Build vs buy: scraping court portals vs a court data API
Every team that needs Indian court data starts with the same thought: the portals are public, so how hard can it be to scrape them? The first prototype usually works in a week. The trouble starts when it has to cover every court, survive portal changes, read scanned orders, match names to real people and stay current after every hearing.
Build if you need a few courts, a narrow question and full control. Buy if you need national coverage, order text, freshness and identity matching without a team to maintain them.
Prices and coverage read from ecourtsindia.com on 15 November 2026.
Last updated: 15 November 2026
Key takeaways
- Collection is the easy part: cleaning, matching and keeping data current is where in-house projects stall.
- Coverage: India has district and taluka courts in every state, 25 High Courts and many tribunals, each with its own portal behaviour.
- Freshness: a record goes stale after every hearing. Someone has to re-read it.
- Identity: turning a party name into a person or company is the hardest problem, and it is not solved by scraping.
- The buy option: 24 crore+ records indexed, refreshes at ₹0.05, orders as text and identity-matched checks, billed per successful call.
What does an in-house pipeline have to solve?

Collect
Court records sit on several public systems: the national eCourts services for district courts, separate High Court portals, the Supreme Court’s site and each tribunal’s own system. They use CAPTCHAs, session limits and formats that differ court to court, and government servers are often slow or down. Our post on India’s court data stack maps the sources.
Clean
Orders are PDFs, many of them scanned, so you need OCR that copes with Indian legal formatting, signature stamps and mixed languages. Case types and statuses are written differently in different courts and need normalising before you can count anything. And party names are free text, so “HDFC Bank Ltd”, “Hdfc Bank LTD” and “H.D.F.C. Bank” are the same party to a reader but not to a database.
Keep current
Every hearing changes a record. A pipeline has to know which cases to re-read, when, and how to detect new orders, without re-reading millions of cases every day. Then it has to keep working when a portal changes its layout. Our from scraping to production post describes what that took us.
How do the costs compare?
| Cost line | Build | Buy |
|---|---|---|
| Engineering | Initial build plus ongoing maintenance as portals change | Integration only |
| Infrastructure | Crawlers, proxies, OCR, storage, search index | None |
| Coverage gaps | Courts you have not built for yet | The provider’s coverage |
| Freshness | Your re-read schedule and its compute | ₹0.05 per refresh at the Enterprise base rate |
| Order text | Your OCR quality | ₹1.75 per order as text |
| Identity matching | Your own matching logic and its errors | ₹33 per identity-matched check |
The honest comparison is engineering time. If one engineer spends a quarter of their time keeping scrapers alive, that cost alone usually exceeds a mid-sized API bill. Our court data cost model turns per-call prices into monthly figures for common workflows.
How big is the freshness problem?
Bigger than most teams expect. In our index, 62,77,157 cases have a next hearing date in the single week of 16 to 22 November 2026. Each of those records can change that week: a new date, a new order, a change of stage or a disposal. Uttar Pradesh alone accounts for 15,87,968 of them, Maharashtra for 8,89,256 and Bihar for 4,56,607.
| What has to stay current | Scale in our index | Why it matters for a pipeline |
|---|---|---|
| Cases with a hearing in one week (16 to 22 Nov 2026) | 62,77,157 | Every listed case may change after its hearing |
| Records marked pending | 8,37,52,146 | Live cases that will keep changing until they are decided |
| Records with at least one order attached | 16,58,39,092 (47%) | Orders arrive as PDFs and need OCR before they can be searched |
An in-house pipeline has to decide which of these to re-read and when, and it has to do so against portals that limit sessions and sometimes go down. On the buy side you only pay to refresh the cases you actually track, at ₹0.05 per refresh at the Enterprise base rate. Tracking 5,000 cases with a refresh after every hearing, at say four hearings a month each, is 20,000 refreshes, or ₹1,000 a month.
What does name matching look like in practice?
Take a bank that appears in a great many cases. Searching party names in our index for the exact phrase “State Bank of India” returns more than a lakh records. Searching “SBI” returns about 1.1 lakh, and “S.B.I.” 22,533. Many of these are the same bank written three ways, and that is before branch names, “the Branch Manager” and spelling errors.
| How the name is written | Records in our index (exact phrase) |
|---|---|
| State Bank of India | More than 1,00,000 |
| SBI | About 1,10,000 |
| S.B.I. | 22,533 |
For a company, normalising these variants is tedious but possible. For people it is much harder, because a name is shared by thousands of strangers. That is why identity-matched checks exist as a separate product, and why our guide to verified-profile badges spends so long on why a name search cannot verify a person.
A worked example: a legal-tech startup choosing in month one
Suppose a startup is building a litigation-tracking product for mid-sized law firms. In its first year it expects to track about 5,000 cases, add 500 new cases a month by search, and show the latest order on each case.
| Need | Buy: calls and monthly cost at the base rate | Build: what the team has to own |
|---|---|---|
| Find and add 500 new cases | 500 searches and 500 case reads: ₹350 | Search across many portals, CAPTCHAs, de-duplication |
| Keep 5,000 cases current | About 20,000 refreshes: ₹1,000 | Re-read scheduler, failure handling, portal changes |
| Daily listing alerts | 1,10,000 CNR listing checks: ₹11,000 | Daily cause-list collection for every court used |
| Show the latest order as text | About 2,000 orders as Markdown: ₹3,500 | PDF download, OCR, layout clean-up |
| Total | About ₹15,850 a month | At least one engineer’s time, plus infrastructure |
At this scale the monthly API bill is roughly the cost of a few days of one engineer’s time. The calculation changes if the product needs only one tribunal, or if the team already runs crawlers for other public data. That is the case for building, and it is a real one.
Questions to ask before you decide
- How many courts do we need in year one, and in year three? A few courts favour building; a national footprint favours buying.
- How fresh does the data need to be? A weekly research snapshot is cheap to build. A same-day hearing alert is not.
- Do we need order text? If yes, budget for OCR quality, not just collection.
- Will we make decisions about people? If yes, identity matching and consent are the hard parts, and they are not solved by scraping.
- Who will own the pipeline when portals change? If the honest answer is “whoever is free”, buy.
If you do decide to build, start with the narrowest slice that answers a real customer question, measure how often it breaks in the first three months, and price that maintenance honestly before you widen it. If you decide to buy, test the provider on the courts and case types your customers care about most, not on a demo case, and check how failed or empty calls are billed. Either way, keep the CNR as your primary key: it is the identifier the eCourts system gives every case, and it makes switching or combining sources far easier later.
When does building make sense?
- You need a handful of courts, not the country.
- Your question is narrow and stable, such as one tribunal’s new filings each week.
- You have engineers who will own the pipeline for years, not months.
- You are doing research where a one-time dataset is enough. Open-source scrapers and public judgment datasets can be a good start.
When does buying make sense?
- You need national coverage across district courts, High Courts and tribunals.
- Your product depends on current status, not a snapshot.
- You need order text for search or AI. See orders as Markdown for RAG.
- You make decisions about people, where identity matching matters more than raw records.
Many teams do both: buy the national layer and build thin, specialised collectors for the few sources their product depends on most.
Sources and method
- Prices from ecourtsindia.com/api/pricing and record count from /api/docs, read 15 November 2026.
- Build-side costs depend on your team and scope; the table lists cost lines, not estimates.
Frequently asked questions
Can I scrape the eCourts portal myself?
Portals are public but use CAPTCHAs, session limits and court-specific formats, and servers are often slow. A prototype is easy; national coverage, order OCR and keeping records current are the hard parts.
What is hardest about building court data in-house?
Keeping records current after every hearing and matching party names to real people or companies. Collection is the easier part.
When should a company build its own court data pipeline?
When it needs a few courts, a narrow stable question and has engineers to own the pipeline long term.
What does buying court data cost?
Per successful call: ₹0.20 per search, ₹0.50 per case, ₹0.05 per refresh and ₹1.75 per order as text at the Enterprise base rate, before GST.
Can I combine both approaches?
Yes. Many teams buy national coverage and build small collectors for the few sources their product depends on most.
How many court cases are listed for hearing in a week?
Our index shows 62,77,157 cases with a next hearing date between 16 and 22 November 2026, each a record that may change after the hearing.
What does it cost to keep 5,000 tracked cases current?
About ₹1,000 a month at the Enterprise base rate, assuming around four refreshes per case a month at ₹0.05 each, before GST.
Why is matching party names so hard?
The same party is written many ways. In our index, State Bank of India, SBI and S.B.I. each return thousands of records, and people’s names are shared by thousands of strangers.
Important: This guide is published by Billion Insights Ventures Pvt. Ltd., the company behind eCourtsIndia. It is not legal advice. Court data shown here is a copy of public court records as read on the date stated at the top of this page and is not a certified copy. A name match is not identity, a pending case is not guilt, and a disposed case is a status, not an outcome. Prices, limits, counts and examples were checked on that date and can change. eCourtsIndia is a private legal-technology platform and is not affiliated with any court, the e-Committee of the Supreme Court or the Government of India. Found an error? Write to support@ecourtsindia.com.



