Skip to content

MadanHaribhat.

I run ecommerce and marketing for a hair and beauty brand across five markets, and I build the AI systems that run underneath it.

See the workWork together
Annual revenue owned
7-figure, end to end
AI systems in production
12 of 15 built
Channel margin
Double-digit, from loss-making
Markets operated
5 under US, EU and Japan rules
Based in
Ireland

22 systems built · 17 running today

GHOSTBOND AI chatbotLive
GHOSTBOND AI chatbot
Amazon Ads dashboardIn build
Amazon Ads dashboard
The GHOSTBOND chatbotLive since March 2026
The GHOSTBOND chatbot
Social and Marketing OSIn use
Social and Marketing OS
AI content governance systemIn use
AI content governance system
Exhibit AResearch
Exhibit A
GHOSTBOND AI chatbotLive
GHOSTBOND AI chatbot
Amazon Ads dashboardIn build
Amazon Ads dashboard
The GHOSTBOND chatbotLive since March 2026
The GHOSTBOND chatbot
Social and Marketing OSIn use
Social and Marketing OS
AI content governance systemIn use
AI content governance system
Exhibit AResearch
Exhibit A
GHOSTBOND AI chatbotLive
GHOSTBOND AI chatbot
Amazon Ads dashboardIn build
Amazon Ads dashboard
The GHOSTBOND chatbotLive since March 2026
The GHOSTBOND chatbot
Social and Marketing OSIn use
Social and Marketing OS
AI content governance systemIn use
AI content governance system
Exhibit AResearch
Exhibit A
Walmart store locatorLive
Walmart store locator
TETHERLive
TETHER
Claude Creator OSIn use
Claude Creator OS
VozaiIn build
Vozai
The Sole StackLive
The Sole Stack
Three Shopify storefrontsLive
Three Shopify storefronts
Walmart store locatorLive
Walmart store locator
TETHERLive
TETHER
Claude Creator OSIn use
Claude Creator OS
VozaiIn build
Vozai
The Sole StackLive
The Sole Stack
Three Shopify storefrontsLive
Three Shopify storefronts
Walmart store locatorLive
Walmart store locator
TETHERLive
TETHER
Claude Creator OSIn use
Claude Creator OS
VozaiIn build
Vozai
The Sole StackLive
The Sole Stack
Three Shopify storefrontsLive
Three Shopify storefronts

There was never a team the size of the work.

I own the P&L for a beauty and hair brand across five markets, and run the ecommerce and marketing under it. The numbers below are the accounts I am accountable for, not ones I advised on. The work still had to happen, so I built the systems that do it: the customer-facing support agent, the ads platform, the reporting and the monitoring underneath.

7figures
Annual revenue
The P&L I own, across five markets
30%net
Margin on it
After cost of goods, fulfilment, ads and returns
656retail media
Walmart stores
Walmart Connect 1P ads. The 3P marketplace is one I registered and run end to end: WFS inventory, listings and ads
53from one
SKUs advertised
The Turnaround: revenue concentration broken out across the catalogue
All 22 systems

Selected work 5 of 10

Selected work.

Production system

A production AI support agent

I ship AI that talks to customers and survives contact with them, including the boring parts: permissions, handover and tests.

Read the case
The live queue, and a conversation at the point it escalates

The live queue, and a conversation at the point it escalates

The agent answers a product question by citing the catalogue, then the customer asks for a person. Top right is the control that hands the conversation to one. Product names and their links are painted out; the rest of the answer is exactly as it shipped.

Status
Live since March 2026
Answer grounding
Retrieval-Augmented Generation. The AI looks up the real answer in your own documents before it replies, instead of inventing one from memory. on real query logs
Escalation
Human takeover, Role-Based Access Control. Who is allowed to do what is set by their job, so a support agent cannot reach anything an administrator can.
Verification
Passing A set of automated checks that run on every change and fail the build if the system stops behaving.
Flagship

The Turnaround

I own a marketplace P&L end to end, and I can take one from losing money to earning it without buying growth.

Read the case
Seller Central, Sep 2024 to Aug 2026 · currency redacted

Seller Central, Sep 2024 to Aug 2026 · currency redacted

Total order items and units ordered are the real figures. Revenue and average sale are painted out, because those are the employer's to disclose and not this site's. The left chart is units a day across two years; the flat line early on is a stock-out.

Channel Profit and Loss. The full account of what a channel earns and what it costs, so it shows whether it actually makes money.
Double-digit net margin
Blended Advertising Cost of Sales. What share of the money made from ads went back out as ad spend. Lower is better: 20% means £20 of advertising for every £100 sold.
Under 20%
Ad spend
Cut by more than half
Revenue concentration
Across 53 Stock Keeping Units. Individual sellable products, each a specific size and variant with its own code.
Delivery

Seven-Day SaaS

My AI work is a standalone capability. This one has no commerce in it at all, and it went from idea to deployed in a week.

Read the case
TETHER, the running app · search and a sourced place card

TETHER, the running app · search and a sourced place card

Every field on the card carries where it came from: verified against a source, or sourced and dated. That was the point of the build, and it is the part that took the time. Deployed and public seven days after the first commit.

Elapsed time
Seven days to live
Commits
374
Scope
A live travel product
Team
Solo, agents governed
Governance

AI safe enough for the whole team

I make AI safe for people who did not build it, which is the difference between a clever demo and something a company can use.

Read the case
The governed pipeline, 66 items in flight

The governed pipeline, 66 items in flight

Every card is bound to a channel and carries a state: idea, brief, drafting, approved. The rules live in eleven skills behind this, so a teammate with no compliance training gets output that already respects them. Nothing moves to a channel on its own.

Rules
Encoded in 11 skills
Safe output
Default for everyone
Review load
Hundreds of hours back
Regulatory risk
Designed out
Paid media

Paid media beyond the marketplace

I run full-funnel paid media across platforms, and I design the measurement that tells you whether it worked.

Read the case
One 4:5 paid social unit from the programme

One 4:5 paid social unit from the programme

The same creative system was cut for Meta, Google and YouTube. What it drove was measured against a store-level geographic holdout rather than against the platforms' own reported credit, which is why the number beside it is a lift figure and not a ROAS.

Channels run
Four, run as one system
Measurement
Advertising is deliberately withheld from some areas so the untouched ones act as a control group. The gap between them is the real effect, rather than what the ad platforms claim credit for.
Targeting
Ads targeted street by street around individual shops, rather than at a whole country.
Creative supply
A 51-Stock Keeping Unit. One specific sellable product: a single size and variant, with its own code. image program

The rest of the work.

5 more, as an index

Systems

What I build.

Live or in use 17In build 3Shipped 1Research 1

Toolkit 20 tools

What I work in.

Claude Codeagent workflows, build
ChatGPTdrafting and research
Antigravityrapid prototyping
ElevenLabsvoice generation
GitHubversion control
Verceldeployment
Amazon Ads APIofficial partner
Helium 10keyword and listing research
ShopifyD2C and B2B
Meta Adsretail-driver campaigns
Claude Codeagent workflows, build
ChatGPTdrafting and research
Antigravityrapid prototyping
ElevenLabsvoice generation
GitHubversion control
Verceldeployment
Amazon Ads APIofficial partner
Helium 10keyword and listing research
ShopifyD2C and B2B
Meta Adsretail-driver campaigns
Claude Coworkday to day
Geminilong-context research
Higgsfieldimage and video
Twiliomessaging and comms
Next.jsapp framework
Google Cloudwhere the agents run
Seller Centralfive marketplaces
Walmart Connectretail media
Klaviyoemail and SMS
Google Adsstore-level geo
Claude Coworkday to day
Geminilong-context research
Higgsfieldimage and video
Twiliomessaging and comms
Next.jsapp framework
Google Cloudwhere the agents run
Seller Centralfive marketplaces
Walmart Connectretail media
Klaviyoemail and SMS
Google Adsstore-level geo

Experience

  1. ● 2023 - present

    Amazon & E-commerce Manager

    GHOSTBOND / Professional Hair Labs, Ireland

    Amazon US end to end including the full ad budget, Walmart Marketplace on WFS, Walmart Connect retail media, five international marketplaces and compliance across all of them.

    Took a loss-making 7-figure channel to double-digit net margin, with blended ACoS down from around 80% to under 20% and ad spend cut by more than half.

    Built the AI systems the operation runs on, from the customer-facing support agent to the reporting and monitoring underneath it.

  2. 2022

    Brand Associate, E-commerce

    Publicis Groupe / Performics, India

    Ran a $200K Garnier Amazon budget at 3x ROAS.

    Built an automated reporting system that was adopted across teams, worth roughly a 40% productivity gain.

  3. 2020 - 2022

    Account Manager, Amazon account

    Cognizant, India

    Brought ACoS from 45% to 22% in eight months and took the category to best seller.

    Grew client budget from $1K to $50K a month. Best Performer award twice.

  4. 2018 - 2023

    Earlier

    Unmatched Studio, Evolusion Innovation, India and Ireland

    Unmatched Studio, 2018 to 2020. Evolusion Innovation internship, Ireland, 2023.

2023

MSc Management & Marketing

University College Cork, Ireland

CUBS Leaders of Tomorrow award.

Contact

Work together.

Commerce

Marketplace profitability audit

I find where the P&L leaks: ad structure, price, fees, SKU mix, and the operational costs that never make it into the spreadsheet.

Commerce

Paid media and measurement

Campaign architecture across marketplace and off-platform, creative systems that keep up with it, and a measurement design that tells you what actually caused the sale.

AI and systems

AI operations for any team

Automation, agent workflows and the governance that keeps them safe once people who did not build them start using them. Commerce optional.

AI and systems

AI product delivery

From idea to deployed system: support agents, internal platforms, research systems. Built, tested and handed over.

Background

I trained as an engineer in Karnataka and came into marketing sideways, through agency work in India. A master's brought me to Ireland and I stayed. The AI half started because the operation needed software nobody was going to build for it, and that half grew its own legs.

Away from a screen I fly drones. Same appeal as the rest of it: you learn the machine, you read the conditions, and the result tells you plainly whether you got it right.

Inchydoney, Co. CorkFlying
Madan Haribhat flying a drone on a beach

Replies come from my own address.

LinkedInBack to top