Skip to content
Naresh Muppalla

BusinessOnBot · Dec 2025 – Present · Bengaluru

Customer Success Manager

I own 42 D2C brands, from onboarding and setup through go-live and daily success.

  • Configure WhatsApp AI agents in Sim Studio and Langflow.
  • Turn brand issues that repeat into one root-cause fix.
  • Write up every fix, so the team and the AI assistants answer it the same way.

Naresh Muppalla, Customer Success Manager at BusinessOnBot

~4

Years in customer-facing SaaS

42

D2C brands I own today

82+

Brands fixed by one root-cause PR

4

People on the support team I lead

7

Live sites designed & built

(03)Journey

From support tickets to owning outcomes.

Read the full journey →
  1. 2016 – 2020

    BCA, Computer Applications

    St. Philomena's College, Mysore

  2. Dec 2021 – Jan 2023

    Technical Support Engineer

    Tata Consultancy Services

  3. Feb 2023 – Apr 2025

    Customer Success Engineer

    Tata Consultancy Services · client Trellix

  4. 2025 · ~7 months

    Higher-education plans in Ireland

    A planned detour

  5. Dec 2025 – Present

    Customer Success Manager

    BusinessOnBot

(04)Case files · how I work

I don't guess. I trace it.

Every tricky ticket goes through my AI ops workspace. Claude Code reads the company codebase (read-only), queries New Relic for proof, and I turn what I find into a fix, a customer reply and a Jira report. Five recent cases, with brand names removed.

  1. Ticket
  2. Codebase
  3. Logs
  4. Fix
  5. Reply
  6. KB + Jira
company repos I read (read-only)
28company repos I read (read-only)
bug reports (CRB) in my log
7bug reports (CRB) in my log
feature & product reports (CFR/PFR)
5feature & product reports (CFR/PFR)
recurring patterns tracked
9recurring patterns tracked
Case 01Agent routing
  • New Relic NRQL
  • Codebase
  • Admin panel
routing · last 2 days
  • No available agent with less than 100 tickets7,806×
  • assigned (13 Oct 16:00 → 14 Oct 10:00)0
  • cap raised → next run15 assigned
  • waiting15 → 0

Tickets stuck unassigned for a day

Symptom
New support tickets sat in the queue overnight. Agents were online, but nothing was being assigned.
What I found
Auto-assignment caps every agent at 100 open chats. Every online agent was at or above the cap, so the router skipped them all, thousands of times.
Fix
Raised the cap for the brand, checked that the save hit the right account (the first one didn't), and asked agents to close finished chats.

✓Queue went from 15 waiting to 0 in the next 2-minute routing run.

Case 02WhatsApp templates
  • Codebase
  • New Relic NRQL
  • Meta error codes
provider error
  • error#132000
  • body params sent1
  • body params expected0
  • other brands, same error1 found

Automation failing with Meta error #132000

Symptom
An abandoned-cart automation kept failing with “Number of parameters does not match”. The dashboard only said “Failed due to API provider”.
What I found
Traced the send path in code, then confirmed in the logs: the automation sent 1 body variable, but the approved template expected 0. The saved copy was out of date.
Fix
Re-map the automation to the current template. Then I searched every brand for the same error and flagged a second brand proactively.

✓Root cause confirmed from logs, plus 4 product improvements proposed (e.g. validate params on save and before send).

Case 03AI agent
  • New Relic NRQL
  • Codebase
  • Escalation
ai-agent · today
  • 429 prepayment credits depleted753×
  • first seen15:54 IST
  • brands affected (1 h)15
  • chats affected (1 h)185

AI agent replying “Sorry, I couldn't process that”

Symptom
A brand reported their WhatsApp AI agent had stopped answering and was sending the fallback message to every shopper.
What I found
Not the brand's setup. The AI provider was returning 429 “prepayment credits depleted” for the whole platform, hitting 15 brands and 185 chats within the hour.
Fix
Escalated to engineering as a platform outage with the exact error, the start time and a ready-made query. Sent the brand a calm holding reply.

✓Reframed one brand's complaint as a platform outage with evidence, and proposed a low-balance alert so it can't happen silently again.

Case 04Custom-store webhooks
  • New Relic NRQL
  • Webhook logs
webhooks · 30 days
  • abandoned-cart events after 18 Sep0
  • order events (same period)76 ✓
  • abandoned-cart cronevery 4 min ✓
  • causestore stopped sending

Abandoned-cart reminders went silent

Symptom
A custom-store brand had no abandoned-cart messages for three weeks and assumed our system was broken.
What I found
Our side was healthy (the cron ran every 4 minutes). The store had simply stopped sending abandoned-cart events after one date, while its order events kept arriving.
Fix
Gave the brand's developers the exact last event, the endpoint and a 4-point checklist, so the fix went to the right team first time.

✓No guesswork and no blame: the evidence pointed straight at the one integration that had stopped.

Case 05Instagram API
  • Codebase
  • New Relic NRQL
  • Meta Graph API
graph.instagram.com
  • comment webhooks for the reel168 ✓
  • GET media → 400 IGApiException126×
  • code / subcode100 / 33
  • BoB trial-reel filternone

A reel missing from comment automation

Symptom
A brand couldn't attach a comment automation to one reel. An earlier reply had blamed a “trial reel setting” that didn't exist.
What I found
There's no trial-reel logic in the code. Instagram's API refused to return that reel (error 100/33) while still sending us its comments.
Fix
Corrected the earlier reply, wrote up a safe workaround for engineering and drafted the escalation to Meta.

✓Earlier reply corrected, plus a permanent product fix proposed (add a reel by link or ID).

(05)What I believe

“Fix the problem once, at the root. Then write it down so nobody has to fix it again.”

01

Root cause over workaround

Solve it once, properly, even if that means reading the code.

02

Documentation is a product

Good self-service beats a fast reply.

03

Value, not complexity

Use the simplest, most cost-effective tool that does the job well.

(06)The stack

The tools I work in every day.

Support desks, AI agents, debugging tools and the build stack behind nine live sites. Hover any tile to see how I use it.

Tap a tool

See where and how I use it.

32 tools · See them in action in the Lab →

(07)Feedback

Didn't work out? Tell me why.

If you've interviewed me or reviewed my profile, honest feedback helps me improve, even if the answer was no. It takes two minutes and can be anonymous.