A modern statistics and AI boutique.
Sales dip, customers go quiet, a new price goes live. We show you what the numbers really say, what is at stake and where to act.
Our work spans payments, healthcare, real estate and mobile games, and the statistics apply wherever you have data.
Clients include
Fractal
Solventis
OmniXM
Apickle
A bad week is not always news. You see which accounts have moved outside their normal range, and which are only having a quiet stretch.
IllustrativeOne fictional customer, week by week. A flag says something changed, not why.
Fractal and Sky Systemz
One live system tracking $100M+ in yearly transactions, with a health score model that tells account managers and sales reps which clients need attention.
“Blinko delivered exceptional work on a highly complex problem. Deep expertise in data science, machine learning, and business domain knowledge – exceeded all expectations.”
2×
clients assessed with the same team
Solventis
40%
more financial reviews completed
Apickle
~2,400
hours saved a year, estimated at handover
Real estate agency
“The statistics for our project were complex - many rejected it, but Blinko took it on.”
When the board asks how sure you are, you have a number you can defend. Credit, fraud, customer loss and revenue risk.
When a journal or an investor asks how sure you are, your evidence holds up. Study design, patient outcomes and testing of clinical AI.
Customers leave, demand moves and decisions need evidence in any business with data. Finance and healthcare are where we show the most, not where we stop.
Also in our work:
Published research
First author. Oxford University Press, on behalf of the European Society of Human Reproduction and Embryology.
Published research
Co-author. Journal of Ophthalmic and Vision Research. A preliminary model, studied in 527 subjects.
A simple first step for any business that suspects revenue is slipping and cannot yet see where, or how much.
The diagnostic shows where revenue is falling and how much is exposed. Historical data rarely proves why, or which action will work. Forecasting models, pricing tests and live alerts are a separate phase, if the findings justify one.
Sample report
Illustrative: synthetic data, not a client
A fictional payments business: 3,960 merchants, over 36 months. Reported attrition is 8.1%. The book lost 14.6% of its revenue.
Where the apparent $690K decline went
Real decline, $543K
Not a decline, $147K
Estimated exposure: $555K over the next 12 months
Assumes current loss rates hold in each segment. An estimate, not a prediction of what will happen.
Prioritized next steps
Needs a later phase: a rate-cut test. Past cuts went only to merchants who asked, so history cannot show whether they work.
At stake means yearly revenue on those accounts, not a predicted saving.
Account for seasonality, changing customer behavior and imperfect data before flagging a problem.
Evaluate models using only the information that would have been available when the decision was made.
Use experiments or appropriate causal methods to assess whether an action improved the outcome, and say where the evidence remains uncertain.
Connect results to your existing tools, with clear ownership, review steps and monitoring.
Tell us what you are seeing in the numbers. A statistician will tell you, honestly, whether your data can answer it.
NDA first. Read-only access or a secure export. We keep no copy of your data after the project.
