NORA · YOUR ENGAGEMENT WATCHER
See which customer needs you first, and why.
Nora is the AI that watches your queue. It reads the conversation before it ranks it, so a price question outranks a thank-you, and it closes an item on its own once the thread shows it was handled.
Nora, WAOversight's engagement watcher, reads a customer's WhatsApp thread when something may need a manager, lists missed replies, overdue promises, customer signals and deals to follow up, and closes an item once the thread shows it was handled. Every finding quotes its message. Nora never replies, and runs on your own AI key: ChatGPT, Gemini, Claude or DeepSeek.
19 → 11Open items once Nora read them. 8 were thank-yous and conversations that had already ended.
Nora
Jordan M. · via Marcus Webb- Summary
- Jordan asked whether the order shipped and needs it before Saturday. Marcus said he would check with the warehouse and reply the same day. He has not. Jordan asked again.
- Waiting for
- A delivery date
- Promises
- Get back to Jordan about the orderoverdue
- Next action
- Give Jordan a delivery date today, or say plainly if Saturday cannot be met.
Every finding quotes its message
Follow-up promised, not kept
2Threads where a rep said they would come back and then did not.
Promises and commitments made
17Delivery dates, prices and guarantees stated in writing to a customer.
Longest wait
27hYour most-ignored customer, by name
01 — WHAT IT READS
One thread, read the way a manager would.
4 messages from Jordan M. A timer sees 3 customer messages and 1 reply. Nora sees a deadline, a broken promise and a customer still waiting.
CONVERSATION REVIEW
Jordan M.
Customer · Marcus Webb
- Jordan · Mon 10:12
Hi, did my order ship? I need it before Saturday, it's a gift.
- Marcus · Mon 10:40
Let me check with the warehouse and get back to you today.
- Jordan · Mon 10:41
Thanks!
- Jordan · Tue 09:31
Any update on my delivery?
A thread is read when something may need you: a customer waiting past half your reply target, a rep message that sounds like a promise, a customer message that sounds like a deadline, a complaint or an order, and once a day on a thread with an open deal.
Each finding names the message it came from and quotes it word for word. If the quote is not in that message, the finding is thrown away before anyone sees it.
Nora reports what it read. Code decides how urgent it is, using your reply target and the time: the model never sets severity, and never decides who is told.
02 — IN THE QUEUE
The message, the reason, and nothing to take on trust.
The attention queue with Nora on: 9 findings on customer threads, each with the customer's or the rep's own words on top and one line of reason underneath.
When the thread shows an item was handled, with a reply sent, a call connected, a promise kept or a conversation that has ended, Nora closes it on its own and records why. If a reason is ever wrong, the message it came from is right above it.
Rule alerts, like a large discount, still fire the way they always did, and Nora closes a rule's item too once the thread shows it was dealt with, unless you switch that off for the rule. Nora's findings replace only the reply-time items on customer threads, and only after a reading has covered the wait.
Attention queue
Nora on · Customers- Missed replies 3
- Overdue promises 2
- Due soon 1
- Customer signals 2
- Follow-ups 1
- 27h
criticalMissed replyNora
“Any update on my delivery?”
Asked twice about an order needed before Saturday. No reply for 27 hours.
Jordan M. · rep: Marcus Webb
- due Mon
highPromise overdueNora
“Let me check with the warehouse and get back to you today.”
Marcus promised an answer on Monday. Nothing followed.
Jordan M. · rep: Marcus Webb
- 82% confidence
highCompetitor mentionedNora
“Saw the same jacket for $64 at another store.”
Compared the price with another store. The price question is still open.
Kevin T. · rep: Sofia Haddad
- due Fri
mediumPromise due soonNora
“Received, thank you! Shipping in 2 days 📦”
Chris promised shipping within 2 days. Due Friday.
Laura Bennett · rep: Chris Delgado
- deal open
mediumFollow up on an open dealNora
“Let me think about it and come back to you.”
Went quiet after a quote for 12 polos. The deal is open in HubSpot, with no follow-up for 4 days.
Andre P. · rep: Hannah Cole
- The exact message, quoted
- Signals show how sure Nora was
03 — YOUR AI, YOUR KEY
Nora runs on the AI you choose. You pay them, not us.
Pick ChatGPT, Gemini, Claude or DeepSeek in Settings and paste your own API key. The provider bills you for what it reads, at its own prices. We add nothing on top.
| Provider | From | We recommend |
|---|---|---|
| ChatGPT | OpenAI | GPT-6.1 Sol |
| Gemini | Gemini 3.8 Flash | |
| Claude | Anthropic | Claude Opus 5 |
| DeepSeek | DeepSeek | DeepSeek Flash |
| A call to your provider | No call |
|---|---|
| A scheduled reading of a thread | Opening, resolving or exporting findings already made |
| Assess now, on a thread you have open | A thread with nothing new said since its last reading |
| A summary or a list of commitments you ask for | A summary asked for again within 24 hours when nothing new was said |
Nora's settings page shows each rep's readings and what they are estimated to cost on your provider, and lets you cap readings per rep per day. If your provider refuses a request, because a key was revoked or an account ran dry, nothing already found disappears: new readings stop, the classic reply-time alerts come back on customer threads, and the workspace owner gets an email.
04 — BEFORE AND AFTER
What changes
TODAY
The queue sorts by the clock. “Thanks!” at 3 hours sits above “Can you hold the order until Friday?” at 40 min, so a manager opens every thread to find the 4 that matter, and most days runs out of time first.
WITH WAOVERSIGHT
The queue opens on what a customer is waiting for, what a rep promised, and what a customer said that should worry you. Each item quotes the message, and Nora clears what is already handled, so a manager reads 9 lines instead of 40 threads.
05 — LIMITS
What this does not do
It does not reply. No linked number can send through WAOversight, so Nora cannot reply for a rep either. It never writes reply text. The most it suggests is one line on what to do next, for a person to do.
It does not score your reps. It points at conversations, and every finding quotes the message it came from, so a manager can check it against what was actually said.
It does not listen to voice notes or calls, or look at images. It knows that one was sent, and whether a call connected, never what was said. A promise kept some other way can still show as overdue until someone resolves it.
It can be wrong. Sarcasm, a very short message, or a thread that switches language halfway are where it slips most. That is why the quote sits above every reason.
It does not read everything. Scheduled readings cover only customer threads, matched by your CRM or marked by you, never broadcast lists, and nothing runs until the workspace owner switches Nora on.
It does not change your response-time numbers. Those are still measured from timestamps on response tracking, not by Nora.
07 — QUESTIONS
What people ask about this
Which plans include Nora?
Every plan. Nora runs on your own AI provider, ChatGPT, Gemini, Claude or DeepSeek, with your own API key, and the provider bills you directly for what it reads. WAOversight does not charge for Nora. It reads customer conversations only: the ones your connected CRM marks as customers, and any contact you mark as a customer yourself.
Who switches it on?
The workspace owner, in Settings, after reading a short disclosure of what is sent and to whom. Nothing is read until then. The owner can switch it off at any time: new readings stop at once, findings already made stay readable, and the queue goes back to sorting by wait time.
What is sent to the AI provider, and who is it?
The provider you choose: ChatGPT, Gemini, Claude or DeepSeek, under your own agreement with them. A scheduled reading or Assess now sends the new messages of a customer thread, a short summary of what came before, and the CRM's deal stage, amount and close date. Phone numbers, email addresses, links and ID numbers are replaced with placeholders first, and each message is labeled only rep or customer. The full detail is on the security page.
Which model should we use?
Any model your provider serves to your key. The settings page recommends one for each provider: GPT-6.1 Sol for ChatGPT, Gemini 3.8 Flash for Gemini, Claude Opus 5 for Claude, DeepSeek Flash for DeepSeek. It shows each model's price per million tokens, and estimates what each rep's readings cost on your provider each month.
Which languages does it read?
It was tested on English, and on conversations that mix English with other languages, including languages typed in English letters. Findings are written in English, and quotes stay in the language and script they were written in. It is least reliable on short messages that mix languages, which is where the quote matters most.
Is a reading recorded in the access log?
Yes. Before a scheduled reading starts, a row is written to the access log as a reading by Nora. When a manager presses Assess now or asks for a summary, it is recorded as that manager's read. If the row cannot be written, the thread is not read.
How quickly does a finding appear?
At the next reading. Readings run 6 times a day, and one always finishes before the 8:00 morning email, so the report covers the night. Assess now reads a thread straight away, once every 30 min per conversation.
Find out what happened last week.
Fifteen minutes, your own numbers, no slides. If you don't see something that surprises you, we'll say so.
WAOversight is an independent product and is not affiliated with or endorsed by WhatsApp or Meta.