How does AI bid scoring work on VEXORS?

When the bids are in, the buyer triggers an evaluation. The system reads every qualifying bid against the same criteria, the priced bill of quantities and any questionnaire the buyer set, and produces a ranked comparison plus a recommended bid with a short reason. The buyer reviews the ranking and makes the final award. The scoring does the consistent reading; the buyer keeps the decision.

Can the AI award a contract on its own?

No. Scoring produces a ranking and a recommendation only. The buyer always makes the award manually and can award a bid that the scoring did not rank first, for reasons that live outside the data such as a known delivery history or a strategic supplier choice. The AI informs the decision; it never makes it.

Does the AI know the market price?

No. The scoring evaluates only the bids submitted, against the buyer's own criteria. It has no access to external market prices and does not claim to know what something 'should' cost. It tells you which submitted bid is strongest on your terms, not whether the market could do better.

What can disqualify a bid regardless of price?

If the buyer marks a requirement as a show-stopper, a certification, a delivery window, a mandatory term, and a bid fails it, that bid is disqualified no matter how low its price. This is deliberate: it stops a cheap but non-compliant bid from ranking ahead of bids that actually meet the requirements.

How much does running an evaluation cost?

AI scoring runs on credits. An evaluation run uses AI credits from your balance, and each plan includes a one-time credit grant on activation, with top-up packs available to every plan if you need more. Credits are only consumed when you run a feature, so drafting and reviewing without scoring does not spend them.

What should I do when two bids score close together?

Treat a near-tie as the scoring telling you the data cannot separate them, which is useful information in itself. The decision then moves to what the data does not hold: delivery history with your team, responsiveness during clarifications, or the supplier's trust score built from completed work on the network. A close score is an invitation to look at those factors, not a coin flip.

How should I structure a request so the scoring is most useful?

Itemize what you are buying as line items with quantities, state the mandatory requirements explicitly, and use a questionnaire for anything you want every bidder to answer. The scoring compares what bidders submit against what you asked; the more structure the request has, the sharper the comparison it can produce.

Back to blogAI

How AI bid scoring actually works (and what it will not do)

The VEXORS TeamJune 17, 20269 min read

"AI bid scoring" is one of those phrases that sounds either impressive or alarming depending on who you ask. To a buyer drowning in five inconsistent quotes, it sounds like relief. To a buyer who has been burned by a tool that overpromised, it sounds like a black box about to make a decision it has no business making. Both reactions come from not knowing what the scoring actually does.

So here is the honest mechanics. What gets scored, how, what stops a bid cold, and the line the scoring is not allowed to cross. This is the deeper look at one feature; for the wider picture of where AI helps and where it does not, start with AI in procurement: where it actually helps.

What the scoring is comparing

Scoring only works because the bids share a structure. When a buyer publishes a request with a bill of quantities, every supplier prices the same line items in the same shape. That is what makes a fair comparison possible in the first place. Five PDFs in five formats cannot be scored. A set of structured bids against one list can.

The evaluation reads each qualifying bid against the criteria the buyer already defined:

Criterion	What it weighs
Commercial	Price, quantity compliance, and delivery against the bill of quantities
Questionnaire	The buyer's own qualifying questions, where they used them
Show-stoppers	Pass-or-fail requirements that gate the whole bid

The buyer sets what matters before any bid arrives. The scoring then applies it the same way to every supplier, which is the entire point.

Why consistency is the real benefit

The value of a scoring pass is not that a machine is smarter than you. It is that a machine is not tired, biased, or anchored. A person comparing the eighth bid late in the day does not read it the way they read the first. They have a favourite by now, and they are looking for confirmation.

A scoring pass treats bid eight exactly like bid one. That consistency, not any cleverness, is what catches the strong bid you were about to overlook and the cheap one that is weak where it counts.

The output is a ranking with a recommended bid and a short reason for it. It surfaces the offer that looks good on price but misses on compliance, and the one that is suspiciously low on a single line. You still read the bids. You just start from an even-handed comparison instead of a gut feeling.

What disqualifies a bid

Not every criterion is a slider. Some are gates. If the buyer marks a requirement as a show-stopper, a required certification, a hard delivery date, a mandatory term, then a bid that fails it is disqualified regardless of its price.

This is one of the most useful things the scoring does, because it enforces the discipline humans let slip. A very low price is psychologically loud. It is tempting to wave through a missing requirement because the number is attractive. A show-stopper does not get tempted. The non-compliant bid drops out, and the comparison stays honest.

The line the scoring will not cross

Here is the part that should reassure the skeptics. The scoring ranks and recommends. It does not award.

The award stays with the buyer, always, and as a manual action. The buyer can accept the recommendation, or award a lower-ranked bid for a reason that is not in the data: a supplier whose delivery slipped twice last year, a strategic choice to diversify away from one vendor, a relationship that matters this quarter. None of that lives in a bid document, so none of it belongs to the model.

The scoring also does not know the market. It evaluates the bids in front of it against your criteria. It has no feed of external prices and makes no claim about what something "should" cost. It tells you which submitted bid is strongest on your terms. Whether the market could do better is your judgment, informed by your own knowledge of the category.

That boundary, the model prepares the comparison and the human owns the decision, is not a limitation to apologise for. It is what makes the feature safe to use.

What it costs to run

Practical detail, because it shapes how you use it. Scoring runs on credits. An evaluation uses AI credits from your balance, each plan includes a one-time grant of credits on activation, and top-up packs are available if you run more evaluations than your grant covers. Credits are only spent when you actually run a feature, so you can draft a request, review it, and collect bids without touching your balance. You spend credits at the point you choose to score. The current credit costs are shown on the pricing page.

How to read a score, and when to override it

A ranked list with a recommendation invites a lazy reading: top score wins, done. That is not how it is meant to be used, and the buyers who get the most out of scoring read it differently.

The score is an answer to one precise question: of the bids submitted, which best satisfies the criteria this buyer wrote down? Three readings follow from that.

A wide gap is a strong signal. When the top bid clears the second by a distance, the data is saying the comparison was not close on your own terms. Overriding that requires a reason worth writing down, and writing it down is exactly what you should do, because the award record will outlive your memory of the moment.

A narrow gap is a different signal. Two bids within a whisker of each other mean the written criteria cannot separate them. That is where judgment belongs: the supplier's trust score on the network, their behavior during clarifications, your team's delivery history with them. The scoring has done its job by telling you the paperwork is a tie.

An outlier price deserves suspicion in both directions. A bid far below the cluster either found an efficiency everyone else missed or misread the scope. The written rationale beside each score helps here, but so does a clarification question before award. Cheap surprises after award are the most expensive kind.

Overriding the recommendation is legitimate and the platform treats it that way: the buyer always makes the award. The discipline is not "follow the machine." It is "when you depart from the ranking, know why, and record why."

Preparing your request so the scoring works harder

The quality of an evaluation is set before any bid arrives, by the shape of the request. Scoring compares bids against what you asked for, so vague requests produce mushy comparisons no matter how good the model is.

Three habits raise the ceiling:

Itemize. A request whose scope is one paragraph gets bids whose prices are one number, and one number resists comparison. Line items with quantities and units force each bidder to price the same structure, and give the scoring something concrete to align.
Declare the show-stoppers. If a certification, a delivery window, or a payment term is genuinely mandatory, mark it so. That is what lets a non-compliant cheap bid be excluded automatically instead of anchoring your comparison.
Ask everyone the same questions. A questionnaire attached to the request means every bidder answers the same prompts in the same structure. Free-form cover letters are where inconsistency creeps back in.

None of this is extra work invented for the AI. It is the same discipline a good manual evaluation always needed. The difference is that a structured request plus scoring turns that discipline into minutes instead of an afternoon.

The short version

AI bid scoring is not a decision engine and should not be sold as one. It reads structured bids against the criteria you set, applies them consistently to every supplier, disqualifies bids that miss your hard requirements, and hands you a ranked comparison with a recommendation. Then it stops, because the award is yours. Used that way, it removes the tired, biased re-reading and leaves you the part that needs a human: the decision.

Want to see the scoring in context, alongside drafting and supplier matching? Explore the AI features on VEXORS and see exactly where the model helps and where you stay in control.

Frequently asked questions

How does AI bid scoring work on VEXORS?: When the bids are in, the buyer triggers an evaluation. The system reads every qualifying bid against the same criteria, the priced bill of quantities and any questionnaire the buyer set, and produces a ranked comparison plus a recommended bid with a short reason. The buyer reviews the ranking and makes the final award. The scoring does the consistent reading; the buyer keeps the decision.
Can the AI award a contract on its own?: No. Scoring produces a ranking and a recommendation only. The buyer always makes the award manually and can award a bid that the scoring did not rank first, for reasons that live outside the data such as a known delivery history or a strategic supplier choice. The AI informs the decision; it never makes it.
Does the AI know the market price?: No. The scoring evaluates only the bids submitted, against the buyer's own criteria. It has no access to external market prices and does not claim to know what something 'should' cost. It tells you which submitted bid is strongest on your terms, not whether the market could do better.
What can disqualify a bid regardless of price?: If the buyer marks a requirement as a show-stopper, a certification, a delivery window, a mandatory term, and a bid fails it, that bid is disqualified no matter how low its price. This is deliberate: it stops a cheap but non-compliant bid from ranking ahead of bids that actually meet the requirements.
How much does running an evaluation cost?: AI scoring runs on credits. An evaluation run uses AI credits from your balance, and each plan includes a one-time credit grant on activation, with top-up packs available to every plan if you need more. Credits are only consumed when you run a feature, so drafting and reviewing without scoring does not spend them.
What should I do when two bids score close together?: Treat a near-tie as the scoring telling you the data cannot separate them, which is useful information in itself. The decision then moves to what the data does not hold: delivery history with your team, responsiveness during clarifications, or the supplier's trust score built from completed work on the network. A close score is an invitation to look at those factors, not a coin flip.
How should I structure a request so the scoring is most useful?: Itemize what you are buying as line items with quantities, state the mandatory requirements explicitly, and use a questionnaire for anything you want every bidder to answer. The scoring compares what bidders submit against what you asked; the more structure the request has, the sharper the comparison it can produce.