Robot Tender Evaluation Criteria

What Should a Robot Tender Be Scored On?

Score on six weighted criteria: coverage capability, five-year cost, integration and data, service and uptime, safety and compliance, and support maturity. Price carries 30 to 40 per cent of the weight, not 80, because a service robot's unit price is a minority of what the buyer spends over the contract. The remaining weight should sit on the factors that decide whether the fleet works in year three, and each criterion needs a stated evidence requirement so that scoring is a verification exercise rather than a comparison of prose quality.

The Weighted Matrix

A workable matrix for a mid-sized deployment. Weights below are for a single-site cleaning or delivery fleet with an expected contract value in the low six figures; adjust the price weight upward for commodity units and downward where integration is complex.

Criterion	Weight	Evidence required	Scoring band
Coverage capability against the specification	25%	Measured throughput in m² per hour on a comparable floor type; site visit or video of a live unit	0-5 against stated floor area
Five-year total cost	30%	Costed schedule: hardware, service, consumables, energy, connectivity, exit	Formula-based, see below
Integration and data access	15%	Published API documentation; sample export; statement of BMS/CMMS integration	0-5 against named systems
Service, uptime and response	15%	Named SLA with credits; local engineer count and response time; parts stock location	0-5 against SLA floors
Safety and regulatory compliance	10%	Test certificate or declaration; risk assessment; conformity documentation	Pass/fail gate plus 0-5
Support maturity and references	5%	Three reference sites of comparable size and sector, contactable	0-5 against verifiable references

The 25 per cent on coverage is not a courtesy to the technical team. Coverage is the criterion where bids most often diverge in ways that are invisible at award and obvious at month three: a spec-compliant unit can be half the real throughput of another on the same floor once doors, lifts and congestion are included. Demand measured throughput, not the datasheet figure.

The safety criterion is deliberately split into a gate and a score. A bid that cannot produce conformity documentation fails the gate regardless of how well it scores elsewhere, because awarding to a non-conforming unit creates a liability the evaluator owns. The requirement itself is described in the safety standards and compliance guide.

Price-Quality Ratio Arithmetic, Worked

Where the framework allows a price-quality ratio, the arithmetic has a specific shape and the shape matters. The standard formulation divides the lowest compliant price by the bid price, or the lowest score by the bid score, applies the weight, and sums. Two worked bids make the behaviour visible.

Bid	Five-year cost	Quality score (out of 100)	Price component	Quality component	Total (30/70)
A	180,000	62	1.00 x 30 = 30.0	0.62 x 70 = 43.4	73.4
B	210,000	88	0.857 x 30 = 25.7	0.88 x 70 = 61.6	87.3

Bid B costs 16.7 per cent more and wins by 13.9 points. That is the intended behaviour of a 30/70 split, and it is also the outcome a panel that "just looked at the numbers" would have got wrong. The reverse split, 70/30, produces 70.0 for A and 59.9 for B, and the buyer saves 30,000 today while paying an extra 30,000 in cost over five years, at lower quality. Neither split is universally correct; what is not acceptable is leaving the split unstated until the scores are visible.

Two properties of the ratio formula are worth knowing before the tender is published. First, the formula is sensitive to the lowest price, so a single abnormally low bid distorts every other price component. Second, a bid with a very low quality score can still win on price if the price weight is high enough, which is why the safety gate exists as a floor rather than a weighted score.

Under EU public-procurement rules the ratio must be set out in the tender documents with the method for calculating it, and it cannot be changed after bids are opened. The same discipline is good practice in private procurement even where no rule compels it, because an evaluation method that appears after the bids is the most common ground for challenge.

Award Criteria That Survive a Challenge

Three properties make a criterion defensible. It must be linked to the subject matter of the contract, it must not confer unrestricted freedom of choice on the buyer, and it must be capable of objective assessment from evidence the bidder can supply.

Weak criterion	Why it fails	Rewritten version
"Quality of the solution"	Unrestricted freedom; two panels could score the same bid differently	"Measured throughput in m² per hour on a comparable floor, verified by site visit or timestamped video"
"Good local support"	Not measurable; no stated threshold	"Engineer within 90 minutes travel of site, stated parts stock location, SLA with financial credits"
"Reputable supplier"	Discriminatory risk; substitutes brand for evidence	"Three reference sites of comparable size and sector, contactable during evaluation, operating more than 12 months"
"Future-proof technology"	Unverifiable at award	"Published API documentation and a stated integration commitment for named BMS and CMMS products"

The pattern in every rewrite is the same: replace an adjective with a measurement and name the evidence. A criterion that cannot state its evidence will be scored on the quality of the bidder's writing, which is a real and common failure mode in technical panels.

The Five-Year Price Trap

Unit price and five-year cost are different numbers, and the gap between them is where evaluation budgets go wrong. The trap is not that bidders lie; it is that bids are structurally incomparable unless the buyer defines the cost schedule.

A worked comparison on a six-unit cleaning fleet. Bid A quotes a lower unit price and Bid B a higher one. The full schedule tells a different story.

Cost line, five years	Bid A	Bid B
Hardware, 6 units	108,000	126,000
Service and maintenance contract	54,000	36,000
Consumables (brushes, pads, detergent)	9,600	7,200
Energy	4,800	4,200
Connectivity and software subscription	18,000	12,000
Expected unplanned downtime cost	21,600	6,000
Battery replacement (year 4)	0, included	7,800
Total	216,000	199,200

Bid A is 14 per cent cheaper on hardware and 8 per cent more expensive over the contract. The line that reverses the ranking is unplanned downtime, and it is the line no bid includes unless the buyer requires it. Requiring a downtime estimate forces each bidder to state its own availability assumption, which can then be tested against the SLA it offers.

Defining the cost schedule in the tender is the single highest-value preparation step in the whole evaluation. It makes bids comparable, it surfaces the consumable and subscription costs that are otherwise discovered later, and it removes the argument at moderation. The underlying ledger structure is set out in the cleaning robot total cost of ownership guide, and the grants, tax and depreciation guide covers the treatment of the capital line, which can change the ranking again if incentives are available.

Verification: What Evidence You Can Demand

Every criterion in the matrix has a verification question, and the tender should state it. An evaluator who cannot name the evidence will score the bid's confidence rather than the bid's substance.

• Coverage: a site visit to an operating unit, or timestamped video of a full cleaning cycle on a comparable floor, with the area and duration stated.
• Cost: a completed cost schedule in the buyer's own template, signed, with an explicit statement of what is excluded.
• Integration: API documentation supplied with the bid, plus a live demonstration of an export into the buyer's CMMS or BMS format.
• Service: the SLA text as it will be signed, naming response times, credit percentages and the escalation path.
• Safety: the conformity documentation and risk assessment, with the standard versions cited.
• References: three contactable sites, with the buyer permitted to call them during evaluation.

The reference calls deserve a script rather than a chat. Ask each reference three questions: how many units, how many months, and what broke. The third question produces the useful answer, and a reference who hesitates on it has told you something. The contract terms that follow award are covered in the uptime SLA contract guide and the warranty contract terms guide; both are worth reading before the tender is drafted, because their terms should appear as requirements rather than arrive as negotiation.

Exclusion Grounds vs Score Loss

These are different mechanisms and conflating them is the most common procedural error in a robot tender. Exclusion removes a bid from consideration; score loss ranks it lower. Getting the distinction wrong either awards to a bid that should have been excluded or excludes a bid that should simply have scored badly.

Issue	Correct mechanism	Why
Missing safety conformity documentation	Exclusion, or mandatory pass/fail gate	Cannot be remedied by a good score elsewhere
Higher five-year cost	Score loss	A legitimate commercial choice by the bidder
No local engineer	Score loss, unless the tender made it mandatory	An evaluation factor, not a compliance failure
Abnormally low price	Verification request, then exclusion if unexplained	Requires a process, not an automatic rejection
Late submission	Exclusion	Procedural integrity
Unclear API statement	Clarification request, then score loss	Ambiguity is remediable within the rules

The abnormally low price row is the one that gets handled badly. An unusually low bid is not automatically excluded; it triggers a request for explanation, and only an unsatisfactory explanation justifies rejection. Skipping that step, in either direction, is a challenge waiting to happen. The bidder-side view of the same arithmetic is set out in the cleaning contract tender pricing guide, and it is worth reading before writing the tender because it shows where bidders will compress costs and therefore where your verification should concentrate.

Moderating the Panel

Individual scores are inputs, not the award. Moderation is where a panel converts them into a decision it can defend, and the process needs three properties to work.

First, independent scoring before discussion. Each evaluator scores every bid against the matrix alone and submits before the panel meets. Consensus-first scoring produces anchoring, where the most senior voice sets the range and the others converge on it.

Second, recorded reasons per score. For any score at the extremes of a band, the evaluator writes one sentence citing the evidence. A record that says "4, good coverage" is not a reason and will not survive a challenge.

• Score each bid against every criterion, with a written justification for any score outside the middle band.
• Discuss only the criteria where scores diverge by two or more points.
• Do not re-score a criterion because a bidder is stronger overall; that is how the highest-scoring column becomes a halo.
• Record the moderation changes and their reasons in the evaluation report, not just the final numbers.

Third, separate technical and commercial opening where the rules require it. Where price is not to influence technical scoring, the technical panel scores blind and the prices are opened afterwards. In practice the strongest protection against price anchoring is that the evaluators have never seen the numbers.

Publishing the Outcome

The evaluation report is the document that protects the award. It should contain the matrix as published, the individual scores, the moderated scores with reasons, the cost schedules as submitted, and the arithmetic that produced the ranking. Where the framework requires a standstill period or a debrief to unsuccessful bidders, the report is what those are written from.

A debrief that can cite the specific criterion and the specific evidence is a debrief that ends the conversation. One that says "your bid was not successful on this occasion" invites the next step. The same documentation discipline serves the contract management phase, where the awarded SLA and cost schedule become the baseline against which performance is measured. The vendor evaluation framework covers the pre-tender shortlisting stage, and the acceptance testing and commissioning guide covers the point where the promised numbers are actually tested, which is where an evaluation that scored on evidence rather than prose pays for itself.

A tender written this way takes perhaps two additional days of preparation. It routinely saves more than it costs on the first contract, and it avoids the failure that costs the most: a procurement restarted after a challenge, with the deployment a full year behind the original plan.