AI Assistant (Leat MCP)
July 20
→
Responsible AI controls
Control which customer signals Leat AI may use, how automated decisions are evaluated for fairness, and when a person must approve the next step. Every decision can be examined afterward, so your team can understand what happened instead of being asked to trust an unexplained result.
Control which information Decisioning may consider for each type of action. Purchase history, visit patterns, loyalty status, and declared preferences can be permitted, while age, gender, and attributes that could act as substitutes for protected characteristics can be blocked. A signal allowed for product recommendations can still be prohibited for pricing. Information the AI cannot access cannot influence its decisions.
Choose how Leat AI evaluates fair treatment for your business. This might mean similar incentive rates across customer groups, equal access to the most valuable rewards, or comparable treatment for similar customers across channels. Leat monitors results against the thresholds you set and flags changes that may require review before they become an established pattern.
Set which decisions Leat AI may make independently and which require human review. Approval can be required above a certain cost, for particular customer groups, or when the system has limited confidence in its recommendation. Automated Decisioning can also be paused for a segment, incentive type, or the entire program. Your existing program rules continue to apply until automation is resumed.
Review the customer signals used, the rules applied, the available options, the expected value of each, and why one action was selected. Explanations are written for people, allowing customer-facing teams to answer questions and compliance teams to examine how a decision was reached. Every automated choice remains open to review.
Signal definition
Teams can decide which inputs AI may use for each type of decision and block protected attributes or related signals.
Fairness thresholds
The business can define which customer groups should be compared and how much difference should trigger a review.
Fairness outcome review
Incentive rates, incentive value, and access to important rewards can be compared across groups to identify changes in treatment.
Human approval boundaries
Decisions can require a person’s approval based on their cost, customer group, or the system’s level of confidence.
Incident response
Automated Decisioning can be paused for a customer group, incentive type, or the full program while the issue is investigated.
Decision explanations
Teams can review the information, rules, options, and expected results behind an automated decision when a customer or regulator asks.
