Data Analyst Case Study: 5 Surprising Google Trends Lessons

A data analyst case study showing how I developed, constructed, and deployed an end-to-end forecasting application, transforming from a business issue to a live, interactive solution.

This data analyst case study describes my first full data analytics project in the same way I would explain it to a hiring manager: the business problem, why I made the decisions I did, what the data looked like, how the model performed, and, most importantly, what I would do differently next time.

Data Analyst Case Study: Project Summary

Live App: keywords-trends-time-series-forecasting.streamlit.app

1. Background & Motivation

I chose this as my first project purposefully. I didn’t want to start with a clean, pre-packaged Kaggle CSV—I wanted to work with something more like what a working analyst would encounter: live, noisy, API-sourced data with no guaranteed structure.

Personally, I am interested in search trends. As someone who is actively looking for work in data analytics, I was curious: is there an increase in need for these skills? Is it seasonal? Can it be predicted with any reliability? That interest sparked the idea for the project.

This data analyst case study also provided me with the opportunity to work across the whole analytics lifecycle, from data collection to implementation.

2. The Business Problem

The goal of this data analyst case study was to turn descriptive search trends into actionable forecasts.

Problem Statement: Analyze Google search interest data and estimate future trends for key technical areas to help with marketing planning, content strategy, and skill demand analysis.

Search interest is a leading indicators that businesses already use informally—content teams track “what’s trending,” and recruiters track which skills are most frequently searched. However, nearly none of this is forecasted in advance. I intended to close the gap by extending the trend line rather than simply showing it.

Hypotheses

3. Data Source & Collection

• Source: Google Trends, accessible using the Pytrends API.
• Granularity: The daily search interest values.
• Time window: Rolling last 6 months, retrieved dynamically at runtime.
• Keywords: Data analytics, machine learning, and artificial intelligence.

I made an early design decision that affected the entire project: the data needed to be live rather than a one-time static export. Every time the program runs, it pulls current numbers from Google Trends, therefore the pipeline has to be adaptable to changing data each time, rather than adjusted to a single constant snapshot.

4. Data Cleaning & Preparation

1. Fetched raw interest-over-time data per keyword using Pytrends.
2. Converted date fields into proper datetime objects for time series compatibility.
3. Audited for missing values and inconsistent entries across the three keyword series.
4. Reshaped each series into Prophet’s required two-column format (ds = date, y = value).

This stage was more important than I expected—Google Trends data isn’t totally uniform across keywords, and getting the structure correct before modeling saved a lot of time debugging later.

5. Modeling Approach

For this data analyst case study, I assessed each keyword individually to see how the forecasting approach worked across different search trends.

I trained a separate Prophet model for each phrase rather than a combined model because the volume and pattern of the three search terms differed. For each keyword, the model divided the series into:

• Trend — the underlying long-term direction
• Seasonality — recurring weekly/periodic patterns
• Noise — short-term fluctuation smoothed out of the forecast

Model performance for each keyword was assessed using RMSE (Root Mean Squared Error), which provides a solid, comparable measure of forecast accuracy across the three series rather than depending just on visual assessment.

Key KPIs Tracked

6. Tech Stack

If you’re interested in how I use Python across different projects, check out my data analytics projects.

7. The Dashboard

The final deliverable is not a notebook, but rather a deployable, interactive Streamlit app created for non-technical users to explore without touching any code. The final dashboard brings the different stages of this data analyst case study together in one interactive application.

• Keyword selector – toggle between the three tracked terms.
• Adjustable time range for the analysis window.
• Historical trend chart
• Forward-looking forecast graphic.
• Component breakdown for trends and seasonality.
• Download forecast output as a CSV file with just one click.

8. Results & Findings

One of the most significant aspects of this data analyst case study was understanding how the technical model translated into obvious business signals.

• Search interest levels differ significantly across the three keywords—they don’t move as one combined “tech trend,” supporting H3.
• Artificial intelligence shows the strongest upward momentum of the three, noticeably outpacing data analytics and machine learning.
• Seasonal patterns were detectable even within a relatively short 6-month window, supporting H1.
• Prophet handled the noise in daily Google Trends data reasonably well once the input was properly structured, supporting H2.
• The short-term forecasts produced clear directional signals (rising/stable/declining) that could realistically inform a content or hiring decision, supporting H4.

The major change for me wasn’t technical; it was learning that a “trend chart” and a “forecast” answer two very distinct business issues. One person tells you what happened. The other advises you what you should do next.

9. Business Impact

• Supports content and marketing calendar planning around rising topics
• Helps forecast which skills are likely to be in higher demand
• Enables earlier identification of emerging topics before they’re obvious from raw numbers
• Demonstrates a deployable, practical use of predictive analytics rather than a theoretical exercise

10. Limitations & Next Steps

Current Limitations

• A 6-month window limits long-term seasonal detection (e.g., yearly cycles)
• Only 3 keywords tracked — not yet generalized to arbitrary search terms
• RMSE alone doesn’t capture all aspects of forecast quality (e.g., directional accuracy).
• No automated model retraining schedule — runs fresh only on app load

What I’d Build Next

• Add user-input custom keywords instead of fixed terms
• Extend the historical window once enough data accumulates
• Add confidence intervals to forecasts for transparency
• Compare Prophet against a baseline (e.g., ARIMA) for model validation

11. What This Project Demonstrates

• Time series analysis and forecasting fundamentals using Python
• Practical experience with Prophet, including trend/seasonality decomposition and RMSE evaluation
• Handling live, dynamically fetched, real-world data—not just static files
• Building and shipping an interactive analytics application end-to-end
• Translating a technical model into a business-relevant, decision-ready tool

Reflection: What This Data Analyst Case Study Taught Me

This was my first project, and I developed it without a roadmap to follow, so many decisions (live data fetching, per-keyword modeling, what to include on the dashboard) were entirely mine to make and defend. That’s the kind of decision I believe is more important than any one tool or library, and it will serve as the guiding principle for any projects following this one.

If you’d like the shorter, story-style version of this project, check out my behind-the-scenes blog post on how this build came together.

Live App: keywords-trends-time-series-forecasting.streamlit.app