Case study · Fintech · AI agents & workflow automation · Internal tool

Social monitoring that answers customers and wakes the right team.

Every mention of the company and ten competitors is classified as it lands. Rules decide what gets an automatic reply, what goes to Support, Pricing or On-call, and what is only logged. Every action records the rule that fired.

Client
Fintech · EU · consumer app
Services
AI Agents & Workflow Automation · Platform Solutions
Stack
Python · collectors for social, app stores and news · classifier with evaluation set · rules engine · paging integration
Timeline
8 weeks to live
Status
Live
mentions · brand + 10 competitors · today
10
competitors tracked beside the brand, across social, app stores and news
12
routing rules: auto-reply, Support, Pricing, On-call, or log only
1
human review above a severity threshold; everything else is logged with its rule
4 min
from an incident spike to the on-call page, with the comms lead copied
The problem

Two analysts read everything, and incidents were found on social before they were found in monitoring.

Mentions of the brand and ten competitors arrived across social networks, app-store reviews and news, in volumes two analysts could not read in a day. Customer complaints waited for a reply; pricing moves by competitors were noticed a week late; an outage showed up as a spike of angry posts before any alert fired.

The team did not want a bot that talks to customers unsupervised. They wanted the classification and the routing done for them, with every automatic action traceable to a rule they had approved.

What we built

Classify everything, reply to the safe cases, route the rest, log it all.

Collectors
Social, app stores and news for the brand and ten competitors, with source attribution on every item and no private data.
Classifier
Topic, sentiment and severity, with an evaluation set labelled by the analysts and re-run on every change.
Rules
Twelve rules the team wrote and approved decide what is auto-replied, routed to Support, Pricing or On-call, or only logged.
Incidents
A spike of incident mentions pages On-call and copies the comms lead within minutes, with the baseline it was measured against.
Outcome

Live, with the analysts reading the digest instead of the firehose.

Auto-replies cover the approved cases; everything else reaches the right team with the rule that sent it. Outcome figures (analyst hours returned, reply times, incidents caught before monitoring) are published here once the first full quarter closes, with their dates and sample sizes.

“The on-call page arrived before our own monitoring did. That was the moment the team trusted it.”

Head of Customer Operations · fintech
Request a reference call under NDA · next: pricing-move alerts, app-store review replies
Start a project

Which conversations about your brand reach you a week late? Tell us, and we reply within one business day.

Send a few lines about the problem and the people it affects. The reply comes from the engineer who would lead the work.

  1. 01 We read the brief and reply in writing.within 1 business day
  2. 02 A 30-minute call to understand scope and constraints.same week
  3. 03 A written proposal: options, assumptions, estimate.within 5 business days
  4. 04 Kickoff with the team that scoped it.on your timeline
NDA on requestEU data residencyCET · overlap with London and New York
Project brief
about 3 minutes
Optional