What Matters
A procurement agent can choose the lowest quoted number and still select the wrong purchase. DN’s eight synthetic RFQ cases separate price normalization from eligibility and savings measurement. The rule that checks eligibility first chooses seven admissible offers and holds one award. These are reproducible decisions on authored inputs, not evidence of live supplier negotiation, AI performance or realised business savings.
The decision needs a comparable purchase
The interesting procurement question is which offer meets the requirement at an acceptable cost. A price detached from pack size, delivery, specifications or payment terms cannot answer it. A clean-looking ranking can conceal an unusable choice.
DN’s agent-to-agent commerce analysis examines discovery, negotiating authority and settlement infrastructure. This benchmark tests a narrower component: what deterministic selection rules do with supplied quotes. It gives a future agent evaluation an explicit decision contract before adding language models, supplier messages or purchasing authority.
Each case specifies the same required quantity within that case, an acceptable delivery window, supplier and specification conditions, permitted overbuy and payment terms. Its incumbent baseline is also authored as a compliant purchase for the stated requirement. The eight purchases are separate examples; they are not a procurement programme whose savings should be added together.
Five workflow stages, with explicit evidence boundaries
A useful evaluation separates the stages that an agent might eventually perform. Passing arithmetic checks does not establish that it can complete the surrounding work.
- Discover and verify suppliers. Collect candidates and evidence of identity, capability and capacity. This release supplies fictional candidates and approval flags; it does not discover suppliers or validate those flags against external records.
- Normalize quotes. Translate pack quantities, minimum orders, discounts, quote currencies and declared charges into a comparable purchase subtotal. This is executed by the benchmark, including explicit handling of an unknown charge.
- Negotiate within authority. Request or evaluate changes to price and other terms. Here, revised terms are authored inputs. No buyer agent persuaded a seller, protected a reservation price or obtained a concession.
- Evaluate and approve. Apply declared requirements, compare eligible offers and record the decision. The rules exercise selection and HOLD logic; they do not implement an organization’s actual access controls or approval system.
- Award, reconcile and measure. Verify the accepted order, delivery, invoice and payment against the baseline. This release calculates conditional savings only. It does not issue orders, receive goods or observe an invoice being paid.
New Zealand Government Procurement’s due-diligence guidance calls for corroborating supplier information, including identity, financial ability and delivery capability, and documenting the checks. Those are useful general practice references here; an authored approved field is not equivalent evidence. [3]
Three rules expose different failure modes
The sticker-price control selects the smallest raw number in the price-per-pack field. It deliberately ignores pack size, currency, discounts, additional charges and eligibility. It is a limited comparison rule, not a representation of a commercial procurement product or an AI model.
The normalized landed-cost control calculates the complete purchase subtotal and selects the lowest. It excludes incomplete costs but ignores the non-cost eligibility conditions. This separates an accounting problem from an authorization problem: a correct subtotal can still belong to an unacceptable offer.
The eligibility-first rule filters the declared requirements before comparing complete costs. It checks supplier approval, specification match, quote validity, delivery, overbuy, payment terms and authorization of revised terms. If no offer remains, it returns HOLD. Equal costs are resolved by offer ID, a deterministic convention rather than a substantive supplier preference.
Across the eight curated cases, the sticker control selects four ineligible offers; the normalized-cost control selects three. The gated rule selects seven eligible offers and holds the remaining case. These counts describe the constructed examples. They are not accuracy estimates, independent observations or evidence that a particular agent will perform similarly.
New Zealand’s evaluation guidance distinguishes critical pass/fail preconditions from other evaluation criteria and cautions against excessive preconditions. DN’s gates should therefore be read as declared fixture requirements, not a universal template. A real buyer must decide which requirements are essential before seeing the preferred price. [2]
Normalize what must actually be purchased
Case P01 shows the pack-size problem. Offer A lists 10 CU per single unit. Buying 100 units and paying 100 CU freight costs 1,100 CU. Offer B lists 90 CU per ten-unit pack; ten packs plus 50 CU freight cost 950 CU. The larger displayed pack price produces the lower complete purchase cost.
Minimum orders require a second check. In P03, a 6-CU unit price requires purchasing at least 200 units to meet a demand of 100. That means 1,200 CU leaves the buyer, with 100 excess units. The purchase costs 12 CU per required unit before any additional charges, and breaches this case’s overbuy limit.
The benchmark gives surplus stock no automatic resale or future-use credit. A business may have evidence that excess inventory has value, but it must model the associated demand, storage, deterioration and timing separately. Dividing expenditure by every purchased unit can obscure the cash required to meet today’s smaller need.
Currency labels also need an explicit conversion basis. P04 uses invented quote currencies and authored conversion factors; none is a market exchange rate. P07 contains unknown freight. Its incomplete offer is not assigned zero freight or declared cheaper than a complete alternative.
The resulting number is a landed purchase subtotal: discounted items after conversion, freight, nonrecoverable tax and onboarding. It is narrower than full lifecycle total cost of ownership. New Zealand’s TCO guide includes broader acquisition, operating and maintenance costs, with residual value considered over the relevant lifecycle. This model does not evaluate those future flows. [1]
A discount can change the wrong term
P06 supplies a 30% discount that produces a 720-CU purchase subtotal, but its immediate-payment requirement falls outside the permitted terms and the revision is not authorized. The cost-only rule selects it. The eligibility-first rule instead selects a different offer with an authorized 5% discount and a 908.50-CU subtotal.
Nothing negotiated those discounts during the test. The case checks whether a selection rule respects supplied terms after a hypothetical revision. It cannot establish negotiation skill, seller acceptance, achievable concessions or the causal benefit of using an agent.
New Zealand’s negotiation guidance treats preparation, deliverables, assumptions, price and other contractual variables as part of the negotiation. That supports evaluating the whole authorized proposal rather than highlighting a discount while overlooking changed service or payment commitments. [4]
In an operating workflow, retain the original quote, revised offer and approval for the specific change. If quantity, specifications, delivery or payment conditions move, run the comparison again. A favourable number calculated against an obsolete requirement is not a decision-ready result.
Use the calculator without turning assumptions into savings
The Procurement Agent Savings Calculator separates purchase cost, workflow cash spending and owner-time valuation. Compare a selected eligible offer with the corresponding compliant baseline. An incomplete, ineligible or absent selection has no reported savings figure; HOLD is not presented as a zero-cost purchase.
View all 24 decisions and conditional savings
N/A means savings are unavailable, not zero. Controls can select an ineligible quote; its known subtotal is shown without claiming a comparable saving. Owner time is a separate noncash valuation.
| Case | Rule | Selected offer | Decision status | Landed cost (CU) | Baseline (CU) | Workflow cash (CU) | Gross purchase difference (CU) | Conditional cash-net (CU) | Conditional cash-net / baseline | Owner-time value (CU, noncash) | Net after time value (CU) | Failed declared checks |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| P01 | Sticker price | P01-A | ELIGIBLE_SELECTION | 1,100.00 CU | 1,200.00 CU | 20.00 CU | 100.00 CU | 80.00 CU | 6.67% | 80.00 CU | 0.00 CU | All declared checks pass |
| P01 | Normalized cost | P01-B | ELIGIBLE_SELECTION | 950.00 CU | 1,200.00 CU | 20.00 CU | 250.00 CU | 230.00 CU | 19.17% | 80.00 CU | 150.00 CU | All declared checks pass |
| P01 | Eligibility first | P01-B | ELIGIBLE_SELECTION | 950.00 CU | 1,200.00 CU | 20.00 CU | 250.00 CU | 230.00 CU | 19.17% | 80.00 CU | 150.00 CU | All declared checks pass |
| P02 | Sticker price | P02-A | ELIGIBLE_SELECTION | 1,200.00 CU | 1,200.00 CU | 25.00 CU | 0.00 CU | -25.00 CU | -2.08% | 60.00 CU | -85.00 CU | All declared checks pass |
| P02 | Normalized cost | P02-B | ELIGIBLE_SELECTION | 940.00 CU | 1,200.00 CU | 25.00 CU | 260.00 CU | 235.00 CU | 19.58% | 60.00 CU | 175.00 CU | All declared checks pass |
| P02 | Eligibility first | P02-B | ELIGIBLE_SELECTION | 940.00 CU | 1,200.00 CU | 25.00 CU | 260.00 CU | 235.00 CU | 19.58% | 60.00 CU | 175.00 CU | All declared checks pass |
| P03 | Sticker price | P03-A | INELIGIBLE_SELECTION | 1,200.00 CU | 1,100.00 CU | 20.00 CU | N/A | N/A | N/A | 80.00 CU | N/A | Overbuy limit exceeded |
| P03 | Normalized cost | P03-B | ELIGIBLE_SELECTION | 970.00 CU | 1,100.00 CU | 20.00 CU | 130.00 CU | 110.00 CU | 10.00% | 80.00 CU | 30.00 CU | All declared checks pass |
| P03 | Eligibility first | P03-B | ELIGIBLE_SELECTION | 970.00 CU | 1,100.00 CU | 20.00 CU | 130.00 CU | 110.00 CU | 10.00% | 80.00 CU | 30.00 CU | All declared checks pass |
| P04 | Sticker price | P04-A | ELIGIBLE_SELECTION | 1,150.00 CU | 1,200.00 CU | 20.00 CU | 50.00 CU | 30.00 CU | 2.50% | 40.00 CU | -10.00 CU | All declared checks pass |
| P04 | Normalized cost | P04-B | ELIGIBLE_SELECTION | 1,000.00 CU | 1,200.00 CU | 20.00 CU | 200.00 CU | 180.00 CU | 15.00% | 40.00 CU | 140.00 CU | All declared checks pass |
| P04 | Eligibility first | P04-B | ELIGIBLE_SELECTION | 1,000.00 CU | 1,200.00 CU | 20.00 CU | 200.00 CU | 180.00 CU | 15.00% | 40.00 CU | 140.00 CU | All declared checks pass |
| P05 | Sticker price | P05-A | INELIGIBLE_SELECTION | 700.00 CU | 1,100.00 CU | 25.00 CU | N/A | N/A | N/A | 40.00 CU | N/A | Supplier not approved; Specification mismatch |
| P05 | Normalized cost | P05-A | INELIGIBLE_SELECTION | 700.00 CU | 1,100.00 CU | 25.00 CU | N/A | N/A | N/A | 40.00 CU | N/A | Supplier not approved; Specification mismatch |
| P05 | Eligibility first | P05-C | ELIGIBLE_SELECTION | 950.00 CU | 1,100.00 CU | 25.00 CU | 150.00 CU | 125.00 CU | 11.36% | 40.00 CU | 85.00 CU | All declared checks pass |
| P06 | Sticker price | P06-C | ELIGIBLE_SELECTION | 950.00 CU | 1,100.00 CU | 30.00 CU | 150.00 CU | 120.00 CU | 10.91% | 60.00 CU | 60.00 CU | All declared checks pass |
| P06 | Normalized cost | P06-A | INELIGIBLE_SELECTION | 720.00 CU | 1,100.00 CU | 30.00 CU | N/A | N/A | N/A | 60.00 CU | N/A | Payment terms outside authority; Revised terms not authorized |
| P06 | Eligibility first | P06-B | ELIGIBLE_SELECTION | 908.50 CU | 1,100.00 CU | 30.00 CU | 191.50 CU | 161.50 CU | 14.68% | 60.00 CU | 101.50 CU | All declared checks pass |
| P07 | Sticker price | P07-A | INELIGIBLE_SELECTION | N/A | 1,100.00 CU | 20.00 CU | N/A | N/A | N/A | 40.00 CU | N/A | Cost incomplete |
| P07 | Normalized cost | P07-C | ELIGIBLE_SELECTION | 950.00 CU | 1,100.00 CU | 20.00 CU | 150.00 CU | 130.00 CU | 11.82% | 40.00 CU | 90.00 CU | All declared checks pass |
| P07 | Eligibility first | P07-C | ELIGIBLE_SELECTION | 950.00 CU | 1,100.00 CU | 20.00 CU | 150.00 CU | 130.00 CU | 11.82% | 40.00 CU | 90.00 CU | All declared checks pass |
| P08 | Sticker price | P08-A | INELIGIBLE_SELECTION | 700.00 CU | 1,100.00 CU | 20.00 CU | N/A | N/A | N/A | 40.00 CU | N/A | Supplier not approved |
| P08 | Normalized cost | P08-A | INELIGIBLE_SELECTION | 700.00 CU | 1,100.00 CU | 20.00 CU | N/A | N/A | N/A | 40.00 CU | N/A | Supplier not approved |
| P08 | Eligibility first | HOLD | HOLD | N/A | 1,100.00 CU | 20.00 CU | N/A | N/A | N/A | 40.00 CU | N/A | No eligible offer — award held |
The inputs, gates and arithmetic are visible
All eight cases require 100 units, delivery within 14 days, no more than 10% overbuy and payment no earlier than 30 days. Quotes are evaluated as of 22 October 2026. Each baseline is an authored compliant incumbent purchase for the same required quantity, specification, deadline and permitted terms. No avoided baseline labor value is assumed.
| Case | Scenario | Required units | Baseline landed cost | Workflow cash cost | Review hours | CU per review hour |
|---|---|---|---|---|---|---|
| P01 | Pack-size mismatch | 100 | 1,200.00 CU | 20.00 CU | 2 | 40.00 CU |
| P02 | Freight and onboarding | 100 | 1,200.00 CU | 25.00 CU | 1.5 | 40.00 CU |
| P03 | Minimum-order overbuy | 100 | 1,100.00 CU | 20.00 CU | 2 | 40.00 CU |
| P04 | Fictional quote currencies | 100 | 1,200.00 CU | 20.00 CU | 1 | 40.00 CU |
| P05 | Eligibility before ranking | 100 | 1,100.00 CU | 25.00 CU | 1 | 40.00 CU |
| P06 | Conditional negotiated discount | 100 | 1,100.00 CU | 30.00 CU | 1.5 | 40.00 CU |
| P07 | Unknown freight | 100 | 1,100.00 CU | 20.00 CU | 1 | 40.00 CU |
| P08 | No eligible offer | 100 | 1,100.00 CU | 20.00 CU | 1 | 40.00 CU |
Declared gates: supplier approved, specification matches, quote not expired, delivery within the case limit, purchased units within the overbuy limit, payment days at least the case minimum, revised terms authorized and every modeled cost input known. These flags are supplied fixture inputs, not verified supplier records or a legal-compliance assessment.
- Packs ordered = max(ceiling(required units ÷ units per pack), minimum packs). Purchased units = packs ordered × units per pack.
- Item cost in CU = packs ordered × quoted price per pack × (1 − discount ÷ 100) × conversion factor in CU per quote-currency unit.
- Landed purchase subtotal = item cost + freight + nonrecoverable tax + onboarding. All three added charges are already in CU.
- For an eligible selection, gross purchase difference = baseline − selected landed subtotal. Conditional cash-net saving = gross difference − workflow cash cost; its percentage divides by the positive baseline.
- Owner-time value = review hours × CU per hour. Net after this noncash valuation = conditional cash-net saving − owner-time value. No owner-time saving is assumed.
- Unknown cost, an ineligible selected offer or HOLD makes all savings figures unavailable. Negative results remain negative. The eligibility-first rule uses the offer ID to break equal-cost ties.
The model pays for all purchased packs without assigning surplus inventory a resale credit. Its landed subtotal is not a full lifecycle cost estimate. The downloadable editable calculator requires a positive demand and baseline; it validates quantities and input ranges and does not verify the truth of the supplied eligibility declarations.
Inspect all 24 quote inputs, normalized costs and eligibility checks
QCU-A and QCU-B are invented currency labels. Conversion factors are authored, not market rates. Unknown freight remains unknown; that quote’s cost outputs are withheld. Every supplier is fictional and identified by its quote ID.
| Offer | Quoted price per pack | Units per pack | Minimum packs | Packs ordered | Units purchased | CU per quote unit | Discount | Freight (CU) | Nonrecoverable tax (CU) | Onboarding (CU) | Item subtotal (CU) | Landed subtotal (CU) |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| P01-A | 10 CU | 1 | 1 | 100 | 100 | 1 | 0% | 100.00 CU | 0.00 CU | 0.00 CU | 1,000.00 CU | 1,100.00 CU |
| P01-B | 90 CU | 10 | 1 | 10 | 100 | 1 | 0% | 50.00 CU | 0.00 CU | 0.00 CU | 900.00 CU | 950.00 CU |
| P01-C | 48 CU | 5 | 1 | 20 | 100 | 1 | 0% | 80.00 CU | 0.00 CU | 0.00 CU | 960.00 CU | 1,040.00 CU |
| P02-A | 8 CU | 1 | 1 | 100 | 100 | 1 | 0% | 300.00 CU | 0.00 CU | 100.00 CU | 800.00 CU | 1,200.00 CU |
| P02-B | 9.2 CU | 1 | 1 | 100 | 100 | 1 | 0% | 20.00 CU | 0.00 CU | 0.00 CU | 920.00 CU | 940.00 CU |
| P02-C | 9.5 CU | 1 | 1 | 100 | 100 | 1 | 0% | 50.00 CU | 20.00 CU | 0.00 CU | 950.00 CU | 1,020.00 CU |
| P03-A | 6 CU | 1 | 200 | 200 | 200 | 1 | 0% | 0.00 CU | 0.00 CU | 0.00 CU | 1,200.00 CU | 1,200.00 CU |
| P03-B | 9.5 CU | 1 | 1 | 100 | 100 | 1 | 0% | 20.00 CU | 0.00 CU | 0.00 CU | 950.00 CU | 970.00 CU |
| P03-C | 98 CU | 10 | 1 | 10 | 100 | 1 | 0% | 0.00 CU | 0.00 CU | 0.00 CU | 980.00 CU | 980.00 CU |
| P04-A | 7 QCU-A | 1 | 1 | 100 | 100 | 1.6 | 0% | 30.00 CU | 0.00 CU | 0.00 CU | 1,120.00 CU | 1,150.00 CU |
| P04-B | 9.6 CU | 1 | 1 | 100 | 100 | 1 | 0% | 40.00 CU | 0.00 CU | 0.00 CU | 960.00 CU | 1,000.00 CU |
| P04-C | 8 QCU-B | 1 | 1 | 100 | 100 | 1.25 | 0% | 50.00 CU | 0.00 CU | 0.00 CU | 1,000.00 CU | 1,050.00 CU |
| P05-A | 7 CU | 1 | 1 | 100 | 100 | 1 | 0% | 0.00 CU | 0.00 CU | 0.00 CU | 700.00 CU | 700.00 CU |
| P05-B | 8 CU | 1 | 1 | 100 | 100 | 1 | 0% | 0.00 CU | 0.00 CU | 0.00 CU | 800.00 CU | 800.00 CU |
| P05-C | 9.5 CU | 1 | 1 | 100 | 100 | 1 | 0% | 0.00 CU | 0.00 CU | 0.00 CU | 950.00 CU | 950.00 CU |
| P06-A | 10 CU | 1 | 1 | 100 | 100 | 1 | 30% | 20.00 CU | 0.00 CU | 0.00 CU | 700.00 CU | 720.00 CU |
| P06-B | 9.3 CU | 1 | 1 | 100 | 100 | 1 | 5% | 25.00 CU | 0.00 CU | 0.00 CU | 883.50 CU | 908.50 CU |
| P06-C | 9.2 CU | 1 | 1 | 100 | 100 | 1 | 0% | 30.00 CU | 0.00 CU | 0.00 CU | 920.00 CU | 950.00 CU |
| P07-A | 7 CU | 1 | 1 | 100 | 100 | 1 | 0% | N/A | 0.00 CU | 0.00 CU | N/A | N/A |
| P07-B | 8.8 CU | 1 | 1 | 100 | 100 | 1 | 0% | 150.00 CU | 0.00 CU | 0.00 CU | 880.00 CU | 1,030.00 CU |
| P07-C | 9.5 CU | 1 | 1 | 100 | 100 | 1 | 0% | 0.00 CU | 0.00 CU | 0.00 CU | 950.00 CU | 950.00 CU |
| P08-A | 7 CU | 1 | 1 | 100 | 100 | 1 | 0% | 0.00 CU | 0.00 CU | 0.00 CU | 700.00 CU | 700.00 CU |
| P08-B | 8 CU | 1 | 1 | 100 | 100 | 1 | 0% | 0.00 CU | 0.00 CU | 0.00 CU | 800.00 CU | 800.00 CU |
| P08-C | 9 CU | 1 | 1 | 100 | 100 | 1 | 0% | 0.00 CU | 0.00 CU | 0.00 CU | 900.00 CU | 900.00 CU |
| Offer | Supplier approved | Spec matches | Valid until | Lead days | Payment days | Revised terms authorized | Excess units | Costs complete | Eligible | Failed checks |
|---|---|---|---|---|---|---|---|---|---|---|
| P01-A | Yes | Yes | 2026-11-05 | 7 | 30 | Yes | 0 | Yes | Yes | All declared checks pass |
| P01-B | Yes | Yes | 2026-11-05 | 7 | 30 | Yes | 0 | Yes | Yes | All declared checks pass |
| P01-C | Yes | Yes | 2026-11-05 | 7 | 30 | Yes | 0 | Yes | Yes | All declared checks pass |
| P02-A | Yes | Yes | 2026-11-05 | 7 | 30 | Yes | 0 | Yes | Yes | All declared checks pass |
| P02-B | Yes | Yes | 2026-11-05 | 7 | 30 | Yes | 0 | Yes | Yes | All declared checks pass |
| P02-C | Yes | Yes | 2026-11-05 | 7 | 30 | Yes | 0 | Yes | Yes | All declared checks pass |
| P03-A | Yes | Yes | 2026-11-05 | 7 | 30 | Yes | 100 | Yes | No | Overbuy limit exceeded |
| P03-B | Yes | Yes | 2026-11-05 | 7 | 30 | Yes | 0 | Yes | Yes | All declared checks pass |
| P03-C | Yes | Yes | 2026-11-05 | 7 | 30 | Yes | 0 | Yes | Yes | All declared checks pass |
| P04-A | Yes | Yes | 2026-11-05 | 7 | 30 | Yes | 0 | Yes | Yes | All declared checks pass |
| P04-B | Yes | Yes | 2026-11-05 | 7 | 30 | Yes | 0 | Yes | Yes | All declared checks pass |
| P04-C | Yes | Yes | 2026-11-05 | 7 | 30 | Yes | 0 | Yes | Yes | All declared checks pass |
| P05-A | No | No | 2026-11-05 | 7 | 30 | Yes | 0 | Yes | No | Supplier not approved; Specification mismatch |
| P05-B | Yes | Yes | 2026-10-20 | 20 | 30 | Yes | 0 | Yes | No | Quotation expired; Delivery deadline exceeded |
| P05-C | Yes | Yes | 2026-11-05 | 7 | 30 | Yes | 0 | Yes | Yes | All declared checks pass |
| P06-A | Yes | Yes | 2026-11-05 | 7 | 0 | No | 0 | Yes | No | Payment terms outside authority; Revised terms not authorized |
| P06-B | Yes | Yes | 2026-11-05 | 7 | 30 | Yes | 0 | Yes | Yes | All declared checks pass |
| P06-C | Yes | Yes | 2026-11-05 | 7 | 30 | Yes | 0 | Yes | Yes | All declared checks pass |
| P07-A | Yes | Yes | 2026-11-05 | 7 | 30 | Yes | 0 | No | No | Cost incomplete |
| P07-B | Yes | Yes | 2026-11-05 | 7 | 30 | Yes | 0 | Yes | Yes | All declared checks pass |
| P07-C | Yes | Yes | 2026-11-05 | 7 | 30 | Yes | 0 | Yes | Yes | All declared checks pass |
| P08-A | No | Yes | 2026-11-05 | 7 | 30 | Yes | 0 | Yes | No | Supplier not approved |
| P08-B | Yes | No | 2026-11-05 | 7 | 30 | Yes | 0 | Yes | No | Specification mismatch |
| P08-C | Yes | Yes | 2026-10-20 | 30 | 30 | Yes | 0 | Yes | No | Quotation expired; Delivery deadline exceeded |
For P06, the authored baseline is 1,100 CU. Subtracting the selected 908.50-CU purchase gives a 191.50-CU gross difference. After 30 CU of workflow cash cost, conditional cash-net savings are 161.50 CU. Valuing 1.5 owner hours at 40 CU/hour deducts a further 60 CU, leaving 101.50 CU after that noncash valuation.
Those are three different quantities. The owner-time deduction is not an invoice, avoided salary or observed saving. No baseline labor saving is assumed. The comparison also retains negative results: a workflow that matches the incumbent purchase price can still cost more once its own expenses are included.
“Conditional” matters because the inputs are authored, the baseline is assumed comparable and no purchase occurred. The model does not turn a negotiated quote into realised savings, or multiply a single opportunity into an annual claim. It also does not establish implementation payback.
Make a real procurement pilot measurable
Freeze the requirement and baseline before evaluating candidate quotes. Record quantity, specification, deadline, payment terms, included costs and the evidence supporting the baseline. Preserve rejected offers and reasons as well as the selected one; otherwise later reviewers cannot reconstruct the decision.
Track implementation and operating costs at an explicit level. A one-time integration expense is different from a per-order service charge. State the allocation period and expected volume before spreading fixed costs across purchases, then compare that assumption with actual usage. Record review and exception time separately from cash spending.
Start with recommendations that a named buyer can inspect. The package for approval should include the normalized quantity, complete cost, failed conditions, current quote version and the specific commitment requested. Restrict tools and accounts to the permitted actions. OWASP’s excessive-agency guidance supports minimal functions and permissions, human approval for consequential actions and authorization enforced downstream. [5]
After an approved award, reconcile delivery, invoice and payment. Track shortfalls, substitutions, additional charges and corrective work. Only then assess the realised difference against the comparable baseline. DN’s AI Agent ROI Index framework explains why operational improvement and financial benefit need separate evidence.
Frequently asked questions
Did DN test commercial procurement agents?
No. This release executes three deterministic rules on eight authored cases. It includes no language-model inference, commercial agent evaluation, live supplier discovery or real purchasing.
Why is the lowest complete cost sometimes rejected?
A known cost does not satisfy missing requirements. An offer can fail supplier approval, specifications, delivery, validity, overbuy or permitted terms even when its arithmetic is complete.
What happens when every quote fails?
The eligibility-first rule returns HOLD and leaves savings unavailable. The buyer needs a compliant alternative or a properly approved change to the requirement; the tool does not silently waive a condition.
Is the landed subtotal a full total-cost-of-ownership estimate?
No. It covers the declared purchase items and charges. Future maintenance, operating costs, disposal and residual value are outside this release.
Can extra inventory be counted as a saving?
The model assigns it no automatic credit. All required packs are paid for. Any future-use or resale value needs separate evidence and an explicit treatment of timing and associated costs.
Are the discounts evidence of agent negotiation performance?
No. They are supplied hypothetical terms. The test checks their arithmetic and authorization, not whether an agent could obtain them or whether a real supplier would agree.
Can the eight savings figures be added into an annual benefit?
No. The fixtures are separate illustrative purchases, not a measured programme or recurring demand sample. Annual benefit requires actual volume, comparable baselines, realised outcomes and operating costs.
Does passing the declared gates establish legal compliance?
No. The checks cover only the authored requirements. Applicable law, organizational policy, contractual obligations and supplier due diligence need their own assessment.
Method, version and limitations
Version 1.0 — 22 October 2026: initial release of eight curated RFQ fixtures, 24 fictional quotes, three deterministic rules and 24 resulting decisions. Independent recomputation covered quantities, purchase costs, eligibility, selected offers and savings availability. Reproducibility establishes consistency with the declared inputs, not their realism.
The benchmark does not test document extraction, adversarial supplier content, actual inventory quality, changing market conditions, integration reliability or a model’s ability to follow instructions. Several failures are deliberately present together. The cases cannot isolate a population-level effect or support a vendor ranking.
The procurement references are attributed New Zealand public-sector practice guidance, not universal legal requirements for every DN reader. No affiliate offer or paid product ranking appears here. The 2027 URL identifies the editorial series; the research and verification date is 22 October 2026 (SAST).
Send corrections through the DN contact page, identifying the fixture, offer, field or source with non-sensitive supporting evidence. Changes to inputs, rules or calculation methods should receive a new version and a recorded explanation.
Download the decisions and editable calculator
The CSV contains all 24 decisions. The research package contains the 24 quote inputs, normalization outputs, exact model source, methodology, independent review and standalone editable calculator.
Extract the ZIP and open dn-procurement-agent-savings-calculator-v1.0.html in a browser for a local single-quote comparison. The calculator does not send inputs or place orders. To reproduce the benchmark with Python 3’s standard library:
python3 build_model.py --output reproduced-resultsA separate assistant recomputed every quote and decision without importing the model builder. The fixed-fixture review passed. The package records calculator parity, input validation and reproduction checks separately; none is an external expert audit or evidence of real procurement performance.
Primary sources
- New Zealand Government Procurement: Total Cost of Ownership guide — lifecycle cost boundaries.
- New Zealand Government Procurement: Decide your evaluation criteria — preconditions and value evaluation.
- New Zealand Government Procurement: Conducting due diligence checks — corroborating supplier evidence.
- New Zealand Government Procurement: Negotiating with your preferred supplier — preparing and evaluating terms.
- OWASP: LLM06:2025 Excessive Agency — permissions and consequential actions.
