Data Engineering & Analytics
What a data platform actually costs
Data work is quoted on the state of your source systems, not the number of dashboards at the end, and almost everything that moves the figure is checkable before anyone starts.
Get a scoped estimateWhy quotes for the same project vary so wildly
Nobody is really quoting the dashboards. They are quoting the condition of the data underneath: how many systems it lives in, whether the customer ID means the same thing in each of them, and how much has to be repaired before a chart can be trusted. A supplier who has not opened your data is guessing at that.
The question that moves the answer most is which decision changes when the number arrives. A team that can name one, such as a reorder point or a churn intervention, needs a narrow pipeline and can have it in weeks. A team that wants general visibility is asking for every table, which is a platform.
Cost drivers
What actually moves the number
Every one of these can be checked before a quote is written. A proposal that mentions none of them is pricing a guess and will be revised later, usually upwards.
How many sources, and how awkward
A documented, paginated API loads in days. A supplier who emails a spreadsheet with merged cells every Tuesday is a week of engineering on its own.
Identity across systems
The same customer exists as a Stripe ID, a HubSpot record and a support email. Deciding which one wins is design work, not a settings screen.
Historical backfill
Pulling years of history through a rate-limited API frequently takes longer than building the pipeline that keeps it current from now on.
Refresh frequency
Overnight batch is cheap and boring. Anything close to live means streaming, replay handling and someone on call, and very few decisions are made that fast.
Metric definitions
Finance counts revenue on invoice date, sales on close date. Someone has to arbitrate and write it down, and those meetings are part of the build.
Governance and access
Row-level rules so a regional manager sees only their region, plus masking on anything that would surface in a GDPR subject access request.
Warehouse and query design
Managed warehouses bill for badly written queries. Partitioning and incremental models are the difference between a dull monthly bill and an alarming one.
Where the answer is delivered
A dashboard is the cheap end. Per-tenant analytics embedded inside your own product is product engineering that happens to read from a warehouse.
Activation back into tools
Pushing a computed segment into HubSpot or your ad platform so it changes behaviour, rather than sitting in a report nobody opens on Monday.
Scope tiers
Four honest shapes for a first engagement
Most companies need the second tier. A fair number find that out only after paying for the third, having built tables for questions nobody ever asked.
- TIER 1 — REPORTING STARTER
- Warehouse and core dashboards. Ships in 3–5 weeks.
- ₹1.5L – ₹2.5L$1.7k – $2.8k
- Managed ingestion from a few well-documented sources, a warehouse, and dashboards answering questions you name up front. Suits a team still rebuilding numbers by hand each week. Does not include modelling, quality testing, historical repair or access controls.
- TIER 2 — MODELLED WAREHOUSE
- Agreed metrics and tested pipelines. Ships in 8–12 weeks.
- ₹6L – ₹9L$6.8k – $10k
- Modelled tables, metric definitions written down and signed off, incremental loads, and automated tests on the pipelines that matter. Suits a company where two teams quote different revenue. Does not include streaming, embedded analytics or regulated-data controls.
- TIER 3 — PRODUCTION PLATFORM
- Orchestration, alerting and activation. Ships in 4–6 months.
- ₹18L – ₹28L$20k – $32k
- Orchestrated pipelines with alerting and lineage, reverse ETL back into your CRM and ad platforms, and the DevOps and Cloud Infrastructure work to run it properly. Suits operations at scale. Does not include customer-facing analytics or formal compliance evidence.
- TIER 4 — GOVERNED OR EMBEDDED
- Regulated or customer-facing analytics. Ships in 6–12 months.
- From ₹40LFrom $45k
- Row-level security, PII masking, audit trails, per-tenant analytics inside your own product, and the evidence a regulated review will ask to see. Suits fintech, health and multi-tenant SaaS. Does not include your legal counsel or certification audit fees.
Indicative ranges for scoping conversations, not quotes. They reflect what work of this shape has cost us to deliver — your figure comes out of a paid discovery, against a written scope, and is fixed before any code is written. Gas, licences, cloud and third-party audit fees sit outside these numbers.
How we work
From a vague question to a number you can defend
Timelines below are typical for a first engagement. The survey is deliberately early and deliberately cheap, because that is where most of the bad news lives.
Question audit
1 weekWe list the decisions this data is meant to change, then cut the list to the few someone will actually act on this quarter. Everything else waits.
Source survey
1–2 weeksWe pull real extracts from every system and test whether the join keys hold. Scope shrinks here often, and occasionally we tell you a spreadsheet is still enough.
Model and build
4–10 weeksIngestion, modelled tables, tests, and the dashboard or activation surface. Metric definitions get agreed in writing before any figure is shown to a board.
Run and hand over
OngoingMonitoring, schema-change fixes and query tuning once real usage shows where the bill goes. You keep the code, the models and the runbook either way.
Is this worth budgeting for now?
Worth reading before you take a number upstairs. A data project started too early gets blamed on the data team.
Worth budgeting for now if…
- You can name a decision currently made on somebody's instinct
- Numbers are rebuilt by hand in spreadsheets every single week
- Two teams present different revenue figures in the same meeting
- Your product needs to show customers their own usage data
Wait, if…
- Everything lives in one system that already reports well enough
- Nobody has agreed what a customer or an order is
- The source systems are being replaced within the next few months
- You want visibility, but no decision would change because of it
Frequently Asked
Questions
Common questions about scoping a data platform, phasing it, and what it takes to run afterwards.
Usually the source systems, not the reporting. A documented API loads in days; a legacy ERP with no read replica, or a supplier who sends a spreadsheet by email, is real engineering before anything reaches a chart. The second driver is identity — when the same customer exists under different IDs in billing, CRM and support, someone has to decide which record wins and then defend that decision to finance. The third is refresh speed. Overnight is straightforward. Anything close to live means streaming, replay handling and someone carrying a pager.
Streaming first. Almost every decision a business makes daily survives an overnight refresh, and near-real-time roughly doubles the operational surface you have to run. Cut historical backfill next — start with a recent window and load the rest later, once you know which tables anyone opens. Cut the source list down to the systems that answer your named questions, and be ruthless about dashboards. A small set people actually use beats a library that quietly rots and then gets distrusted wholesale.
Four things, reliably. Data quality work: the deduplication, the missing country codes, the returns never linked back to their orders. Metric definitions, which in practice is a series of meetings where finance and sales agree what revenue means, and that time is not free. Access control, particularly row-level rules so a regional manager sees only their region. And the warehouse bill itself — driven far more by how queries are written than by how much data you happen to store.
Three buckets, none of them fixed. Warehouse compute, which scales with how often pipelines run and how carefully they were written; an incremental model can cost a small fraction of a full refresh of the same table. Tool licences, if you use managed ingestion or pay per BI seat. And maintenance, because source APIs change without warning and an unwatched pipeline fails silently for weeks. Budget for a named owner, whether that is a retainer with us or someone internal.
Phase it, in nearly every case. The first phase should answer one question end to end, because that is what reveals how bad the data really is, and the survey almost always changes the design. Teams who commission a full platform up front tend to build tables for questions nobody asks, then discover the join key they assumed was stable is not. Ship one useful report, watch whether anyone acts on it, then extend. If nobody acts, you have saved the rest.
Have a project in mind?
Fixed price after a paid discovery — no hourly billing. A real engineer reads every enquiry, and we reply within 24 hours.