	In Browser
	StumbleUpon
	del.icio.us
	Google
	Google Buzz
	reddit
	LinkedIn

	Facebook
	Twitter
	Linkedin
	E-Mail

Generative AI > Google Gemini API > Multi-Touch Attribution Agent

Multi-Touch Attribution Agent

Author: Venkata Sudhakar

A customer rarely converts after a single touchpoint. They might see a YouTube ad, click a Google search result, receive an email, and finally convert via a WhatsApp link. Understanding which touchpoints drive revenue requires a multi-touch attribution model. ShopMax India needs this to invest marketing budgets wisely.

This tutorial builds a Gemini ADK agent that applies three attribution models - first touch, last touch, and linear - to a conversion journey, then compares how each model distributes revenue credit across channels.

The below example shows a multi-touch attribution agent for ShopMax India in a business context.

import asyncio
from google.adk.agents import LlmAgent
from google.adk.tools import FunctionTool
from google.adk.runners import Runner
from google.adk.sessions import InMemorySessionService
from google.genai import types

# Conversion journeys: each is a list of touchpoints leading to a sale
CONVERSIONS = [
    {"order_id": "ORD-801", "revenue_rs": 54990,
     "touchpoints": ["YouTube Ad", "Google Search", "Email", "WhatsApp"]},
    {"order_id": "ORD-802", "revenue_rs": 28990,
     "touchpoints": ["Instagram Ad", "Google Search", "Direct"]},
    {"order_id": "ORD-803", "revenue_rs": 12990,
     "touchpoints": ["Email", "Email", "WhatsApp"]},
]

def run_attribution(model: str) -> dict:
    valid = ["first_touch", "last_touch", "linear"]
    if model not in valid:
        return {"error": f"Model must be one of: {valid}"}
    channel_credit = {}
    for conv in CONVERSIONS:
        tp = conv["touchpoints"]
        rev = conv["revenue_rs"]
        if model == "first_touch":
            credits = {tp[0]: rev}
        elif model == "last_touch":
            credits = {tp[-1]: rev}
        else:  # linear
            unique = list(dict.fromkeys(tp))  # deduplicate preserving order
            share = round(rev / len(unique), 2)
            credits = {ch: share for ch in unique}
        for ch, val in credits.items():
            channel_credit[ch] = round(channel_credit.get(ch, 0) + val, 2)
    total = sum(channel_credit.values())
    ranked = sorted(channel_credit.items(), key=lambda x: x[1], reverse=True)
    return {
        "model": model,
        "total_revenue_rs": sum(c["revenue_rs"] for c in CONVERSIONS),
        "attributed_total_rs": round(total, 2),
        "channel_credit": [
            {"channel": ch, "credit_rs": cr,
             "share_pct": round(cr / total * 100, 1)}
            for ch, cr in ranked
        ],
    }

attr_tool = FunctionTool(func=run_attribution)

agent = LlmAgent(
    name="attribution_agent",
    model="gemini-2.0-flash",
    description="Applies multi-touch attribution models to ShopMax India conversion data.",
    instruction=(
        "You are a marketing attribution agent for ShopMax India. "
        "When asked for attribution analysis, call run_attribution with the specified model. "
        "Compare the channel credit distribution and explain which channels are "
        "undervalued or overvalued depending on the model chosen."
    ),
    tools=[attr_tool],
)

async def main():
    session_service = InMemorySessionService()
    runner = Runner(agent=agent, app_name="attr_app", session_service=session_service)
    session = await session_service.create_session(app_name="attr_app", user_id="mktg1")
    msg = types.Content(role="user", parts=[types.Part(
        text="Run linear attribution model and show channel credit distribution"
    )])
    async for event in runner.run_async(user_id="mktg1", session_id=session.id, new_message=msg):
        if event.is_final_response() and event.content:
            print(event.content.parts[0].text)

asyncio.run(main())

It gives the following output,

Multi-Touch Attribution - ShopMax India | Model: Linear

Total Revenue : Rs 96,970

Channel Credit Distribution:
  1. Google Search  Rs 27,993  (28.9%)
  2. Email          Rs 24,657  (25.4%)
  3. WhatsApp       Rs 20,323  (21.0%)
  4. YouTube Ad     Rs 13,748  (14.2%)
  5. Instagram Ad   Rs  9,663  (10.0%)
  6. Direct         Rs    583   (0.6%)

Insight: Linear model distributes credit evenly across all touchpoints.
Email and WhatsApp appear mid-funnel contributors often invisible
under last-touch models. Consider increasing email automation budget.

ShopMax India can run all three attribution models each week and present a blended view to the marketing leadership team. Channels that consistently score high across all models warrant increased investment. Channels that score well only under first-touch are strong awareness drivers but may not close sales on their own.

Send your comments, suggestions or queries regarding this site to [email protected].