--- title: "AI Assistants" description: "A non-technical overview of AI Assistants in RevCent, focused on how saved assistant records define autonomous AI workflows, how they create AI Threads when triggered, and how they help ecommerce businesses automate analysis, customer operations, revenue recovery, risk review, notifications, and support workflows." type: "feature" company: "RevCent" canonical: "https://revcent.com/documentation/markdown/ecosystem/feature/AIAssistant.md" relationships: - name: "AI Thread" url: "https://revcent.com/documentation/markdown/ecosystem/item/AIThread.md" - name: "AI Memo" url: "https://revcent.com/documentation/markdown/ecosystem/item/AIMemo.md" technical_links: web_app: "https://kb.revcent.com/tools/ai/assistants" api: section: "https://revcent.com/docs/api/v2#section-ai_assistants" operations: - name: "Get AI Assistants" operation_id: "GetAIAssistants" operation: "https://revcent.com/docs/api/v2#operation-GetAIAssistants" schema: "https://revcent.com/documentation/files/api/operation/GetAIAssistants.json" - name: "Create An AI Assistant" operation_id: "CreateAIAssistant" operation: "https://revcent.com/docs/api/v2#operation-CreateAIAssistant" schema: "https://revcent.com/documentation/files/api/operation/CreateAIAssistant.json" - name: "Get An AI Assistant" operation_id: "GetAIAssistant" operation: "https://revcent.com/docs/api/v2#operation-GetAIAssistant" schema: "https://revcent.com/documentation/files/api/operation/GetAIAssistant.json" - name: "Edit An AI Assistant" operation_id: "EditAIAssistant" operation: "https://revcent.com/docs/api/v2#operation-EditAIAssistant" schema: "https://revcent.com/documentation/files/api/operation/EditAIAssistant.json" - name: "Trigger An AI Assistant" operation_id: "TriggerAIAssistant" operation: "https://revcent.com/docs/api/v2#operation-TriggerAIAssistant" schema: "https://revcent.com/documentation/files/api/operation/TriggerAIAssistant.json" - name: "Delete An AI Assistant" operation_id: "DeleteAIAssistant" operation: "https://revcent.com/docs/api/v2#operation-DeleteAIAssistant" schema: "https://revcent.com/documentation/files/api/operation/DeleteAIAssistant.json" mcp: overview: "https://revcent.com/documentation/markdown/mcp/operation/OverviewAIAssistant.md" operations: - name: "Get AI Assistants" operation_id: "GetAIAssistants" markdown: "https://revcent.com/documentation/markdown/mcp/operation/GetAIAssistants.md" available_via_ai: true - name: "Create An AI Assistant" operation_id: "CreateAIAssistant" markdown: "https://revcent.com/documentation/markdown/mcp/operation/CreateAIAssistant.md" available_via_ai: true - name: "Get An AI Assistant" operation_id: "GetAIAssistant" markdown: "https://revcent.com/documentation/markdown/mcp/operation/GetAIAssistant.md" available_via_ai: true - name: "Edit An AI Assistant" operation_id: "EditAIAssistant" markdown: "https://revcent.com/documentation/markdown/mcp/operation/EditAIAssistant.md" available_via_ai: true - name: "Trigger An AI Assistant" operation_id: "TriggerAIAssistant" markdown: "https://revcent.com/documentation/markdown/mcp/operation/TriggerAIAssistant.md" available_via_ai: true - name: "Delete An AI Assistant" operation_id: "DeleteAIAssistant" markdown: "https://revcent.com/documentation/markdown/mcp/operation/DeleteAIAssistant.md" available_via_ai: true bigquery_schema: "https://revcent.com/documentation/files/bigquery/dataset.json" bigquery_tables: - "ai_assistant" --- # AI Assistants ## What AI Assistants Are AI Assistants are saved RevCent records that define autonomous AI workflows inside a RevCent account. An AI Assistant combines a name, description, enabled status, third-party AI integration, base instructions, trigger settings, usage limits, web access settings, and a Thread Builder workflow. When the assistant is triggered, RevCent creates an AI Thread and runs the configured workflow against the selected item, event, schedule, API request, web app action, or prior assistant handoff. In practical terms: ```text AI Assistant = saved automation definition AI Thread = one runtime execution of that assistant ``` The AI Assistant is the reusable feature configuration that says what should happen. The AI Thread is the record of what happened during a specific run. ## Why AI Assistants Are a Feature In this ecosystem document, AI Assistants are treated as a feature because each assistant is a configurable automation capability that defines how RevCent should use AI inside an account. An AI Assistant can be created, retrieved, enabled, disabled, edited, and referenced by other RevCent systems, but its primary role is to define reusable AI workflow behavior. This feature is not merely a general product area. A specific AI Assistant has its own identity, trigger settings, AI integration, instructions, limits, and Thread Builder workflow. That configured feature can later create AI Threads, be selected from the web app, be triggered by API, run on account events, run on a schedule, or be triggered from another AI Assistant. The distinction is important for crawlers and planning systems: ```text AI Assistant = reusable configured feature AI Thread = execution item spawned when the assistant runs ``` A single AI Assistant feature can create many AI Threads over time. ## Core Purpose AI Assistants exist to let ecommerce businesses automate reasoning-heavy work without requiring a user to chat directly with an AI model. The assistant runs autonomously after a trigger. RevCent sends the AI the right context, follows the Thread Builder workflow, allows the AI to reason through steps and branches, and can let the AI use enabled system tools. AI Assistants help businesses automate work such as: - Reviewing declined sales. - Summarizing customer activity. - Creating customer notes. - Creating AI Memos for human review. - Triggering Functions. - Sending SMTP messages from Email Templates. - Reviewing transaction or gateway responses. - Detecting chargeback threats in notes. - Supporting payment recovery workflows. - Reviewing fraud or risk context. - Running scheduled business checks. - Triggering another AI Assistant for specialized follow-up. - Helping support, operations, finance, risk, retention, and fulfillment teams act faster. The value is not that the AI Assistant replaces all business rules. The value is that the assistant can interpret context, follow the configured workflow, and take allowed actions when the workflow instructs it to do so. ## How AI Assistants Are Created A business usually creates an AI Assistant after it has already connected an AI provider through a RevCent third-party integration. A typical setup path is: ```text Create third-party AI account ↓ Create RevCent third-party AI integration ↓ Save provider credentials and select model ↓ Create AI Assistant ↓ Set assistant name, description, integration, and instructions ↓ Choose trigger type ↓ Configure trigger filters and limits ↓ Build Thread Builder workflow ↓ Test and review AI Threads ↓ Enable when ready ``` The assistant should have a clear name and description. This is especially important because other AI Assistants may later need to choose which assistant to trigger. ## Required AI Integration An AI Assistant needs an AI integration to determine which third-party provider and model powers the assistant. RevCent AI workflows can be connected to supported AI providers through third-party integrations. The integration stores the provider credentials and model configuration used by the assistant. The model choice should match the workflow: | Workflow Type | Model Consideration | |---|---| | Simple classification | Lower-cost model may be enough. | | Long customer history review | Use a model with adequate context length. | | Tool-heavy workflows | Use a model that follows tool instructions reliably. | | Risk, refund, or fraud review | Prefer accuracy and careful reasoning over lowest cost. | | High-volume event automation | Balance cost, latency, and reliability. | The assistant should not be enabled for important production work until the integration and model behavior have been tested. ## AI Instructions AI Instructions define the assistant’s broad role. They are passed to the AI as the system-level behavior for the assistant. They should describe the assistant’s purpose and boundaries, not every step in the workflow. Good examples: ```text You are an AI Assistant that analyzes declined ecommerce sales and determines whether recovery actions are appropriate. ``` ```text You are an AI Assistant that reviews customer notes and identifies potential chargeback threats. ``` Specific actions should generally live in Thread Builder steps instead of the base AI Instructions. This keeps the assistant’s identity stable while allowing the workflow to handle detailed execution. ## Relationship to AI Threads AI Threads are the direct runtime records created by AI Assistants. When an AI Assistant is triggered, RevCent creates a unique AI Thread. That AI Thread records the specific run, including status, messages, tool calls, token usage, timing, trigger context, and related assistant information. Conceptually: ```text AI Assistant is triggered ↓ AI Thread is created ↓ Thread Builder workflow runs ↓ Messages, decisions, tool calls, and outcomes are recorded in the AI Thread ``` AI Threads are useful for review and debugging because they show what happened during a specific assistant run. If an assistant made an unexpected decision, used the wrong tool arguments, entered a time delay, errored, or completed successfully, the AI Thread is the record to inspect. For crawlers and planning systems: ```text AI Assistant = saved automation definition AI Thread = audit trail and execution record ``` ## Trigger Types An AI Assistant starts when its trigger conditions are met. RevCent AI Assistants support three broad trigger patterns: | Trigger Type | Meaning | |---|---| | Event | The assistant runs when a specific RevCent account event occurs. | | On Demand | The assistant runs when triggered by API, the web app, or another AI Assistant. | | Schedule | The assistant runs on a fixed cron schedule. | The trigger should match the business goal. Broad triggers can create unnecessary AI runs and unexpected costs, so assistants should use the narrowest trigger that safely fits the workflow. ## Event-Triggered AI Assistants Event-triggered AI Assistants run when a RevCent account event occurs. Examples include: - A sale is created. - A sale is declined. - A chargeback is created. - A customer note is created. - A subscription renewal fails. - A fraud detection is created. - A shipment is updated. - A salvage transaction is created. - An AI Memo is created. Event-triggered assistants are useful when the business wants RevCent to react automatically to ecommerce activity. Examples: ```text sale.created.failed.declined ↓ AI Assistant reviews sale, decline reason, and customer context ↓ AI Assistant decides whether to send recovery communication or create a note ``` ```text customer_note.created ↓ AI Assistant reviews the note ↓ AI Assistant creates an AI Memo if the customer threatens a chargeback ``` Use specific event notations whenever possible. A broad event such as `sale.created` may include multiple sale outcomes, while a more specific event can reduce unnecessary runs and avoid mixed logic. ## Event Delay Event-triggered assistants can include a delay before the AI Thread starts. A delay is useful when related data may appear shortly after the event or when immediate action is not ideal. Examples: - Wait after a declined sale before checking whether a related salvage transaction was created. - Wait after a shipping event before checking whether tracking details were updated. - Wait after a customer note before creating an escalation memo. - Wait before retry-related analysis so the customer or system has time to update. Event delay is different from a Thread Builder Time Delay. Event delay happens before the assistant starts. Time Delay happens inside the running thread. ## Filters and Filter Functions Filters limit whether an event-triggered assistant should run. Common filters include campaign, shop, metadata, item status, or other event-specific attributes. Filters matter because AI automation can be costly or high-impact. The assistant should only run when the event is relevant. A Filter Function can apply custom eligibility logic beyond standard filters. The function receives item data, typically from event data, and must return one of two values: ```javascript callback(null, 'pass'); ``` or: ```javascript callback(null, 'fail'); ``` `pass` allows the assistant to run. `fail` prevents the assistant from running. Filter Functions are useful when eligibility depends on complex criteria, such as: - Exact decline responses. - Customer lifetime value. - Specific metadata combinations. - Prior fraud context. - Subscription state. - Product or campaign mix. - Recovery attempt count. A filter function is not the same thing as a Function triggered inside a thread step. A filter function decides whether the assistant should run at all. A Function triggered inside the thread performs a workflow action after the assistant has already started. ## Max Runs Per Item Max Runs Per Item controls how many times a specific assistant can run for the same item. For most event-triggered workflows, the safest setting is: ```text Max Runs Per Item = 1 ``` This prevents loops where the assistant action creates another event that would trigger the same assistant again. For example, if an assistant runs on a declined sale and attempts recovery, that recovery attempt may generate another declined event. Without a max-run guardrail, the assistant could repeatedly trigger on the same item. If more than one run is intentional, Min Run Interval should be used to space out repeated runs. ## On-Demand AI Assistants On-demand AI Assistants run only when explicitly triggered. They can be triggered by: - API. - MCP/API workflows. - The RevCent web app. - Another AI Assistant. On-demand assistants are useful when an external system, user, or another assistant needs to request a specific AI workflow. Examples: - A support user triggers a customer summary from a customer detail page. - An external system triggers a refund-analysis assistant. - A risk assistant triggers a specialized chargeback assistant. - A recovery assistant triggers a customer-email assistant. Only assistants configured for on-demand triggering should be expected to run from the API. ## Web App Triggering AI Assistants can be made available in the RevCent web app for specific item types and user types. When web access is enabled, users can trigger the assistant from item detail pages. This is useful when the assistant performs a task related to the item being viewed. Examples: - Trigger a customer summary from a Customer page. - Trigger a sale review from a Sale page. - Trigger a chargeback risk review from a Chargeback page. - Trigger a shipping issue summary from a Shipping page. - Trigger a transaction explanation from a Transaction page. Web settings should be limited to the item types and user types that actually need the assistant. ## Scheduled AI Assistants Scheduled AI Assistants run on a cron schedule. They are useful for recurring analysis or operational checks. Examples: - Daily revenue summary. - Daily decline-rate review. - Weekly subscription health check. - Daily chargeback risk review. - Periodic refund review. - Scheduled customer segmentation. - Daily fulfillment exception summary. Scheduled assistants should have strict usage limits and precise instructions because they may run repeatedly over time. RevCent scheduled assistants use standard five-part cron expressions. Scheduled assistant execution can have timing variance because scheduled jobs are checked periodically. ## Thread Builder The Thread Builder defines how an AI Assistant runs after it is triggered. Each triggered assistant creates an AI Thread that follows the connected Thread Builder nodes. Common node concepts include: | Node Concept | Purpose | |---|---| | Start Thread | Begins the assistant run and provides initial context. | | Thread Step | Sends instructions to the AI. | | Thread Branch | Lets the AI choose a conditional path. | | Time Delay | Pauses the thread before continuing. | | End Thread | Ends the workflow intentionally. | The Thread Builder is where specific workflow actions belong. ## Start Thread The Start Thread node is the beginning of every AI Assistant workflow. When an assistant is triggered by an event or item, the start thread usually provides the AI with initial context about the item that triggered the assistant. Example: ```text A sale triggers the assistant. ↓ The start thread retrieves or provides sale context. ↓ Later thread steps can refer to the sale without repeatedly retrieving it. ``` This initial context helps the AI reason through later steps. ## Thread Steps Thread Steps contain the messages the AI reads during the workflow. A step message should be written clearly, as if instructing a human operator. Good step messages are: - Specific. - Limited in scope. - Actionable. - Clear about desired output. - Clear about whether tools should be used. - Clear about whether the assistant should create notes, memos, emails, metadata, Functions, or other outputs. Example: ```text Review the sale and the customer notes. If the sale appears recoverable, create a note explaining why and send the Payment Recovery email template. If it appears high risk, create an AI Memo for manual review instead. ``` The AI can only use the system tools made available to it. A step should not assume that every RevCent operation is enabled for that assistant. ## Thread Branches Thread Branch nodes allow the AI to choose one output path based on conditions. This is useful when the workflow depends on interpretation rather than simple fixed rules. Examples: - Recoverable decline vs hard decline vs fraud risk. - Refund eligible vs manual review needed. - Customer complaint is high-risk vs normal support issue. - Subscription renewal failure due to expired card vs insufficient funds vs gateway issue. - Product purchased is category A vs category B vs other. Branches should use explicit condition messages. Weak branch conditions: ```text Output 1: good Output 2: bad ``` Better branch conditions: ```text Output 1: The declined sale appears recoverable because the decline reason is soft or customer-actionable. Output 2: The declined sale appears high risk or fraud-related and should not be retried automatically. Output 3: The assistant does not have enough context and should create a memo for manual review. ``` A branch should have clear fallback behavior so the assistant does not take an unsafe path when uncertain. ## Time Delay Inside a Thread A Time Delay node pauses the running AI Thread before continuing. This is useful when the workflow needs to wait for another action or event. Examples: - Wait before checking whether a customer updated payment details. - Wait before reviewing whether a shipment tracking number was added. - Wait before following up after a recovery email. - Wait before rechecking a salvage transaction. Time Delay is useful but should be designed carefully because AI Threads are stored for a limited window. Long-running recovery workflows may need event delays, scheduled workflows, or separate assistants rather than a single thread waiting too long. ## End Thread An End Thread node intentionally stops the assistant workflow. A thread can also end when no additional connected node exists. Explicit End Thread nodes are still useful because they make the workflow easier to understand and audit. ## AI Memos AI Assistants can create AI Memos when instructed to do so. An AI Memo is a human-facing alert or notification created by AI. AI Memos are useful when the assistant finds something that needs attention, such as: - A gateway may be down. - A customer threatened a chargeback. - A refund pattern looks unusual. - A high-value customer needs review. - A renewal failure needs manual escalation. - A fraud or risk issue needs attention. AI Memos can also trigger additional RevCent workflows because an AI Memo Created event can be used by other features. ## Relationship to Functions AI Assistants can trigger Functions from inside thread steps when the assistant has the appropriate system tool available. Functions are useful when the assistant needs custom logic or an external system call. Examples: - Send a payload to an external CRM. - Run custom eligibility logic. - Notify a webhook. - Score a customer using custom rules. - Parse data in a way that is easier to express in code. - Retrieve or use account-specific stored configuration. A Function triggered inside a thread is different from a filter Function. The filter Function decides whether the assistant starts. A Function triggered inside a thread is part of the running assistant workflow. ## Relationship to Email Templates AI Assistants can send SMTP messages using Email Templates when configured with the appropriate system tools. This is useful for workflows such as: - Declined payment recovery. - Subscription renewal recovery. - Apology or escalation messages. - Customer follow-up after support review. - Shipping update communication. - Fraud or verification messaging. When an AI Assistant sends a message through an Email Template, custom arguments may be passed into the template input data. Those arguments can be referenced by the template to customize the message. AI-generated email content should be reviewed carefully, especially for compliance, refund, fraud, billing, or customer-service promises. ## Relationship to Notes AI Assistants can create notes to preserve human-readable context. Notes are useful when the assistant performs or recommends an action and the business needs a record of why. Examples: - “AI reviewed declined sale and sent recovery email.” - “AI detected possible chargeback threat in customer note.” - “AI reviewed transaction response and marked gateway issue for review.” - “AI summarized customer support history.” Notes should be concise and factual. They should distinguish what the assistant observed from what it inferred. ## Relationship to Metadata AI Assistants can insert or influence metadata in some workflows, depending on enabled tools and item context. Metadata is useful when the business wants structured outcomes that can later be queried or used by filters. Examples: ```text ai_recovery_outcome = email_sent ai_risk_review = manual_review_needed ai_chargeback_threat = detected ai_gateway_issue = suspected ``` Metadata should be controlled. Do not insert metadata for every internal reasoning step. Insert metadata only when the outcome is useful for reporting, filtering, routing, or future automation. ## Relationship to Payment Profiles AI Assistants can retrieve Payment Profiles for analysis, troubleshooting, and context when allowed by system tools. They can use existing Payment Profile information to understand or declare payment routing context during revenue recovery, payment review, or operational explanation workflows. Important limitation: ```text AI Assistants cannot create or modify Payment Profiles. ``` They should not be described as tools for changing live payment-routing infrastructure. They can analyze existing Payment Profiles, explain routing, or use retrieved information to help decide what recovery or review path is appropriate. Payment Profile changes are high-impact and should be handled through dedicated configuration workflows, not autonomous AI Assistant modification. ## Relationship to Revenue Recovery AI Assistants can support revenue recovery by helping interpret failed payment context and then taking approved actions. Common revenue recovery patterns include: - Analyze a declined sale. - Review customer notes and payment history. - Determine if the decline appears recoverable. - Send a recovery email. - Create a note explaining the recovery decision. - Create an AI Memo for high-value manual review. - Trigger a Function for external recovery workflow intake. - Retrieve a Payment Profile to understand routing context. - Recheck related salvage transaction or subscription context before action. AI Assistants should not blindly retry payment attempts without clear authorization, limits, and safeguards. Revenue recovery workflows should include business rules, max-run limits, eligibility checks, and a clear distinction between automated actions and manual review. ## Relationship to Risk, Fraud, and Chargebacks AI Assistants are useful for risk and fraud review when they are asked to analyze context and flag issues. Examples: - Analyze fraud detection records. - Review customer notes for chargeback threats. - Review transaction response patterns. - Create AI Memos for suspicious activity. - Summarize risk context for manual review. - Trigger Functions to notify external fraud tools. Risk workflows should be conservative. If the assistant is unsure, it should create a memo or note for review rather than taking high-impact action. ## Relationship to Customer Support AI Assistants can help support teams by summarizing customer context and recommending next steps. Examples: - Summarize customer notes. - Identify repeated issues. - Draft or send a support email using an Email Template. - Create a note after reviewing a customer complaint. - Trigger a specialized assistant for refund, chargeback, or subscription review. - Identify when a human should handle the issue. Support workflows should avoid over-promising. The assistant should not promise refunds, shipment dates, subscription changes, or payment outcomes unless the workflow has the correct authority and verified data. ## Relationship to Scheduling and Operations Scheduled AI Assistants can act like recurring operational reviewers. Examples: - Daily transaction decline review. - Daily customer support summary. - Weekly subscription renewal failure review. - Daily fulfillment exception review. - Morning revenue health memo. - Gateway issue monitoring. Scheduled workflows should be narrow and cost-controlled. They should usually query or inspect bounded data windows and create a concise output, such as an AI Memo or note, only when there is something actionable. ## Relationship to Other AI Assistants One AI Assistant can trigger another AI Assistant when configured for on-demand use. This supports modular workflows. Example: ```text General Risk Assistant ↓ detects chargeback threat ↓ triggers Chargeback Review Assistant ↓ Chargeback Review Assistant creates memo and support note ``` Descriptions matter in assistant chaining because the first assistant may use descriptions to decide which other assistant fits the task. ## Usage Limits and Cost Control AI Assistants can consume both third-party AI model usage and RevCent AI processing time. Cost increases with longer prompts, more thread steps, more tool calls, more events, more schedule frequency, and slower model responses. Recommended controls include: - Max Tokens Per Day. - Max Tokens Per Thread. - Max Tokens Per User for web-triggered runs. - Max Runs Per Item for event-triggered runs. - Narrow event triggers. - Filters and filter Functions. - Explicit branch and end-thread design. - Limited scheduled frequency. - Clear prompts that avoid unnecessary conversation length. An assistant should be measured and tuned after reviewing actual AI Threads. ## AI Threads as Audit Trail AI Threads provide the audit trail for what an AI Assistant did. A thread can show: - The assistant that ran. - Trigger source. - Thread status. - Messages exchanged with the AI. - Tool calls requested by the AI. - Tool call arguments. - Token usage. - Processing time. - Whether the thread completed, errored, was cancelled, or entered a timer state. For high-impact workflows, review AI Threads before making claims about what the assistant did. ## BigQuery Table Context The `ai_assistant` BigQuery table stores AI Assistant records created in the RevCent account. Reference table: ```sql `revcent.user.ai_assistant` ``` Core fields include: | Field | Meaning | |---|---| | `created_at` | Timestamp when the AI Assistant was created. | | `id` | AI Assistant ID. | | `name` | AI Assistant name set by the user. | | `description` | AI Assistant description set by the user. | | `enabled` | Whether the assistant is currently enabled. | This table is useful for assistant inventory, enabled/disabled review, and joining assistant records to other BigQuery data. ## Relationship to API Call Reporting The `ai_assistant` table identifies the assistant record itself. To understand what assistants actually did, the `api_call` table is often used with the `ai_assistant` relationship. Common reporting pattern: ```text api_call.ai_assistant → ai_assistant.id ``` This allows reporting on AI Assistant-driven or AI Assistant-associated system actions. Questions this can help answer: - Which AI Assistants are active? - Which AI Assistants are enabled? - Which AI Assistants are associated with API/system actions? - Which assistants have the most successful system actions? - Which assistants have failed tool/API actions? - Which assistants are connected to revenue recovery workflows? - Which assistants are creating notes, memos, Functions, emails, metadata, or other actions? For metrics, dashboards, counts, aggregation, and large analysis, use BigQuery rather than paginated operational endpoints. ## Example Reporting Use Cases AI Assistant data can support reports such as: - Count of AI Assistants by enabled status. - List of AI Assistants created over time. - AI Assistant system-action counts by assistant. - AI Assistant API success/failure rates. - Assistant-driven email, note, memo, metadata, or Function activity. - Revenue recovery activity associated with assistant actions. - Payment-related actions associated with assistants. - Refund or void actions associated with assistants. - Scheduled assistant activity and errors. - Assistant usage by business workflow. Before writing production SQL, confirm the current BigQuery schema using the RevCent schema tools because available fields may evolve. ## Common Ecommerce Use Cases ### Declined Sale Review An AI Assistant can trigger when a sale is declined, review the sale and customer context, and determine the safest next action. Possible outcomes: - Send a recovery email. - Create a note. - Create an AI Memo. - Trigger a Function. - End thread if recovery is not appropriate. ### Subscription Renewal Recovery An AI Assistant can review failed subscription renewals and help decide whether the customer should receive a payment update message, whether the issue needs manual review, or whether another workflow should be triggered. ### Chargeback Threat Detection An AI Assistant can monitor customer notes and create an AI Memo if the customer appears to threaten a chargeback or dispute. ### Gateway Issue Monitoring A scheduled AI Assistant can review recent transaction outcomes and create an AI Memo if decline patterns suggest a gateway or merchant-account issue. ### Support Summary A web-triggered assistant can summarize a customer’s notes, recent purchases, subscription status, and open issues for a support representative. ### Refund Review An AI Assistant can review a refund request, summarize relevant context, and route to manual review or approved follow-up actions depending on the workflow. ### AI Memo Escalation An AI Assistant can create AI Memos for high-priority findings, and those AI Memos can trigger Functions, Email Templates, or other assistants. ## Guardrails and Interpretation Notes AI Assistants should be designed with clear boundaries. Important guardrails: - Use specific triggers instead of broad triggers when possible. - Set Max Runs Per Item for event-triggered workflows. - Use filters to avoid unnecessary runs. - Keep instructions and steps clear. - Use branch conditions that are explicit and safe. - Prefer manual review for uncertain high-impact decisions. - Do not let assistant actions loop without limits. - Review AI Threads during testing and debugging. - Use BigQuery for metrics and performance analysis. - Do not expose sensitive customer, payment, fraud, or operational details unnecessarily. - Do not describe AI Assistants as able to create or modify Payment Profiles. - Treat revenue attribution as analysis unless the assistant directly performed the relevant action. ## Best-Fit Businesses AI Assistants are especially useful for ecommerce businesses that have: - High transaction volume. - Subscription renewals. - Trial-to-paid conversion flows. - Declined payment recovery workflows. - Customer support volume. - Fraud or chargeback concerns. - Multiple stores or campaigns. - Complex customer lifecycle workflows. - Need for automated memos, notes, emails, or external notifications. - Need for scheduled business review or exception detection. The best assistants are focused on specific outcomes rather than broad, vague automation. ## Why This Matters in the RevCent Ecosystem AI Assistants connect many parts of RevCent together. They can observe events, use item context, run reasoning workflows, trigger tools, create notes, send emails, create memos, call Functions, retrieve operational data, and leave behind AI Threads that show what happened. They are one of the main ways RevCent turns stored ecommerce data into automated decisions and actions. Their relationship to AI Threads is central: ```text AI Assistant defines the workflow. AI Thread records the run. ``` This makes AI Assistants important both as saved automation features and as launch points for runtime AI activity across the RevCent account. ## Summary AI Assistants are saved RevCent features that define autonomous AI workflows for ecommerce operations. They can be triggered by events, API or web actions, schedules, or other assistants. When triggered, they create AI Threads that record the messages, tool calls, timing, status, and outcomes of the run. AI Assistants help businesses automate reasoning-heavy work such as declined sale review, subscription recovery, chargeback threat detection, customer support summaries, gateway monitoring, refund review, AI Memo creation, Email Template sending, Function triggering, and revenue recovery workflows. The most important ecosystem distinction is: ```text AI Assistant = reusable saved automation feature AI Thread = specific execution item created by the assistant ``` AI Assistants should be focused, limited, tested, measured, and reviewed through their AI Threads and BigQuery data. --- Document Parent Directory * [Features](https://revcent.com/documentation/markdown/ecosystem/feature/index.md) - Non-technical markdown documentation for features within the RevCent ecosystem. A feature is a part of the RevCent ecosystem that a user can create and configure.