Bidding on Chile public procurement with AI: 2026 guide for state suppliers

Bidding on Chile’s Mercado Público (Chilecompra) consumes time from small teams: read the terms, decide whether to bid, write, calculate the price. AI can accelerate much of it without replacing human judgment. This guide covers what to automate well, what not to automate, and what tools to use in 2026.

The concrete problem

A medium-sized Chilean state supplier faces, every week:

Volume: hundreds of tenders published, most irrelevant.
Long terms: 40-100 page PDFs with technical and administrative requirements.
Short deadlines: typically 7-15 calendar days.
Repetition: if they bid on something similar last year, much can be reused.
Informed decision: is it worth bidding? how competitive is the proposal?

A spreadsheet plus manual search on mercadopublico.cl solves this at the cost of many hours. Well-designed AI reduces it drastically.

What to automate well

1. Discovery (filtering)

AI discovers new tenders daily, filters by UNSPSC code of your industry, minimum amount, region and deadline. It hands you a short list of 5-15 per week instead of 200.

2. Technical analysis

For the ones that pass the filter, AI extracts: critical requirements, evaluation criteria, delivery deadline, required guarantees, minimum experience. Instead of reading 60 pages, you read a structured sheet.

3. Scoring

Each tender gets a deterministic score (not a black box): calculated with clear rules about your fit. If you fail a hard requirement (e.g. minimum 3 years experience, you have 1), the system says it before you waste time.

4. RAG over your previous bids

For the ones you decide to bid on, the AI searches your corpus of previous bids (the ones you won) and proposes reusable sections: experience, team, methodology. You edit and adjust to the case.

5. UNSPSC analytics

Dashboard of which categories the State buys most, where it competes, where you stand. Useful to decide whether to get certified in a new UNSPSC code.

What NOT to automate (yet)

Final decision to bid: human judgment, considers client relationship, real operational capacity.
Pricing: depends on internal costs the AI does not know well.
Cover letter / personal stamp: the human touch makes the difference in qualitative evaluation.
Legal guarantees and bonds: manual paperwork with your bank.

Common mistakes AI can make worse if not designed well

1. Bidding on everything the filter shows

If your AI only filters by keywords, it will flood you with irrelevant opportunities and burn the team. The AI must rank by real probability of award.

2. Invented citations in the bid

If the AI writes sections without RAG over your corpus, it invents background. Winning with false info means disqualification.

3. Black box

If the AI says “this tender has score 87” without explaining why, you cannot appeal to judgment. You need deterministic scoring with a breakdown.

4. Sensitive data sent to OpenAI

Your previous bids are IP. If the tool sends everything to a foreign provider’s public API, consider confidentiality implications and clauses with your own clients.

How quelicitar.cl does it

quelicitar is aGo’s B2B product for Chilean state suppliers. In pre-launch. It covers:

Daily discovery with filters by UNSPSC, region and amount.
Deterministic explainable scoring (not a black box, each score component is visible).
RAG over your corpus of previously awarded bids.
UNSPSC market analytics.
Add-on marketplace (factoring for awarded bids, insurance).
Stack: Django 5 + Flow + OpenFactura. Data on controlled infrastructure.

If your SMB bids recurrently on Mercado Público, request early access.

Frequently asked questions

Can the AI write the full bid for me?

It is not advisable. Writing with AI without RAG or review invents background. RAG over your previous bids plus human editing is the correct flow.

How much time can I save?

It depends on volume, but filtering + technical analysis + scoring can save 50% to 70% of the time before the decision to bid.

Does it comply with data regulations?

quelicitar processes personal data under Chile’s Law 21.719 (see our Law 21.719 guide). Data lives on controlled infrastructure in Chile.

Is it for large or small companies?

It works from companies that bid twice a month. For very low volumes it may not be worth it.

Conclusion

AI in public procurement does not replace the commercial team: it frees them from repetitive tasks so they can focus on higher-value decisions. The key is to choose tools with explainable scoring and RAG over your own corpus, not generalist agents that hallucinate.

Request early access to quelicitar.cl.

Sources