One Research Internship: What “Specific” Really Means
Here’s a set of edits I shared before. The original was a single line:
Conducted in-depth financial analysis, secondary research, and due diligence on companies in the xxx sector.
When I revise something like this, I take the three big terms in the sentence and question each one separately.
Financial analysis: which years of financials? Which metrics did you compare? Did you build forecasts, or mainly analyze historical data? If you only looked at ratios, the next question is what those ratios told you.
Secondary research could mean researching the company, the industry, the supply chain, or competitors, or it could mean looking at opportunities to enter a new market. Write only the term, and the reader doesn’t know what you actually looked into.
Due diligence is broad too. Checking public information, making site visits, and interviewing employees and management are different kinds of work. You have to say which of them you actually took part in.
The revised version in the original material expanded it into three lines:
Drafted a 20-page research report covering five leading global firms and emerging Chinese companies in the industry.
Attended more than 20 industry meetings and identified increased investment in lithium battery production linked to favorable government policies, highlighting potential investment opportunities.
Recommended an investment in a company specializing in small lithium batteries based on its market position and growth potential.
These lines were written from the information in the original case, with no new results added. Don’t copy the page count or the number of meetings. Look instead at what job each piece of information does: the first line sets out the scope of the research, the second covers what was found and on what basis, and the third gives the judgment.
The lines also don’t say how much the investment eventually made, because the original material didn’t include that result. Making the existing content specific is already enough for a reader to understand the experience.
A Full Case: Turning “I Used EDA” into Resume Lines, Then Preparing for the Interview
Below is a student’s telecom customer churn analysis. I’ve organized what came out of my questions in class into a complete process, so you can follow along and dig into your own experience the same way. The English at the end is a teaching example I wrote from this information. It isn’t a claim about the final resume the student submitted at the time.
The original problem: only tools and actions.
The student’s original description said, roughly, “used EDA to study behavioral differences across customer groups and built a churn prediction model.” That shows the general direction, but not the scope of the project or what she got out of the analysis.
I first asked what the data was and what problem the project was solving, then how much data there was. She said 7,043 records and 20 variables. This project used a public sample dataset. Just because it involves a telecom company’s business scenario doesn’t mean you can present it as work for a real employer or as a system that went live.
Next I asked: how did you segment the customers, and which ones were more likely to churn? If there were differences, how did you see them in the data? Only then did we get to handling class imbalance, comparing models, and final performance.
Here’s what we pulled together in class. She looked at customers’ payment methods and service usage and found that some groups were more likely to churn. She applied SMOTE, and she compared models, reporting about 84% accuracy for XGBoost.
Each piece of information has a different use.
| What we had | What it does in your materials |
|---|---|
| Public telecom churn data, 7,043 records, 20 variables | Establishes the type and scope of the project |
| Looked at differences by payment method, service usage, and so on | Shows how the analysis was done |
| Identified groups with different churn tendencies | Presents what the analysis found |
| Handled class imbalance, compared models | Explains technical choices and validation |
| Reported about 84% accuracy in class | Can serve as an in-project evaluation result, but you need to be able to explain how you got it and what it applies to |
Based on this, a first pass at two example lines could look like this:
Analyzed a public telecom churn dataset with 7,043 records and 20 variables, examining differences in churn across payment methods and service usage patterns.
Applied SMOTE to address class imbalance and compared classification models, with XGBoost achieving approximately 84% accuracy in the project evaluation.
This version first makes the project’s nature, scope, and methods clear. To turn the first line into a business finding, you’d need to go back to the results, confirm exactly which customers and how large the differences were, and then add a conclusion you can support. Don’t turn a fuzzy memory into a precise number.
Once the resume is done, interview prep goes one level deeper.
An interviewer might ask why you focused on these variables, how you defined churn, what the original class distribution looked like, why you chose SMOTE, which models you compared, and where the evaluation results came from. If you mention accuracy, you also need to know what it shows and which questions that number can’t answer.
When the interviewer asks about business implications, discuss which customers these differences would lead you to focus on first and what you’d want to test next. But if no retention campaign actually ran, you can’t write up the analysis as “reduced the company’s churn rate.”
You’ll find that a project that looks very ordinary has no shortage of things to say. What you need to do is connect every “I used X” to “why I used it, what I saw, and what judgment it supports next.”