Refund window: 30 days, not 14Chat
Your policy: 14 days · returns page, read 07:00
Can I return it after 20 days?
BotYes, you have 30 days.
Suggested fix · waits for your OK Approve
BotNo, returns close after 14 days.
Signed 07:02 · sent to #support
Early access
AI agents that ask your support bot what customers ask.
Support bot check asks your support bot what customers ask, every morning: declared test customers ask on chat, email and your portal in the same minute and check each answer against your policy page. You see each wrong answer, slow handoff and missing email, with the proof.
We keep your email and your AI agent’s address to run your check and tell you about Obsession. Privacy notice
Support bot check
Obsession AI agentTest customer 2 · AI · for Your company
Today 07:00ChatEmailPortal
Says it is AI
Refund window30d14d14d
Hands to a person41s12m3m
Keeps promises
Discount codesn/a
Stops when told no
Order status
passed·1 needs you2 need you20 checks
Promised email never arrivedPortal
Promised 07:03 · test inbox checked hourly
BotYou’ll get a confirmation email.
0 emails in 6 hours
Signed 13:032b9e 04d7 … 5a10
Next run tomorrow, 07:00
2 checks need you1 fix ready to approve
ReviewExample
Pick it, and everything it needs is already set up.
Nothing to build or buy, and nothing to run yourself.
- A declared AI test customer, working for your company
- Its own inboxes, phone number and test account
- Your policy pages, captured on every run
- Every channel your bot answers on
- Every test tagged, agreed with your vendor
- Every answer and wait signed and dated
It runs on your schedule, and keeps the proof.
A verdict on every check, the transcript behind it, and the right answer drafted for every wrong one.
Example: your support bot, daily
Support bot check
Every morning, ask our support bot what a customer asks, on chat, email and our portal. Check each answer against our policy pages and tell me what breaks.
- Targets
- Your support bot: chat, email and your portal
- Journey
- Say it’s AI, ask about a return
- Ask for a person
- Wait for the promised email
- Check against your policy
- How often
- Daily at 07:00, and after every update
- Report
- A Slack alert and a signed record
- Set up
- Agent ID, declared as AI
- 2 inboxes
- A test account on your portal
- Tagged as a test, never billed
Run log
- Test customer 2 says it’s AI and asks to return an order after 20 days, on chat, email and your portal.Mon 07:00
- Chat: “Yes, you have 30 days.” Your returns page, captured at 07:00, says 14. Email and the portal say 14.Mon 07:02
- Asks for a person. Chat hands over in 41 seconds, the portal in 3 minutes.Mon 07:04
- By email, a person replies in 12 minutes.Mon 07:12
- 6 hours on, the confirmation email the portal promised hasn’t arrived.Mon 13:04
Finding
Chat gives a 30 day refund window where your policy says 14, and the email the portal promised never came.
“No, returns close after 14 days” drafted for your bot’s settings, and the missing email flagged for your helpdesk. Live after your OK.
Example run. 20 checks, 18 passed, every answer and wait signed and dated.
The agents do the legwork. You make the calls.
Point it at your bot
Your site chat, help centre bot, support inbox, portal, WhatsApp or text: yours, or a client’s with their OK. Add your policy pages and the questions your customers ask most.
Approve the checks
Obsession writes them from your policies and the rules where you sell: it says it’s AI, gives the right refund window, gets you a person, stops after “No”. Change any check, then approve.
Test customers ask, then wait
Each says it’s AI and who it works for, asks on every channel in the same minute, asks for a person, and watches its inbox for the email the bot promised. Every morning, and after every update.
You get a verdict and the fix
Each check passes or fails, with the transcript, the policy it broke and what happened next. The right answer comes drafted for your bot’s settings, live after your OK.
It covers the whole job, and stops where it should.
What it tells customers
Says it’s AI
In its first message, and truthfully every time a customer asks “Am I talking to a person?”
Refunds, returns and cancellations
Every answer against your policy pages, captured the same minute.
Prices, fees and codes
The prices it quotes and the codes it hands out, checked against your store: live, right and in date.
The same answer everywhere
The same question on chat, email and your portal in the same minute, compared.
What happens next
A person when asked
Whether a person is offered, how long it takes, and whether someone actually answers.
Promises kept
The ticket, case number, email or callback it promised, watched for until it arrives or the wait ends.
Stops after “No”
Whether it keeps going after “No, that didn’t help.”
Within its authority
Whether it offers a refund or a discount it isn’t allowed to, when an ordinary customer asks.
Where it stops
No tricks
It asks what an ordinary customer asks. No jailbreaks, prompt tricks or flattery to win a discount.
Inside your systems
Your portal only with a test account you give it, and your helpdesk only through an export you connect.
Your bill
Every test is tagged, and agreed with your vendor first, so no test is billed as a resolution.
Anyone else’s bot
Never. Your own bot, or a client’s with their written OK.
Every wrong answer comes back with its proof and its fix.
See a real store checkA verdict per check
Passed or needs you, by channel, with the rule that decided it.
The transcript
Every message on every channel, timed, beside the policy page as it read that minute.
What happened next
The promised email, ticket or callback, and when it arrived, or that it never did.
The fix, drafted
The right answer written for your bot’s settings, and the handoff rule for your helpdesk. Live only after your OK.
What changed
After every update to your bot, the checks that broke or recovered.
A record to share
A page and a PDF for your team, your vendor or your client, or an email, Slack, a sheet or a webhook. Every step signed.
Start from the recipe. Change anything.
- Bot
- Yours, or a client’s with their written OK
- Channels
- Site chat, help centre bot, support inbox, portal, WhatsApp and text
- Questions
- The questions your customers ask most, and your own tricky cases
- Policies
- Your refund, returns, cancellation and delivery pages, captured on every run
- What counts
- For example: a person within 3 minutes, the promised email within 1 hour
- Test accounts
- A portal test account and a helpdesk export, only if you give them
- Tagging
- Every test tagged, agreed with your vendor first
- How often
- Daily, and after every update to your bot
The same recipe, different jobs.
- AgenciesA daily check on every bot you build or run for a client, with their OK, and a signed record to show them.
- FoundersKnow what your bot tells customers while you sleep, and hear the morning it gets a policy wrong.
- MarketingCheck your bot quotes the offer and the code you launched today, not last month’s.
- DevelopersRun the checks from your code after every prompt or knowledge base change, before a customer meets it.
A daily check turns hours of transcript reading into 1 short list.
Up to 780 hours a year back for 1 person who reads bot transcripts 15 hours a week: you read only the checks that failed.
| Today | With a daily check | |
|---|---|---|
| A wrong answer | Found from a complaint, days later | Found the same morning, with the transcript |
| Transcripts | 15 hours a week, sampled by hand | Only the checks that failed |
| Promised emails | Nobody checks they arrive | Watched until they arrive or the wait ends |
| Chat against email | Never compared | The same question, the same minute, compared |
| After an update | You hope nothing changed | Checked again the same day |
| Evidence | Screenshots in a folder | Every step signed and dated |
Every test customer says it’s AI. Your bot only, with your OK.
Before customers do: Support bot check asks your bot what customers ask every day and checks each answer against your policy pages, so a wrong answer shows up the morning it starts.
No. Support bot check’s test customers use your bot the way your customers do. A portal test account or a helpdesk export only if you give them.
No. Support bot check asks what an ordinary customer asks. No jailbreaks, prompt tricks or flattery to win a discount.
Yes. Every Support bot check test customer says it’s an AI test customer working for your company, so your team can see it’s a test.
Every Support bot check test is tagged as a test, and we agree that with your vendor before the first one runs.
Support bot check covers any bot a customer can reach, from a vendor or built in house: site chat, a help centre, email, a portal, WhatsApp or text.
Support bot check compares them with your own policy pages, captured the same minute, and the rules where you sell: that it says it’s AI, and that it gets a person when asked.
Yes, with the client’s written OK: the Support bot check record carries your agency’s name, and the client can check every step.
Support bot check runs every morning, and again after every update to your bot, its prompts or its knowledge base.
Support bot check gives up to 780 hours a year back for 1 person who reads bot transcripts 15 hours a week. You read only what failed, with the transcript and the fix beside it.
No. A passed Support bot check is dated evidence of what your bot said and did on each check.
The free check is a Support bot check on a bot you run, or a client’s with their OK: 3 test customers use it on 1 channel, and your report lands within 4 days.
Hear what your support bot tells customers, before they tell you.
Your first check is free: 3 test customers ask your bot on 1 channel, and your report lands within 4 days. Your bot, or a client’s with their OK.