Incubated by NSRCEL · IIM BangaloreCausal AI · Built in India
Causal AI that names the lever behind your numbers, with an estimator your data lead can audit. Built for Indian founders and the analysts who report to them.
Recent work
Four engagements · audited Q1 2026
Video-ad drop-off named
Identified the exact frame where viewers leave. 40% Meta-spend savings projected.
Read case studyReturn on engagement
41 dormant accounts reactivated. ₹15L+ recovered in two weeks.
Read case studyMarket sized
Investor-ready 90-day GTM, with the math behind the addressable market.
Read case studyCampaigns audited
CPL projected to drop 40% across two cities. The exact creative pulled, named.
Read case studyHow it works
We don't sell a tool you have to learn. We sell an answer you can forward. Here's how a typical engagement runs.
Sales, marketing spend, inventory, customer logs, support tickets. Tabular, time-series, whatever shape your stack happens to produce. No clean room required.
Correlation tells you which numbers travel together. Causal inference tells you which one, if changed, would have changed the result. We use the established estimators (DoWhy, EconML, frontdoor and backdoor adjustment) and we show our work.
The deliverable is a short list of decisions, each with the expected effect, the confidence interval, and the trade-off. The kind of memo a founder forwards to a CFO without rewriting it.
Why this is different
Standard analytics is a chart of the past. Causal AI is a model of cause and effect. Same data, completely different deliverable.
What most analytics tools do
What we do
Worked example · 10-min delivery economics
The shape every Indian operator already knows from Zepto, Blinkit, Swiggy Instamart. Hover any edge to see the estimated effect; hover a node to see what it represents in the model.
Hover an edge for the causal claim and effect size. Hover a node for what it represents in the model.
Estimator: DoWhy backdoor + frontdoor adjustment with IV check (Pearl 2009). Synthetic example calibrated on Indian quick-commerce benchmarks.
See more use cases →Working with
“They watched our video ads frame by frame and pointed to the second viewers drop off. We pulled three creatives that same week. Meta spend was projected 40% lower by next quarter.”
“Hundreds of dormant accounts and no one to call first. They ranked the list with a model, not a guess. ₹15L came back in two weeks.”
“Most agencies hand you a deck. ProjektAnalytics handed us a shortlist of decisions, each with the expected lift and the math behind it. The ones we acted on landed where they said.”
“They didn't pitch a tool. They sat with our data, named the lever, and showed us what to expect when we pulled it. That is the team you want, not the dashboard.”
Five questions you'll ask
Short answers below. Longer answers on the call. If yours isn't here, ask it directly.
BI tools chart what your KPIs did. Analytics consultancies write a deck about it. We do something narrower and more useful: we name the variables that caused the result, write down the assumptions, and run an estimator (DoWhy, EconML) that anyone with a stats background can audit. The deliverable is a list of decisions, not a dashboard you have to learn.
Most in-house data teams are full to the brim with reporting, instrumentation, and pipeline work. Causal inference is a different discipline with its own tooling and a steep ramp. We work with your data lead, not around them. They keep the keys; we bring the playbook.
Where to spend the next ad rupee. Which onboarding step is silently driving churn. Which city, channel, or cohort is masking a 40% margin leak. Whether the price change you're considering will move volume or just margin. The pattern: any decision where you have data but the data doesn't explain itself.
A first audit lands in two to three weeks. The Gangpur Ventures case study (linked above) was a 14-day engagement with a documented 65x ROI. Longer engagements turn into a retainer once the first audit pays for itself, which is the only sane way to price this work.
Yes. Book a 30-minute walkthrough and we'll trace one of our case studies end to end: the dataset, the DAG, the estimator, the assumptions, the sensitivity check, and the call we made. If your data lead pokes a hole, we'll tell you so.
