At one of our recent Google & Amplitude meetups in Berlin, analytics experts Askar Abdullaev and Sven Herschel gave a presentation about Google Analytics Advisor vs. MCP, through the lens of a real customer case study. They've helpfully distilled those insights into a blog post for us, and this is the result!
Google Analytics recently released the Analytics Advisor, aka "Ask Advisor". It follows a shift in the way data is being analyzed in general. With tedious manual report creation increasingly a thing of the past, many users are looking more towards so-called agentic analytics solutions. Google's Analytics Advisor is one of these: a conversational analytics tool, designed to simplify analytics and make users more independent. The idea is that users can ask questions about their data in plain English and get insights back almost instantaneously.
In this blog post, we want to find the answer to the following questions:
What happens if a user needs more than just standard analytics data?
What if the insights require complex context from, say, a CRM system, or additional info about your profit margins, or even specific market benchmarks?
That's where the Model Context Protocol (MCP) comes into play.
Along the way, we'll explore the real-world strengths and limitations of GA4's native Analytics Advisor, and how MCP is opening up entirely new possibilities for AI-driven decision-making.
The Google Analytics Advisor is a beta feature inside GA4 that brings conversational AI directly into your analytics. Powered by Google Gemini, it understands your tracking data and lets you ask questions like "Why did organic traffic drop last week?" instead of having to manually set up a report to find out.
Understanding performance: Allows natural language queries for quick insights
Investigating anomalies: Detects unusual behavior and comes up with potential reasons for it
Supporting with analyses: Provides step-by-step guidance for creating segments and explorations
Finding opportunities: Gives recommendations where it sees potential for improvement
We asked the Advisor to detect suspicious bot traffic and analyze it. It identified unusual activity patterns based on metrics like session duration, bounce rate, and geographic data concentration. It even cross-referenced suspicious locations against known data center locations, which we thought was a pretty impressive use of context.
The Advisor gave us step-by-step guidance on creating a segment to investigate further. While it didn't fully automate the exploration report creation, it significantly reduced the manual work involved.
We tested its ability to identify Personally Identifiable Information in our dataset. The Advisor automatically scanned dimensions including page location, page path, and query parameters for common PII markers (such as "@" symbols in email addresses).
It then guided us through creating an exploration to visualize potential PII and recommended solutions for minimizing future collection. Impressive, but still no automated data deletion.
Not everything went smoothly. There were clear boundaries where the Advisor couldn't deliver desired outcomes.
We asked it to automatically configure unwanted referral exclusions. Unfortunately, it couldn't do it. The Advisor can't modify GA4 settings; it can only provide a recommendation, so the actual configuration remains a manual process.
When we asked it to analyze revenue decline, the Advisor identified "items purchased" as a driver. But it interpreted this as multiple purchases, when in reality, it was one user placing a single order with 12 items.
The Advisor saw the volume number but missed the nuance: a critical difference when making strategic recommendations.
Data silos: It can't access CRM, ERP, or revenue systems beyond GA4
Context loss: Previous chats and analyses are not available in future sessions
Limited semantic understanding: Custom event parameters can confuse it without explicit guidance
No automation: It advises but doesn't configure or execute changes
Most importantly, it only knows your GA4 analytics data, and nothing else beyond that.
Enter the Model Context Protocol (MCP), an open standard that lets any AI model talk to your analytics data. Unlike the Analytics Advisor, which is locked to Google Gemini and GA4 data, MCP works with any AI model (Claude, GPT, etc.) and can incorporate data from multiple sources.
The key difference: MCP isn't just an analytics tool. It's a bridge that lets AI systems access and reason about data across your entire organization.
We tested MCP with Forever Thirsty, our own online natural wine shop. To get a feel for it at different levels, we asked three increasingly complex questions:
"Can you provide the main insights for May 2026 for property 239442503?"
MCP delivered key insights about revenue, conversions, and active users.
But compared to the Analytics Advisor, the MCP analysis included actionable channel breakdowns, landing page performance, and strategic recommendations, all in one coherent narrative.
Key insight: Revenue and conversions grew much faster than users, indicating significantly improved traffic quality. The homepage alone drove the majority of revenue.
"Can you combine this with official wine industry statistics for the DACH region and make recommendations?"
This is where MCP's power becomes obvious. It synthesized:
Forever Thirsty's May performance data
DACH region wine market trends
Consumer behavior shifts toward "natural wine" and "no-and-low" categories
The insights & specific findings:
The shop was "bucking the trend": while total wine market consumption in DACH is shrinking, Forever Thirsty experienced growth. Why? They're perfectly positioned for the "Drink less, but better" movement. But there is still room for improvement. Non-alcoholic wines are the fastest-growing segment, but the shop's alcohol-free wine page received minimal conversions.
Recommendations generated:
Capitalize on the "no-and-low" boom immediately
Leverage "social commerce" (Instagram Reels, shoppable videos) for natural wine discovery
Focus on Switzerland as a high-value premium market
Optimize product filters for "organic" and "sustainability" certifications
"Can you analyze the uploaded profit calc.csv (some dummy data) together with GA4 data and give recommendations?"
Now MCP demonstrated its true superpower: cross-system analysis.
The reality check:
May has the highest revenue, but the profit picture was different: for every €100 in revenue, the business kept only €9. This is far below the typical margins for boutique wine retailers.
Channel profitability analysis:
Paid Search was the strongest performer in margin, while Direct and Referral underperformed due to smaller orders and heavier discounting.
Why are margins low? Item costs, shipping, and discounts exceeded revenue on individual orders, making them unprofitable.
Data-driven recommendations:
Raise the free shipping threshold
Increase Google Ads CPC budget by 15-20%
Shift SEO strategy toward super-premium wines
We asked the same request multiple times and occasionally received different answers. When you specify your metrics, dimensions, date range, and desired aggregation, you get better results.
It's also crucial to understand that MCP surfaces patterns in data, not business reality. It doesn't know your pricing constraints, brand strategy, or market positioning. Cross-check recommendations with your business team before implementation.
| Aspect | GA4 Analytics Advisor | MCP for GA4 |
|---|---|---|
| AI model support | Google Gemini only | Any model (Claude, GPT, etc.) |
| Data access | GA4 & linked Google services | GA4 & unlimited external data |
| Use cases | Analytics questions within GA | End-to-end business workflows |
| Flexibility | Limited | High. You can add new data sources easily. |
| Implementation | Built-in, no setup | Requires initial setup (technical knowledge) |
| Business value | Analytics insights | Cross-system insights & decisions |
| Cost | Included in GA4 | Implementation effort required |
Use GA4 Analytics Advisor for quick analytics questions, simple data audits, anomaly detection, and learning GA4 faster.
Use MCP for business decisions that require context beyond analytics, such as revenue forecasts that factor in profit margins, or market strategies informed by both data and industry trends.
The future of data-driven decisions isn't about choosing one or the other. We recommend starting with GA4 Advisor for fast analytics wins, then graduating to MCP when you need the full picture.
Both tools share a common truth: the quality of your insights depends on the quality of your questions.
GA4 Advisor might struggle with semantic nuance or multi-source context, but ask the right question and it delivers.
MCP requires more precision but rewards you with cross-system insights that would take a data analyst hours to compile.
The shift from "manual reporting" to "ask and get answers" is real. But the next shift — from "analytics-only" to "business-wide AI reasoning" — is even more significant.
If you're looking to get started with MCP, we're more than happy to support you.