP servers and loyalty assistants so they can only query or act on the minimum data required. Revalidate scopes whenever a new workflow, channel, or partner is added.
• Add policy gates before autonomous reward changes Require approval or deterministic guardrails for actions that alter offer value, customer eligibility, or redemption state. Preserve rollback paths and event logs for every AI-driven decision.
• Unify fraud and identity telemetry Correlate unusual reward activity, API token use, and account takeover signals so loyalty abuse can be detected as an access problem, not only a marketing anomaly. This is especially important when AI agents can trigger transactions.
• Audit partner integrations for entitlement drift Review cross-industry partnerships, shared APIs, and delegated access to ensure customer value cannot be expanded beyond intended business rules. Remove stale tokens and unused connectors.

Key takeaways

• AI is turning loyalty programmes into governed access systems because rewards, tiers, and customer data are now exposed through APIs and assistants.
• The article’s own data shows that more than 35% of consumers already use AI tools to research loyalty options, which raises the stakes for discoverability and control.
• Practitioners should combine IAM, fraud telemetry, and policy gates so AI-driven loyalty actions cannot exceed the customer’s intended scope.

Key terms

• Loyalty Entitlement Sprawl: The uncontrolled spread of reward balances, tier rules, partner permissions, and API access across multiple systems. It creates weak visibility into who can redeem what, where automated decisions are made, and how access should be revoked when a partner or workflow changes.
• Model Context Protocol: Model Context Protocol is an open protocol that lets AI agents connect to tools and data sources. It expands what an agent can reach, so governance has to cover not only the model and its prompts, but also every system that can receive or return agent-driven data.
• Agentic execution environment: A runtime in which an AI system can choose actions, call tools, and continue a task with limited human intervention. In identity terms, it becomes an access-bearing environment that can amplify whatever credentials and permissions it inherits, so governance must treat it like a privileged workload.
• First-Party Data: Data collected directly from customers through a brand’s own channels, such as app activity, purchase history, and loyalty interactions. It is valuable for personalisation, but it also becomes sensitive when exposed through APIs, assistants, or partner integrations.

What's in the full article

Comarch's full article covers the operational detail this post intentionally leaves for the source:

• AI-enabled loyalty platform capabilities and how they map to marketing execution.
• Audience Q&A on MCP-connected loyalty interfaces and conversational commerce.
• Practical differences between standalone and coalition loyalty models in larger markets.
• Key takeaways on moving from static tier mechanics to dynamic personalisation.

👉 Comarch's full post covers the webinar recap, audience questions, and AI loyalty operating model in more detail.

Deepen your knowledge

The NHI Foundation Level course, the industry's only accredited NHI security programme, covers NHI governance, machine identity security, secrets management, and identity lifecycle control. It helps security and identity practitioners apply governed access principles to the systems and workflows their programmes increasingly depend on.