# TRINTA vs building tender monitoring in-house

> Scraping one procurement portal is a weekend. Scraping a region is a product with an on-call rota. An honest look at the build-versus-buy decision, including when building really is right.

Source: https://trinta.ai/compare/in-house · Format: Markdown mirror of that page · Language: en

## Verdict

Building tender monitoring in-house is realistic for one or two portals and unrealistic across a region, and the reason is maintenance rather than difficulty. The first scraper is a weekend; the problem is the next several hundred, each of which breaks when a government portal changes its markup, adds a session token or moves behind a login, usually without notice. TRINTA runs 860+ connectors as its product, plus semantic scoring, document retrieval and multilingual matching on top. If your target market is one national portal with a clean open API, building it is the honest answer.

## What building it in-house gets you

- Total control over the data model, the storage and the integration into your own systems.
- No recurring licence fee, which matters if the build is genuinely small and stays small.
- Feasible and even easy where the target market publishes open data with an API, as Brazil's PNCP and Ukraine's ProZorro do.
- The output stays yours, with no vendor to depend on for continuity.

## Where they run out

- Portals break scrapers constantly, and each break is silent: the feed does not error, it just goes quiet, and nobody notices until a deadline is missed.
- Most procurement portals have no API. Many require sessions, some require registration, and a few are effectively PDF archives behind navigation.
- Scraping is the small half. Semantic scoring against a product catalogue, multilingual matching and annex extraction are each their own project.
- The work never finishes, so it becomes a permanent claim on engineering time that was hired to build your actual product.
- Coverage tends to freeze at whatever the original developer built, which means expansion into a new market restarts the project.

## Side by side

| Dimension | Building tender monitoring in-house | TRINTA |
| --- | --- | --- |
| Time to first useful output | Weeks for one portal, months for several | One demo call, then the profile is built for you |
| Breakage | Yours to notice and yours to fix, silently and forever | Maintained as the product, across the whole fleet |
| Scale of the fleet | What you have time to write and keep alive | 860+ connectors, 1,539 platforms mapped |
| Scoring | A separate project, usually never started | Semantic fit scoring 0 to 100 against your catalogue |
| Documents | A second scraping problem, harder than the first | Annexes fetched and read into a case file |
| Ongoing cost | Engineering time, permanently, and it grows with coverage | From €2,000 per month, fixed |

## Build it in-house instead when

- Your market is one or two portals that publish clean open data through an API.
- You have specific internal systems the data must land inside, and integration matters more than coverage.
- You have engineering capacity that is genuinely spare, and an owner who will still be there in two years.
- The monitoring is incidental to a larger data product you are building anyway.

## Questions

**How hard is it to scrape government procurement portals?**

The first one is easy and the fleet is the problem. Government portals change markup without notice, add session requirements, sit behind registration walls and sometimes publish tenders only as PDFs inside navigation. Each break is silent, so a stalled feed looks identical to a quiet week until a deadline is missed. Maintenance, not construction, is the real cost.

**Could we build the scraping and buy the scoring?**

In principle, but the two are harder to separate than they look, because scoring quality depends on how completely and how consistently each notice was extracted in the first place. TRINTA is not sold as a scoring API over someone else's feed. If you have a specific integration need, it is a reasonable thing to raise on the demo call rather than a settled no.

**What does TRINTA maintain that we would have to?**

A fleet of 860+ connectors against a mapped registry of 1,539 platforms, the multilingual matching that lets a French notice score against an English catalogue, document retrieval from behind portal navigation, and the scoring model itself. Each is a standing commitment rather than a delivered project.

---

Full site index for agents: https://trinta.ai/llms.txt · Whole site in one document: https://trinta.ai/llms-full.txt