Commercial flow 06 · Support
Resolve, do not forward.
The agent reads the history, checks the account status and replies. It escalates to a person only when the answer requires a decision it cannot make.
Customers don’t need help during office hours. They need it when something breaks.
Which is usually late at night, on a weekend, from a phone, with a screenshot and not much patience.
Traditional chatbots make it worse. They loop people through “please type your order number”, escalate everything to a human and then leave that human without context. And when an incident hits at scale, the wave of complaints arrives long before any communication does.
- HoursIt breaks at night, on a weekend, from a phone.
- The loop“Please type your order number” — then a human with no context.
- IncidentsThe complaint wave arrives long before any notice does.
Resolve what can be resolved instantly. Act inside your systems. Hand the sensitive cases to a person with everything ready.
Evidence is read as it arrives, claims are filed mid-conversation, and escalation happens with a case summary — while a human can step in with one click and hand back with one more.
How Medra runs this flow, step by step
- STEP 1
Resolve the frequent, instantly.
Order status, how-to questions, billing basics. Answered in seconds, in the customer’s language, at any hour.
- STEP 2
Diagnose from evidence.
Receipts, error screenshots and documents are read on the spot and used to understand the case. No more “please type your order number” loops.
- STEP 3
Act in your systems.
The agent files claims, updates records and triggers back-office workflows in the middle of the conversation.
- STEP 4
Escalate with judgment.
Complaints, legal keywords and requests for a human go to the right inbox with a case summary. The agent pauses only on that conversation.
- STEP 5
Close the loop.
Resolved cases trigger CSAT surveys inside the chat with WhatsApp Flows, and escalation reasons feed analytics that tell you which flows to fix next.
- STEP 6
Get ahead of incidents.
When something breaks at scale, proactive notices reach affected customers before the complaint wave forms.
Under the hood
- Multimodal diagnosisPhotos, screenshots, documents and voice notes understood natively.
- Actions in external systemsClaims and updates filed mid-conversation.
- Human on the loopOne click to intervene, one click to hand back. The case stays flagged until it is solved.
- Escalation rulesComplaints, legal terms, compensation decisions and loops, detected automatically.
- WhatsApp Flows CSATSatisfaction measured where the conversation happened.
- Mass proactive noticesIncident communication that reaches people before they reach you.
Faster answers at 3 am than most companies give at 3 pm.
Figures are illustrative, written to look and feel real. Before publishing they are replaced with a client result or kept labeled as illustrative — and we tell you in the first session what your own baseline says.
Autonomous resolution rate
20%→72%Baseline, then the same metric with the flow running.
Time to resolution
18 hours→6 minutesBaseline, then the same metric with the flow running.
CSAT
3.4 / 5→4.6 / 5Baseline, then the same metric with the flow running.
Escalation triggers in this flow
Complaint or conflict detected·Legal or compliance keyword·Explicit request for a human·Compensation decision required·Flow blocked or looping
What it moves
- Autonomous resolution rate
- Time to resolution
- CSAT via in-chat surveys
Your team keeps the hard cases. The agent keeps the night.
The Turing Test, passed in production.
Support is the most demanding version of the test: 11 pm, a voice note, a photo of a receipt and a customer who is already annoyed. At Medra it runs there every day, in more than 19 countries, with customers who have real problems and no idea they are the judges.
out of 100 people can’t tell Medra from a human.
The other 3 catch us on one detail: we reply at 3 am in four seconds.
The moment a customer realizes it’s a bot, they answer in monosyllables and type “agent” until somebody shows up. When it feels human, they tell you what actually broke, what they already tried and what would make it right. That honesty is what a 94% autonomous resolution rate is built on.
How we do it in Support
There’s no single trick. Six mechanisms work together, each one closing a gap that gives other bots away. These three carry the most weight in this flow.
- Native earsVoice notes, photos, screenshots and documents are understood as naturally as text. Nobody is ever asked to “please type your question”.
- Human rhythmPacing, message length and the way replies are split feel like a person typing, not a form printing out an answer.
- Invisible handoffsWhen a person from your team steps in, the conversation keeps its tone and its context. The customer doesn’t feel the seam.
Medra next to the usual options
Humans on the loop, not stuck in it
Passing for human doesn’t mean replacing your team.
A complaint, a legal keyword, a compensation decision: one click and a person from your team takes that thread — only that thread — with a case summary, logs the reason and hands it back when it’s solved.
Every intervention feeds analytics, so you know which flows need tuning and why.
We never hide that we’re AI. We just make sure it never feels like it.
Pick an hour. We’ll bring your numbers to it.
A first session is 30 minutes with the people who would build it — not a sales call. Bring one flow you’d like to fix and we’ll tell you, in the session, whether it’s worth doing.
What happens in the session
- 30 minutesWith the people who would build it. Not a sales call.
- One flowBring the flow you would like to fix first, and the numbers behind it.
- A straight answerWe tell you, in the session, whether it is worth doing.
- The inviteYou get the confirmation and the video link by email.