How the score works
The percentage on a tender is our estimate of the chance that ITTECH wins it if we bid — built from our own bidding record and the portal's history of the category, every factor shown with the numbers it came from. It is a starting point for a decision, not the decision.
1. It starts from our base rate
Of every tender we have bid on that reached a decision, we won a share: that is the base rate, and the number every score starts from. Right now that is 20 won of 56 decided (36%), from 60 bids in total.
2. Five factors move it up or down
Each factor compares a slice of our record with the base rate. Where we have won more than our average in that slice, the score goes up; where less, down.
| Factor | What it asks | Where the evidence comes from |
|---|---|---|
| Our line of business | Does this read like the things we have actually bid on? | The tender's title compared with our 60 past bids (text similarity), and the buyer's own CPV classification — a code we bid in repeatedly settles it on its own. |
| Expected competition | How crowded will the field be? | How many bidders this category usually draws across the whole portal (48,352 tenders), and how often we win in fields of that size. |
| Deal size | Is this the size of deal we win? | Our win rate in the value band this tender falls in: under ₾10k, ₾10k–50k, ₾50k–200k, over ₾200k. |
| Category fit | Do we play in this exact category? | Our wins and losses under the same CPV code. |
| This buyer | Have we won with this buyer before — and does it have a favourite supplier? | Our record with the buyer, and whether one supplier keeps winning this category from it (the incumbent). |
3. Thin evidence is not allowed to shout
Two bids in a category cannot mean as much as twenty-nine. Every slice's win rate is pulled toward the base rate in proportion to how little stands behind it:
So a 2-bid category barely moves the number, while a 29-bid one mostly gets its way. No single factor may swing the odds by more than a fixed amount (a clamp of 1.6 in log-odds, roughly ×5 either way), and the scale is capped at 3–92%: nothing is ever certain, in either direction.
4. The factors combine as odds, not as points
Each factor is converted to log-odds and added, then converted back to a percentage. That is what lets "this is a category we know" and "this is four times bigger than anything we have won" combine sensibly instead of one silently overwriting the other.
5. What it deliberately does not know
Whether we can actually meet the specification. The score reads the buyer, the category, the size and the field — it does not read the tender documents. That is what the AI review on a tender's page is for: it reads the specification, the buyer's history and past winners, and gives a second opinion on the number.
Worked example — SPA260000807: კომპიუტერული აქსესუარების შესყიდვა
| Step | Evidence | Effect |
|---|---|---|
| Base rate | 20 of 56 decided bids won | 36% |
| ▲ Expected competition | this category averages 1.6 bidders, 40% go single-bid. In 1–3 bidders we have won 12 of 21. | +0.68 |
| ▲ Deal size | ₾10k–50k: we have won 9 of 16 bids in this range. | +0.61 |
| ▲ Our line of business | reads like our own bids (კომპიუტერული, აქსესუარების) — 65% match to our bidding history. Closest thing we have bid on: "კომპიუტერული მოწყობილობებისა და აქსესუარების სახელმწიფო შესყიდვა". | +0.50 |
| ▼ Procedure | SPA: 0 wins from 1 bids. | -0.23 |
| ▲ Buyer relationship | we have bid 2× with this buyer and won 1. | +0.15 |
| — Category fit | 30200000: 11 wins from 29 bids. | +0.08 |
| Score | 77% — strong, helped by expected competition and helped by deal size. | 77% |
Effects are in log-odds; +0.69 doubles the odds, −0.69 halves them. Open this tender →
Reading it
- Above 60% — the shape of a deal we win: right size, a category we know, a field we can beat. Worth a proper look, and the AI review.
- 30–60% — mixed evidence; read the factors and decide which one you believe.
- Below 30% — usually the wrong category, the wrong size, or a buyer with an incumbent. Bid only with a reason the model cannot see.
The profile behind the score is recomputed whenever a tender we bid on is decided; the category odds come from the whole portal and refresh daily.