A consumer research project I built as part of my application to Decision Lab's Market Research Consultant Programme (May 2026). The goal was to demonstrate the full consulting capability chain: structure a business problem → design research → analyse data → produce a commercial recommendation.
The hypothetical client is Techcombank, one of Vietnam's leading private commercial banks. The research question: who are the next-wave adopters of Techcombank's digital banking platform, what stops them from converting, and what should Techcombank do about it?
├── data/
│ └── Findex_Microdata_2025_updateViet_Nam.csv # World Bank Global Findex 2021, Vietnam module
├── scripts/
│ ├── phase3_step1_explore.py # data loading, SQLite setup, SQL-based demographic analysis
│ ├── phase3_step2_regression.py # logistic regression, marginal effects, decision tree, segment profiling
│ └── phase3_step3_excel.py # styled Excel workbook output
├── outputs/
│ |
│ └── Techcombank_Digital_Adoption_Analysis.xlsx # analytical workbook (3 sheets)
└── README.md
World Bank Global Findex 2021 — Vietnam module
- 1,000 nationally representative Vietnamese adults (998 usable after cleaning)
- Individual-level survey: account ownership, digital payment use, saving, borrowing, demographics
- Free download from microdata.worldbank.org (catalog 5861), requires registration
- This is the main analytical dataset
Decision Lab Connected Consumer Q4 2025
- Quarterly tracker of Vietnamese digital platform usage, n = 1,427
- Used for market context — platform penetration trends, generational breakdowns
- Free PDF from decisionlab.co/library
Decision Lab AI and the New Financial Decision Journey (Dec 2025)
- Consumer financial decision journey research, n = 894
- Used for trust barrier evidence and journey-stage analysis
- Free PDF from decisionlab.co/library
Decision Lab State of Consumer AI in Vietnam (Aug 2025)
- AI adoption and barrier research, n = 600
- Used to corroborate age and education patterns found in Findex
- Free PDF from decisionlab.co/library
You need Python 3.x with the following libraries:
pip install pandas numpy statsmodels scikit-learn openpyxl
Put the Findex CSV in the same directory as the scripts and run them in order:
python phase3_step1_explore.py # creates findex_vietnam.db + prints demographic tables
python phase3_step2_regression.py # runs the regression, decision tree, segment profiles
python phase3_step3_excel.py # generates the Excel workbookScript 1 needs to run first because it creates the SQLite database that scripts 2 and 3 load from.
Three findings that matter commercially for Techcombank:
1. The access paradox. The next-wave adopter segment — banked Vietnamese adults who haven't gone digital yet — has 90.8% internet access. The barrier isn't infrastructure. It's behavioural inertia. Internet access raises adoption probability by 23.5 percentage points in the regression (the single biggest predictor), but the next-wave segment already clears that bar. The problem is the absence of a compelling first-use trigger.
2. The formal saver is the highest-value target. Saving formally with an institution raises adoption probability by 12.8pp — statistically identical to moving up one full education level (also 12.8pp). 60.5% of the next-wave segment already saves formally. These are existing Techcombank customers who haven't been activated digitally. This is a retention and activation play, not cold acquisition.
3. Age beats geography as a segmentation axis. Rural consumers adopt at 68.1% vs urban at 57.5% — geography isn't the barrier and isn't statistically significant in the regression (p = 0.47). Age is. Adoption peaks at 83.3% for 25-34 year olds then drops to 18.3% for 55+. The 45-54 cohort at 59.1% is the tractable swing segment — still above 50%, likely already in a Techcombank relationship, but not yet digitally active.
Model performance: logistic regression AUC = 0.93, pseudo R² = 0.37. Decision tree AUC = 0.90 for comparison.