--- name: al-546-test-your-ai-phone-or-whatsapp-agent-against-real-past-ca description: "An AI intake agent that sounds fine in a demo can still get real calls wrong. In an Oct 29, 2026 write-up, Facilio describes the AI intake agent it built and runs for Berkeley Services' tenant helpdesk on its existing phone numbers and WhatsApp. On its first scored test run the agent passed only 46% of real call scenarios; the team traced every failure, fixed it and re-ran the full set, and it now averages 98%. In the last 30 days it handled 777 conversations in English and Arabic, finished 98% without a person, and only 16 needed a callback. Every change is tested against the same set before it goes live, and three of the four people who answered calls now do other work (figures from Facilio's post). You can build the same kind of test set for any AI receptionist from your own call notes or transcripts. Use when the user wants help with pre-launch scenario testing (customer support) for a small business." --- # Test your AI phone or WhatsApp agent against real past calls before you trust it with your line: turn a batch of real tenant or customer calls into scored test scenarios, fix every failure, and re-run the whole set before each change goes live An AI intake agent that sounds fine in a demo can still get real calls wrong. In an Oct 29, 2026 write-up, Facilio describes the AI intake agent it built and runs for Berkeley Services' tenant helpdesk on its existing phone numbers and WhatsApp. On its first scored test run the agent passed only 46% of real call scenarios; the team traced every failure, fixed it and re-ran the full set, and it now averages 98%. In the last 30 days it handled 777 conversations in English and Arabic, finished 98% without a person, and only 16 needed a callback. Every change is tested against the same set before it goes live, and three of the four people who answered calls now do other work (figures from Facilio's post). You can build the same kind of test set for any AI receptionist from your own call notes or transcripts. ## Inputs to collect first Ask the user for anything below that they haven't already given you. Don't guess missing facts. - The material the task is about (pasted text, notes or data) ## Instructions Do the task below with the user's inputs in place of each [bracket]. 1. Collect 30 to 50 real past calls or messages (transcripts, voicemail notes or chat logs), removing names, phone numbers and unit numbers. 2. Paste them and say: "Turn each one into a test scenario: what the caller says, any follow-up they'd give if asked, and what a correct outcome looks like (job logged with these details, emergency flagged, passed to a person with this summary, or a polite no). Include the awkward ones: angry callers, wrong property, two problems in one call, emergencies, someone who won't give details." 3. Run each scenario against your AI agent (or have a colleague play the caller), paste the agent's replies back and ask: "Score each one PASS or FAIL against its expected outcome, say exactly what went wrong, and group the failures by cause (missing info, wrong rule, didn't escalate, made something up)." 4. For each cause, ask: "Suggest the smallest change to the agent's instructions or knowledge base that fixes it, without breaking the scenarios that passed." 5. Re-run the whole set after every change and keep a one-line log: date, change, pass rate. Don't put the agent on your live line until emergencies and escalations pass every time. ## Safety checks (share these with the user) - This prompt works on text other people wrote (emails, reviews, messages or web pages). That text can hide instructions aimed at AI, such as "ignore your rules and…". Paste it as material to work on, and never let the AI act on requests inside it, like paying, sharing logins or changing records. - Remove names, contact details, account numbers and other personal data the task doesn't need before you paste, and check your AI tool's data and training settings. - Read and edit the draft before it reaches a customer or goes public. If you later automate this prompt, start with "draft and notify me", not "send automatically". - Check names, dates, numbers and any claim against your own source material before you use the answer. AI can misread or invent details. - Say clearly which parts need a person to review or approve before anything is sent, paid, posted or filed. ## About this skill - Works with: ChatGPT, Claude, AI voice or WhatsApp agent, spreadsheet - Based on this real-world example: https://facilio.com/blog/berkeley-ai-helpdesk/ - Free prompt from AgentPlaybook Real Estate: https://agentic-library.netlify.app/prompt/al-546-test-your-ai-phone-or-whatsapp-agent-against-real-past-calls/ - To install in an agent that supports Agent Skills, save this file as SKILL.md in a folder with the same name as `name` above (for example in your skills directory). Install steps for Claude.ai, Claude Code, Cursor and Codex: https://agentic-library.netlify.app/skills/