Data analytics and machine learning
Krikir builds data analytics and machine learning for Shopify brands and SaaS teams, numbers first.
Fixed quote after one callReply within one working day
Example: one brand's revenue, from several tools to one number
What we build, in order
Data analytics and machine learning means moving your numbers into one place, agreeing what each one means, and predicting only where a model beats a simple rule. It suits teams with several tools and repeat decisions, and it is the wrong choice with one source and a working report.
Source audit
We list every source, who holds the login, how much history it keeps and how it treats refunds and time zones.
Signed metric sheet
Each metric in writing: formula, source of record and owner. Nothing is built on an unsigned metric.
Data engineering services
Pipelines load your tools into a warehouse you own, with tests for duplicates, missing keys and late data.
Dashboards and reporting
Business intelligence consulting on a clean data model, with each number linked to its definition.
Predictive analytics
Machine learning development for demand, churn or lifetime value, where a decision repeats and enough events exist.
How far up the ladder do you need?
Each step is right until its trigger arrives, and many businesses can stop before the top. When the question is about what already happened, plain SQL and a dashboard beat a model.
One or two sources, and the question stays inside Shopify, Stripe or GA4.
The same export gets rebuilt by hand, or two people get two answers.
A few agreed metrics in a tool such as Data Studio (formerly Looker Studio), Power BI or Metabase.
A question joins several tools, or you need history a platform will not keep.
Several sources must join on one customer or order, in a warehouse such as BigQuery or Snowflake.
The numbers are trusted, and a decision keeps asking what happens next.
You meet the data thresholds of a tool such as Klaviyo or GA4, and its output fits the decision.
It does not cover the decision, or you need your own margin or stock data in it.
A decision repeats, enough events exist and the model beats a simple baseline on past data.
It stops beating the baseline. Then it comes off.
A sample metric sheet
This sample was written for this page and is not from a client. Every number on a dashboard has a line like this, signed before anything is built on it.
Talk it through on a call
We look at what you have now and what you want it to do.
You hear how we would approach it and who would work on it.
You get a fixed quote in writing after the call, with nothing to sign until you are ready.
How the data work runs
Founder and CEO Krishnkant Patidar leads the project, and AI and ML developer Deepak Patidar brings ten years in this field. Earlier client work stays under NDA, so this page shows the method. Reviews run in writing, and call times show in your own time zone.
Name the decisions
We start with the decisions the numbers must serve, such as what to reorder and where to spend. Everything else waits.
Audit the sources
We compare totals across your tools over one period, and note how each treats refunds, time zones and currency.
Agree the metric sheet
Each metric is written down and signed. It stays the reference for every later argument.
Build pipelines and dashboards
Raw data loads into the warehouse, tests run on every load, and totals tie back to Shopify or Stripe.
Set a baseline and backtest
A simple rule sets the bar. The model trains on earlier data and is tested on the period after it, including peak trading.
Run it beside your process
The model scores in the background while people work as they do now. Alerts go to a named owner.
Rules for any model we ship
A model keeps its place only while it beats something simple.
A baseline comes first
Before any model, we write down what a simple method scores: the same period last year for demand, or "no order in 90 days" for churn.
Tested in date order
The model never sees the future while training. If the baseline wins the backtest, we stop and say so.
No accuracy promise
No honest figure exists before a backtest on your data, and we guarantee none after it. Results are reported as error against the baseline.
A rule for switching off
A model that stops beating the baseline comes off. The retrain and switch-off rules are written down before launch.
Personal data stays out
Reporting tables leave out names and emails where a metric does not need them, and hashed data is still treated as personal.
What you keep
What your tools stop keeping
Read on 7 November 2026. These limits can change.
GA4
Event-level detail can be kept for two or fourteen months, and that setting applies to explorations and funnels. Standard aggregated reports are not affected.
Meta Ads
Since 12 January 2026 the Ads Insights API no longer returns 7-day and 28-day view windows. Unique-count fields go back thirteen months.
Shopify
The Admin API limits orders to the last 60 days unless an app is granted wider access. That is a limit on the API, and your store keeps the orders.
Built-in predictions
Klaviyo needs at least 500 customers with an order and 180 days of history. GA4 needs 1,000 returning users who met a condition and 1,000 who did not.
Where your data sits
A BigQuery dataset cannot be moved after creation, so the region is chosen with you on day one. Data handling is agreed in writing before any real data is used.
Other pages for this work
Questions before you hire a data team
Ownership, mismatched numbers, warehouses, data size and accuracy.
Book a 30-minute call
After a data analytics project with Krikir, the warehouse, pipelines, code, metric sheet, dashboards and models sit in your accounts. They are transferred to you on final payment, under the contract.
Shopify, GA4 and Meta show different revenue because each counts differently: refunds, browser tracking, attribution windows and time zones all vary. They will not match exactly. We choose one source of record for each metric, usually Shopify or Stripe for money, write it down and watch the gap.
Often a data warehouse is not needed at first. A warehouse earns its place when questions join several tools, or when a platform will not keep the history you need.
Whether you have enough data for machine learning depends on the number of events, such as churned customers, more than on rows. The source audit checks this first. No honest fixed minimum exists, though Klaviyo and GA4 publish thresholds for their own predictions.
A forecast's accuracy cannot be stated honestly before a backtest on your own data, and we guarantee no figure after it. We report error against a simple baseline, and if the baseline wins we say so. Accuracy falls as conditions change, so it is measured after launch too.
Your customer data sits in your own accounts, in a region chosen with you at the start. Personal fields stay out of reporting tables where a metric does not need them, and data handling is agreed in writing before any real data is used.
Let's talk about your next project
Book a 30-minute call or send a brief. A founder replies within one working day.