There's a dirty secret in public procurement that nobody talks about openly: most bid evaluations are a rush job.
Not because the people doing them are lazy. The opposite, actually. Procurement specialists are some of the most meticulous professionals you'll meet. The problem is math. When you have 6 proposals for a complex tender, each running 80-150 pages, and you need to evaluate them against 30+ technical criteria within a two-week deadline — something has to give.
Usually what gives is depth. Page 1 gets careful attention. Page 80 gets a skim. The technical annex gets opened, scrolled through, and closed. We've spoken with procurement teams across Latvia who openly admit this. Not with pride, but with the resigned honesty of people who know the system demands more than it allows.
What the numbers look like
Latvia processed 11,421 public procurement procedures in 2024, awarding 21,558 contracts worth EUR 5.45 billion. That's about 13% of GDP flowing through a process where evaluators are constantly fighting the clock.
And 73% of those contracts were awarded on lowest price alone, according to 2021-2023 data. Not because procurement teams don't want to evaluate quality. Because proper quality evaluation takes time they don't have.
The State Audit Office (Valsts Kontrole) published an audit in December 2024 that didn't mince words: Latvia's procurement regulation is "complicated and inflexible." Only 16.5% of IUB's own users find the regulations understandable. The Cabinet of Ministers approved structural reforms in August 2025 — reducing exclusion criteria from 12 to 2, eliminating national-level thresholds to keep only EU ones.
But streamlining procedures only helps so much. The core bottleneck remains: someone has to actually read those documents.
Where AI fits (and where it doesn't)
Here's where we need to be honest. AI is not going to replace procurement specialists. Full stop. The judgment calls — is this supplier credible? Does this pricing make sense in context? Should we weight innovation over cost in this particular case? — those remain human decisions.
What AI can do is the part that procurement teams wish they had more time for but never do: reading everything, thoroughly, every time.
We built our AI agent to work the way a perfect junior analyst would. It reads the full RFP. It reads every proposal, cover to cover. It compares each requirement against what the bidder actually wrote. When it finds a gap or a weak point, it pulls the exact quote so you can verify in seconds.
The result isn't faster procurement. Sometimes the analysis takes hours on a complex multi-lot tender. The result is that nothing gets missed. Page 120 gets the same scrutiny as page 3.
What we've seen in practice
One thing that surprised us early on: the AI doesn't just find things evaluators miss. It finds things evaluators couldn't reasonably be expected to catch.
Cross-references between technical specifications and financial proposals that don't quite add up. Promises in the executive summary that are quietly contradicted by caveats buried in an annex. Compliance claims that technically answer the question but don't actually address the requirement.
These aren't mistakes by evaluators. These are things that would require reading the same 400-page proposal three times with perfect recall. Nobody does that. Nobody can.
The uncomfortable question
Every procurement team we've talked to eventually arrives at the same uncomfortable realization: if AI can catch things that humans structurally cannot, then going back to purely manual evaluation feels like choosing to be less thorough.
That's not a comfortable thought. It shouldn't be. But it's honest.
The teams that are getting the most value from AI evaluation aren't the ones trying to "automate procurement." They're the ones treating AI as a tool that lets their experts be better at the parts that matter — the judgment, the strategy, the decisions that ultimately protect public interest and public money.
Because that's what this is about, at the end of the day. Public money, spent well. And the people responsible for that deserve better tools than a PDF viewer and a tight deadline.