Find out what is really moving your numbers, what to do next, and whether it worked. For teams in payments, SaaS, operations and healthcare.
We build it with statistics, causal inference and AI: from assistants that answer from your documents to models your team checks every morning.
Clients include Fractal, Solventis and OmniXM.
400 customers, one mark each, sized by revenue. Who is about to leave, and what it would cost.
Six services. Each one starts from a business question and ends in something your team uses every week.
Your team asks questions in plain English and gets answers from your own documents, with the source quoted.
Typical first version: 3 to 6 weeks
See how it works: AI assistants on your documentsRepetitive work done by AI agents, with a person approving anything that matters.
Typical first version: 3 to 6 weeks
See how it works: AI agents and automationForecasts of demand, churn, risk or revenue that your team can plan with.
Typical first version: 2 to 6 weeks
See how it works: Predictive modelsClear answers to what changed, by how much, and whether it is more than noise.
Typical first version: 1 to 3 weeks
See how it works: Statistics you can defendKnow whether a price change, an offer or a program actually caused the result, and what it was worth.
Typical first version: 2 to 4 weeks
See how it works: Measuring what really workedYour systems connected into one live view, so every team works from the same numbers.
Typical first version: 4 to 8 weeks
See how it works: Analytics systemsNot sure which one fits?
For Fractal we connected payments, sales and operations into one live system. Every role sees its numbers as they happen, and alerts explain themselves in plain English.
Read the Fractal case studyFractal, construction fintech, United States. Figures from the case study.
Blinko delivered exceptional work on a highly complex problem. Deep expertise in data science, machine learning, and business domain knowledge – exceeded all expectations.
Nearly 100 client engagements delivered since 2023.
Solventis, consumer fintech
2×
clients assessed with the same team
Real estate agency, real estate
~2,400
hours saved a year, estimated at handover
Apickle
40%
more financial reviews completed
Automated loan assessment and prioritization increased the number of customers processed through financial review.
Most reports stop at what happened. You also get the next two answers: what to do about it, and whether it worked. Pick a problem to see it worked through.
400 customers, one dot each. The 23 in red are pulling back faster than their normal pattern.
What is happening?
You see which customers are pulling back, why, and what each one costs you, while there is still time to act.
Each dot is a customer, placed by how much its activity changed in the last 12 weeks. 23 are pulling back well beyond normal, putting $1.9M of revenue a year at risk. 14 more are only having their usual seasonal dip, so they stay grey.
What should we do?
You know which customers are worth acting on and which to leave alone, with a dollar figure and an honest level of confidence.
The 23 at-risk customers, placed by what acting would earn after its cost. Acting on 9 of them is worth about $340K. For the other 14, the action would cost more than it brings back, so we leave them alone.
Did it work?
You see how much the action really kept, measured against similar customers who did not get it, with an honest range around the number.
See how we measure itTo check the action really works, 400 similar customers were split at random. Of the 200 who got a retention offer, 22 left (11%). Of the 200 who did not, 36 left (18%). The offer kept about 7 more customers in every 100, and the likely range is 1 to 13.
Then AI engineering keeps it running every day: the right person gets an alert, someone on your team approves each action, and every result sharpens the next decision.
Each step stands on its own. Start with the diagnostic, and go further only when it pays off.
How we work, step by step2 weeks, fixed scope
Two weeks on your own data. You get what is driving the problem, what it is worth, the limits of your data, and a practical next step.
One high-value problem solved on your own data first: a model, an AI assistant or a controlled test. We agree up front what it must beat.
First version live in 4 to 8 weeks
The solution, deployed into your existing workflows with integrations, monitoring and alerts, and optional ongoing support.
2 weeks on your own data, with a written readout at the end. Built for teams with data spread across several systems and one decision they cannot yet make with confidence.
Start with a secure export from the systems you already use. We assess feasibility before proposing any integration. NDA signed first.
Read-only access only, and we keep no copy of your data after the project.
In total: 23 customers flagged, $1.9M of revenue at risk a year. Acting on 9 of them is worth about $340K.
Data limits: refund history covers only the last seven months, so refund-driven losses carry a wider range.
You keep the tools you already have. We start with the business question, test what your data can support, and build only what is worth it to you.
Tell us the one decision you would most like to get right. You will get an honest answer on whether your data can support it.
No pitch, just an honest answer. We reply within one business day.
NDA first. Read-only access or a secure export. We keep no copy of your data after the project.
