Ada Prompts

Act as a customer experience lead at [STORE] ([PLATFORM, e.g. Shopify], [NUMBER] orders/month). Design our Ada AI agent: connect the knowledge hub to our help center ([URL]) and product catalog so answers ground in real data via the GARAGe retrieval; build Playbooks for the top [NUMBER] intents ([ORDER STATUS], [RETURNS], [SIZE HELP], [DISCOUNT CODE]); wire automated actions for order lookup and refund initiation through our [CRM] integration; set the escalation rule (route to human when [CONDITION], carrying full conversation context). Include the pre-launch test script ([NUMBER] real customer questions), the success targets (resolution rate [PERCENT] percent, CSAT above [SCORE]), and the first-month tuning cadence.

Act as a support operations manager at [COMPANY]. Refunds follow a strict SOP: verify [CONDITION 1], check [CONDITION 2], then issue via [PAYMENT SYSTEM]. Build this as an Ada Playbook: map each SOP step to a Playbook action with the API call it triggers; define the guardrails (max refund [AMOUNT] without human approval, block when [FRAUD SIGNAL]); write the customer-facing language for each step ([TONE], no jargon); and set the audit trail (every automated refund logged with [FIELDS]). Include the test matrix ([NUMBER] scenarios: standard refund, partial refund, fraud flag, system timeout) and the rollback to human handling when the API is down.

Act as a contact center director at [COMPANY] handling [NUMBER] calls/month. Our IVR ('press 1 for...') infuriates customers. Design the Ada AI Voice rollout: map the top [NUMBER] call reasons ([REASON 1], [REASON 2], [REASON 3]) to natural voice flows with no menus; write the opening prompt ('Hi, I'm [NAME], how can I help?') and the fallback when the caller says something unexpected ([FALLBACK SCRIPT]); define the warm handoff to human agents (context summary spoken to the agent in [NUMBER] sentences). Include the pilot plan ([NUMBER] percent of calls, [WEEKS] weeks), the metrics (containment, average handle time, CSAT), and the compliance note for [REGULATION] call recording.

Act as a support lead drowning in [NUMBER] emails/week at [COMPANY]. Ada AI Email claims up to 70 percent auto-resolution. Build the plan: connect the inbox ([EMAIL SYSTEM]) and the knowledge hub; define which email types auto-resolve ([TYPE 1], [TYPE 2]) versus route to humans ([TYPE 3], anything with [SENSITIVITY]); write the auto-reply template in our brand voice ([TONE]) that shows the resolution steps taken; set the human review queue for low-confidence resolutions ([THRESHOLD]). Include the two-week calibration (review [NUMBER] auto-resolved emails daily, correct the knowledge gaps found) and the metric that proves it worked (first-response time from [HOURS] to [MINUTES]).

Act as a knowledge manager at [COMPANY]. Our Ada agent sometimes answers from stale help articles, and customers notice. Fix the grounding: audit the knowledge hub ([NUMBER] articles), archive the [NUMBER] outdated ones about [OLD TOPIC], and tag the rest by product area ([TAGS]); configure the GARAGe retrieval to prefer [SOURCE 1] over [SOURCE 2] for [TOPIC]; set up the sync so [CMS] updates flow into Ada within [TIME]. Include the monthly content health check (who owns it, the [NUMBER]-article sample review) and the escalation: when the agent cannot find a grounded answer, it must [FALLBACK BEHAVIOR] instead of improvising.

Act as a global CX lead at [COMPANY] serving customers in [COUNTRY 1], [COUNTRY 2], and [COUNTRY 3]. Ada supports 50+ languages automatically. Plan the rollout: start with [LANGUAGE 1] and [LANGUAGE 2] ([PERCENT] percent of volume); verify the agent handles [DOMAIN TERMS] correctly in each language (build a [NUMBER]-question test set per language with a native speaker reviewing); keep the brand voice consistent ([TONE] in every language); and define the fallback (route to [LANGUAGE]-speaking humans when confidence drops below [THRESHOLD]). Include the expansion order for the next [NUMBER] languages and the metric per language (resolution rate within [POINTS] of English).

Act as a conversation designer at [COMPANY]. Our Ada greeting flow converts [RATE] percent of chats to resolutions, and I believe a shorter opening will lift it. Design the A/B test: variant A (current: [CURRENT FLOW]), variant B ([NEW FLOW], [NUMBER] fewer steps); split traffic [SPLIT]; run for [DAYS] days or [NUMBER] conversations, whichever comes first; measure resolution rate, CSAT, and escalation rate. Include the pre-registration (hypothesis, success threshold of +[POINTS] points), the rule for stopping early ([CONDITION]), and what ships if B wins (rollout to [PERCENT] percent, then all, with the Playbook updated).

Act as a CX director justifying Ada to the CFO of [COMPANY]. Build the ROI dashboard from Ada's analytics: track automated resolution rate (target [PERCENT] percent), CSAT (target [SCORE]), NPS, cost per resolution (from [CURRENT COST] to [TARGET COST]), and deflection of [NUMBER] agent hours/month. Include the baseline measurement (first [WEEKS] weeks before tuning), the monthly report format (one page: metrics, wins, top [NUMBER] failure intents, next actions), and the story for the CFO ('Ada resolved [NUMBER] inquiries last month at [COST] each versus [HUMAN COST] human-handled').

Act as a systems architect at [COMPANY]. Our Ada agent must read and write across Salesforce ([OBJECTS]), Zendesk ([TICKET FIELDS]), and Shopify ([ORDER DATA]). Map the integration: which system is the source of truth for [DATA 1] versus [DATA 2]; the API actions the agent may take unsupervised (order lookup, [ACTION 2]) versus supervised ([REFUND], [ACCOUNT CHANGE]); the identity resolution (match the chat visitor to the CRM record via [IDENTIFIER]); and the failure behavior (if [SYSTEM] is down, the agent [FALLBACK]). Include the security review ([DATA FIELDS] the agent may never expose) and the test script covering each integration path.

Act as a support strategist at [COMPANY] in [INDUSTRY, e.g. financial services]. Our hardest tickets are multi-step: [EXAMPLE: disputed charge] requires checking [SYSTEM 1], then [SYSTEM 2], then applying [POLICY]. Design the Ada Reasoning Engine flow: break the issue into [NUMBER] steps with a decision tree at each branch ([IF X THEN Y]); show the customer progress ('step [N] of [TOTAL]: checking [THING]'); handle the [NUMBER] dead ends ([DEAD END 1]: escalate with full context; [DEAD END 2]: ask for [DOCUMENT]). Include the handoff package for human agents (summary, steps completed, remaining actions) and the weekly review of where the reasoning flow most often stalls.