Shubham Bubna

I find the one finding that changes the decision.

Customer research that companies actually act on, from the survey design to the one slide that changes the roadmap.

See my work Get in touch
Things I shipped

A browser extension, an organ, and a pile of spreadsheets.

Chrome extension, built solo, early access

de-slop

A Chrome extension that rewrites AI slop into writing a human would actually say.

Highlight anything you typed, pick a tone, and it rewrites the corporate sludge in place. Three settings run from Ghost to Casual to CEO, with a built-in em-dash and stock-phrase killer for the words nobody means. Built solo, end to end, including the Stripe billing.

Go de-slop something
Solo buildStripe billingTone rewritingEm-dash killer
The one that wasn't a side project

Part of my liver

My most personal project, for the most important person.

I donated part of my liver to my mother. The surgery went through, the recovery hurt more than I will pretend it didn't, and I would do it again tomorrow without a second thought. My life is hers. Some things you do not weigh, you just do, for the person who gave you yours. This is the one I am proudest of, and it has nothing to do with work.

Five years, the day job, genuinely useful

Boring work projects

The unglamorous research that actually paid the bills.

Five years of regressions, MaxDiff studies, and satisfaction trackers that told large companies what to fix and what to leave alone. Not dinner-party material, but real, and it changed what they did next. Click through for four of them, numbers and all.

The actual projects
ProblemA subscription service watched CSAT fall from 74 to 62. Leadership was certain billing clarity was the culprit and was ready to pour the budget into fixing it.
ApproachRather than trust the loudest hunch, I ran multiple regression across 14 constructs and read the standardised betas as derived importance. That separates what customers say bothers them from what statistically moves the score, which a simple satisfaction ranking never could.
FindingBilling clarity came in at 0.31. Response time at 0.61. Personalisation at 0.67. The thing they wanted to rebuild was the weakest driver in the model.
OutcomeThey redirected the spend toward personalisation, and the next wave recovered from 62 to 71.
Derived importance (standardised beta)SPSS
Billing clarity (client's hunch)0.31
Response time0.61
Personalisation (actual driver)0.67
They wanted to rebuild billing. The numbers said spend on personalisation. The weakest driver was the one they nearly funded.
CSAT across wavestracker
SPSSMultiple regressionDerived importanceTracker design
ProblemTwelve features on the table, and a client convinced price was the deciding factor in conversion.
ApproachRating scales let people call everything important and quietly inflate price sensitivity, so I ran best-worst scaling in Qualtrics to force genuine trade-offs. People had to pick a best and a worst, which exposes real priority instead of polite agreement.
FindingPersonalised workout plans ranked first, nutrition guidance second, and price came third.
OutcomeThey rebuilt the plan tiers around the top two, and conversions rose after relaunch.
What people actually chose (rank)best-worst
1
Personalised workout plansmost wanted
2
Nutrition guidance
3
Pricethe client's bet
How the study was set upMaxDiff
12features evaluated
Best-worst scalingforced trade-offs, not ratings
2top features chosen for the rebuild
MaxDiffBest-worst scalingQualtricsTrade-off analysis
ProblemA financial institution was sure salary was driving people out, and the instinct was to simply raise offers.
ApproachI tested the salary theory directly across 300 employees, controlling for the other factors so a single loud variable couldn't take the credit. Hypothesis testing alongside regression keeps you honest about what is significant and what only looks that way.
FindingSalary was not statistically significant once the other drivers were controlled for. Career development and training were.
OutcomeThey built real progression paths instead of just raising offers, and attrition fell 20% within 12 months of acting on the finding.
Leadership's theory
Salary
Not statistically significant once the other drivers were controlled for.
What the data showed
Career development
Training
Tested vs significanthypothesis testing
DriverTestedSignificant
Salary✓✕
Career development✓✓
Training✓✓
20% lowerattrition within 12 months of acting on the finding
Attrition, indexed (100 = before)before vs after
Before100
After80
Indexed to the starting rate. Building real progression paths brought attrition down by 20% within twelve months.
Hypothesis testingRegressionPeople analyticsSignificance testing
ProblemWork out which product and brand factors actually relate to whether someone buys from a given smartphone retailer, rather than the factors people claim matter.
ApproachSurveyed consumers, built a multiple linear regression in SPSS across 11 predictors, then cross-checked each factor with chi-square tests, reading standardised betas for relative weight and p-values for significance. Running both methods means the relationships have to survive more than one test before you trust them.
FindingBrand image and after-sales service carried the heaviest standardised weights, both negative in this sample, with price change close behind. The full model explained about 30% of the variance. No single factor cleared significance on its own.
OutcomeA clean reminder that detecting real effects needs a larger sample. The value was in running the full pipeline end to end: hypotheses, SPSS modelling, beta and p-value interpretation, and the write-up.
Coefficient matrix (six predictors)SPSS
PredictorStd. betaSig.
Brand image↓-0.28n.s.
After-sales service↓-0.26n.s.
Price change↓-0.20n.s.
Credit facilities↑+0.16n.s.
Price consideration↑+0.14n.s.
Competitor pricing↑+0.11n.s.
Amber cells pull purchase down, blue cells push it up, shaded by absolute weight. Significance reads n.s. for every row: no single factor cleared significance on its own.
Standardised beta, by directionSPSS
pulls down negativepositive pushes up
Brand image-0.28
After-sales service-0.26
Price change-0.20
Credit facilities+0.16
Price consideration+0.14
Competitor pricing+0.11
0.30R squared (variance explained)
11predictors tested
The honest takeaway. No single factor cleared significance on its own, a clean reminder that detecting real effects needs a larger sample. The value was in running the full pipeline end to end: hypotheses, SPSS modelling, beta and p-value interpretation, and the write-up.
Multiple regressionChi-squareSPSSBeta and p-value reading
The toolkit

The right tool for the question.

Statistics earn their keep only when they answer something real. Here is what I reach for, and what each one is actually asking.

Key driver analysis

What actually moves the score, not what customers claim moves it.

Multiple regression

Which factors drive the outcome, and how much each one really weighs.

MaxDiff (best-worst scaling)

What people truly prioritise when they can't rate everything a five.

Hypothesis testing

Whether a difference is real or just noise wearing a confident face.

Text analytics

What the open-text comments are really saying, once you read every one at scale.

Segmentation

Who these customers actually are, grouped in ways that change decisions.

Statistics

SPSS
R
Python
Stata

Survey

Qualtrics
Alchemer
Microsoft Forms
Google Forms
Medallia
SurveyMonkey

BI and visualization

Power BI
Tableau
Advanced Excel

Data

SQL
Excel
About

Hi, I'm Shubham.

Shubham Bubna, painted portrait
Toronto, Canada
CFA Level 2, PG, Business Analytics

I help product, CX, and insights teams find the one finding that changes the roadmap, not just the report.

Numbers aren't a second language for me. They're the first one.

Five years at NetSolz running research end to end: survey design in Qualtrics, key driver analysis in SPSS, text analytics, and readouts for people who had ten minutes and needed a straight answer. A PG diploma in analytics, a finance degree, and CFA Levels 1 and 2 cleared. Level 3 and I are still in mediation.

I moved from India to Toronto, applied to an unreasonable number of jobs, and somewhere in the middle, tired of reading messages that said nothing, I built de-slop. That is roughly the level of annoyed it takes to get a product out of me.

What I want next is a research or consumer-insights role where the findings actually change what the company does. Agency or client-side, in Toronto.

The path

Where I've worked, what I've studied.

Experience
2026 to now
Freelance Researcher
Independent
Taking on consumer and insights research directly with teams: survey design, analysis in SPSS and Excel, and clear readouts, project by project.
2023
Donated part of my liver
To my mother
Living-donor surgery. She needed part of a liver, I had one to share. The most important thing on this page, by a distance.
2020 to 2025
Senior Research Analyst
NetSolz
Ran 20+ tracker projects a year end to end: Qualtrics survey design, key driver analysis and regression in SPSS, text analytics on open-text, and stakeholder readouts. Findings adopted in 90%+ of client engagements, with zero client escalations.
2018 to 2020
Research Analyst
Eximius Research
Designed questionnaires, sampling plans, and interview guides across sector programs. Ran cross-tabs and trend analysis in SPSS and Excel, turning consumer attitudes and satisfaction data into evidence for product positioning and customer experience.
Education
2021 to 2022
Post Graduate Diploma, Business Analytics
BITS Pilani, India, evaluated by WES
2018
CFA Level 2, passed
Chartered Financial Analyst program (Level 3 and I are still in mediation)
2017
CFA Level 1, passed
Chartered Financial Analyst program
2015 to 2018
Bachelor of Commerce (Hons.)
University of Calcutta, India, evaluated by WES
Contact

Let's find your finding.

Open to research and consumer-insights roles in Toronto, agency or client-side. Tell me what your data won't admit, and I'll tell you what it actually says.

Email me Connect on LinkedIn