← Kõik uurimused

4. november 2026 · 5 min lugemist · Sander Gansen

Viešieji Pirkimai: tender discovery is not tender authority

The Lithuanian Viešieji Pirkimai AI/TenderGlass case shows how document search, alerts and historical procurement data can help suppliers prepare a bid without proving autonomous bidding, verified accuracy or customer ROI.

Viešieji Pirkimai AI tender-discovery service — organisation mark from viesiejipirkimai.ai
Juhtumi visuaalne tähistus. Pilt või märk pärineb allikast viesiejipirkimai.ai. See ei tähista kliendisuhet ega heakskiitu.

Allikakirje: C15

Artikli väljaanne valmis: 4. november 2026. See ei ole kõigi viidatud allikate avaldamiskuupäev.

Uurimuse ajahorisont: 23 and 27 October 2026; source-specific dates retained in article

Algne artikkel on inglise keeles. Lehe navigeerimine on eestikeelne.

External-source research synthesis, not an OfferSignal client result. Based on the retained October 2026 research and source-specific dates.

A supplier can miss a relevant public tender because the opportunity is buried in long documents, classified under an imperfect code or scattered across historic procurement systems. Finding an opportunity, understanding its requirements and checking comparable awards are useful jobs for software. They are not the same as signing and submitting a bid.

A research layer for Lithuanian tenders

Viešieji Pirkimai AI began as a Lithuanian supplier-side service associated with Polipo, MB. In an April 2025 interview, founders Giovanni Proietta and Massimo Cairo described a system that scanned published tender documents and used a public large language model within the service to answer questions about contacts, technical specifications and participation information. The original product let users select CPV codes or subcategories and keywords, save filters and receive daily email alerts. The founder said alerts still needed improvement and that the service could not yet track an individual tender’s procedure. 1

By January 2026, the provider described the product as TenderGlass, with full-text search using keywords, synonyms and contextual phrases, theme-based discovery, competitor and winner-price analysis, and repeat-procurement forecasting. It says the AI does not decide for the user, assess legal requirements unless specifically programmed or guarantee a win. These are provider-described capabilities, not a published evaluation: no test corpus, precision or recall, answer-accuracy measure, forecast back-test or customer outcome is disclosed. 3

Historical data needs a visible chain of custody

A January 2026 announcement says TenderGlass was integrated with eViesiejiPirkimai.lt and could search concluded procurements by supplier, contracting authority or CPV, with winner, price, contract and scope views and filters for year, procedure and value. The post claims Lithuanian national and above-EU-threshold history from 2016. 2

That history is not a single unchanged database. Lithuania’s Public Procurement Office says that since 1 December 2024 public procurements have been carried out in the new Central Public Procurement Information System, while the old system remains relevant for historical searches. 4 The current portal provides public search and separate supplier registration and login routes for participation. 5 The legacy site describes the split between procurements from 1 December 2024, notices from 1 July 2017 to 30 November 2024 and earlier records, with searchable fields including buyer, participant or winner and CPV. 6

A stitched 2016-to-present view may be useful, but the vendor has not published its ingestion method, source-record links, refresh cadence, completeness check, entity matching, deduplication or treatment of changed classifications, cancellations and amendments. “Integrated” establishes a product claim, not an independently verified official integration or reliable historical benchmark.

The only disclosed outcome is early adoption

The only public adoption figure is founder-reported: almost 20 paying customers after roughly 1.5 months, at a listed price of €9.99 per month, reported in April 2025. No customer names, conversion rate, retention, active use, submitted bids, wins, saved time, accuracy or revenue are provided. 1

The same interview says an internal customer survey suggested that a supplier without a team could spend up to five hours per week finding relevant opportunities. The sample, question wording, period and actual before-and-after saving are not disclosed. This is a workload estimate, not a measured productivity result. No located source evaluates retrieval coverage, document-answer correctness, historic-price accuracy, legal-compliance effect or supplier return on investment.

Research authority is not submission authority

The official procurement system remains the place where an authorised supplier registers, uses electronic signatures and submits a bid. Nothing in the retained Viešieji Pirkimai AI or TenderGlass evidence shows that the service can log in to CVP IS, sign for an organisation, communicate bindingly with a buyer, negotiate, award a tender or make a tender payment. 4 5

Polipo’s terms describe subscription billing through Stripe, including possible auto-renewal, and cap liability at the lesser of the amount paid and €100. That is SaaS payment, not escrow, invoice settlement or procurement authority. 7 The TenderGlass landing advertises Estonia, Lithuania, Latvia and Poland, but those country links are a coverage claim rather than independently verified operating performance. 8

An illustrative supplier test

For an illustrative equipment manufacturer, use the service to create a sourced bid-or-no-bid brief. Preserve the official notice URL or identifier, source system, extraction date, procedure state, document version, matched CPV or keyword and the exact passage supporting each answer. A commercial or technical owner should verify scope, qualification criteria, deadline, delivery assumptions and comparable prices in the official record.

Measure missed relevant tenders, irrelevant alerts, source delay, answer corrections and review time against a defined set of official searches. For historic prices, retain lot, quantity, currency and tax basis where available, contract term, amendments, procedure and matched supplier identity. The authorised representative—not the research assistant—owns eligibility, pricing, signature and submission.

The durable lesson is a constrained procurement-research assistant: better discovery and evidence-backed preparation before a controlled hand-off to the official system. This case supports that front end. It does not support autonomous bidding, negotiation, payment or claims of customer ROI.

Sources and next reading

Explore OfferSignal · Focused Business & AI Workshop · All research

References

  1. [1] LRT interview on the Lithuanian public procurement AI platform
  2. [2] Tenderglass integration with the Lithuanian procurement archive
  3. [3] AI in public procurement practical use in 2026
  4. [4] Public Procurement Office e-public procurement guidance
  5. [5] Central Public Procurement Information System home
  6. [6] Legacy Central Public Procurement Information System
  7. [7] Viešieji Pirkimai AI terms and conditions
  8. [8] TenderGlass public product landing

Valikulised sätted on alguses välja lülitatud. Keeldumine ei takista veebilehe, kirjutiste ega broneerimislinkide kasutamist.