Procurement AI agent · Should-cost
Should-Cost and Benchmarking AI Agent
Cost a part up from its material grade, process route, labour basis, yield and overhead treatment, carry the date of each input, and show a named analyst where a quote sits against the build.
- Typical delivery
- 4–6 weeks
- Deployment
- Your stack
- Cost analyst
- Build visible
- After launch
- Agent Care
What this agent does
Builds the estimate, never sets the price
In
- 01
A build is raised from stated inputs, and each input is written down beside the figure it produced.
- 02
An estimate is finished, and it says what a part ought to cost, not what a supplier will accept.
Reason
- 03
An input ages past the term it is good for, and the estimate leaning on it is flagged for rebuild.
- 04
An assumption carries no source, and it is shown as unsourced rather than folded into an overhead line.
- 05
A quote arrives, and the distance to the build is reported with the part of the build it lands against.
Decide
- 06
A benchmark comes from another volume, specification or region, and that difference is named before use.
- 07
A build is amended, and the assumption that changed travels with the new figure rather than replacing it.
Out
- 08
A part will not model at all, and that is recorded as the finding rather than filled with a placeholder.
- 09
Execute write actions only inside the approval boundaries agreed during implementation.
Product statement
Building, dating and comparing belong to the agent. The target price, the reading of a gap and any claim that a quote is unreasonable belong to a named cost analyst.
Example workflow
One estimate, inputs to acceptance
- 1Inputs receivedIndex series, rate tables, drawings, bills of material or prior quotations
- 2Build assembledMaterial grades, the process route, the labour basis, the yield assumption and the overhead treatment
- 3Estimate raisedThe build, the date each input was read and the confidence
- 4Controls appliedInput-age checks, assumption-source checks, arithmetic checks and comparability checks
No human action required
Stages 1 to 4 run unaided, and no estimate leaves the file at any of them — the agent is building, and the cost analyst lane opens at the confidence gate.
Goes to the named cost analyst to accept.
Adds a senior analyst read first.
Cost analyst review
The estimate is held with its build, the date of each input and the confidence.
Accept · Rebuild · Send to senior review
Accepted — by the named cost analyst
- 6Estimating and category records updatedOnly where write access and records policy allow it
- 7Outcome evaluatedInput currency, assumption sourcing, analyst rebuilds and what review found
Corrections
Each rebuild made in review is counted in the evaluation.
What should not run autonomously
Human approval stays in control
Outside the boundary — human approval required
8 items
- Setting the target price a negotiation opens at.
- Telling a supplier what its own costs are.
- Deciding that a quoted price is unreasonable.
- Publishing an estimate as a market benchmark.
Automation boundary
Agent acts unaided
- ✓Build the estimate up from the inputs as stated.
- ✓Show each assumption in the build beside the source that supplied it.
- ✓Carry the date each input was read alongside the estimate itself.
- ✓Show which part of the build the gap lands against.
Nothing leaves as an estimate except by a named cost analyst, inside agreed boundaries.
- Judging whether an estimate is fit to be used at all.
- Standing behind an estimate in front of a board.
- Choosing the margin a build is drawn at.
- Changes to the model, the inputs or the assumptions.
Example output
One should-cost estimate, annotated
This serves a procurement team who may have to defend a build to the supplier it was drawn against; below is one estimate exactly as the agent leaves it.
Should-cost build · single part
Illustrative example- Part as specified
- Machined aluminium bracket
- Process route
- Milled from bar, two operations
- Material basis
- Wrought grade, index-priced
- Estimate raised
- Index prices read 14 July 2026
- Confidence
- Held unaccepted
- Held for
- The cost analyst, by name
As received
Taken from the drawing and the index series as they stand — nothing on this side is rewritten by the agent.
Evidence used
- Drawing and grade
- Index series, dated
- Rate and yield tables
Why this is not a price
An estimate says what a thing ought to cost, not what it will fetch.
What the score decides
Below the configured threshold an estimate gets a senior read before the analyst sees it.
Action
AcceptRebuildSend to senior reviewValue
Where AI adds value
The same four claims, placed at the point in the workflow where each one applies.
Where the value lands
Value 01 – 04
Every estimate
From the inputs that built it
Where the estimate is used
The autonomous negotiation agent is what happens once a number goes into a negotiation; this page stops at the estimate, which is not a position anyone has taken.
An estimate is not a price
A build says what a part ought to cost given inputs, process and a normal margin. It does not say what any supplier will accept.
What was checked, and not found
No index series consulted prices the exact grade, form and volume this part is bought at, and nothing here rests on the cost principles that bind federal contracting, so the nearest series is reported with its distance.
Build an evidence trail
The estimate, the build behind it and the analyst who accepted it stay together.
Integrations
Typical integrations
Five system groups connect to the same agent. Which of them are in scope is decided in discovery.
Index and commodity data
- Metals
- resins
- energy series
- Grade and form index movements
Process and labour data
- Cycle times
- labour rate tables
- Machine and labour bases
Drawings and specifications
- PLM
- CAD records
- bills of material
- Grades, tolerances and volumes
Agent
Should-cost build and benchmarking
- Reads the inputs
- Builds the estimate
- Holds for the analyst
Quotes and purchase history
- Quotation history
- past orders
- Quoted prices and what they include
Observability & evaluation
- OpenTelemetry
- Langfuse
- Supported monitoring/evaluation sources
Integration availability depends on the client's existing systems and API access.
Agent controls
Six rulers between the model and the analyst
Six rulers laid end to end, the last the truest. What measures up is set out in the map below.
L6 · Outermost — last line of defence
Inward → L1 · closest to the model
- L1 – L2
- Test whether an estimate may stand
- L3
- Leaves the acceptance to a named analyst
- L4 – L5
- Keep the estimate and the build behind it
- L6
- Publishes no estimate at all when signals degrade
How Nestack evaluates it
Evaluate the whole build — not only the estimate that comes out.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the estimate a buyer carries
- E1Final-output evaluationDid the estimate record the inputs it was actually built on?
- E2Step-level evaluationDid the agent read the right series, the right period and the live rate tables?
- E3Tool evaluationDid it read and write the correct part and the correct estimate?
- E4Confidence calibrationDo low-confidence estimates actually attract more analyst rebuilds?
- E5Slice evaluationHow does performance change across specific cost drivers?
- E6Business outcomeHow many estimates needed a rebuild before the analyst accepted?
Floor — the inputs an estimate rests on
Failure modes
Where each failure originates in the agent
Seven failure modes, each set at the stage that first exposes it.
Agent lifecycle
Direction of processing →
01 · Retrieval
1 mode
- OU-03
Stale index series read
The series read is not the one now published.
Stage gathers
The series, the rates, the drawings and dates
02 · Reasoning
2 modes
- OU-04
Gap attributed to margin
A specification difference is booked as margin.
- OU-06
Retired rate table read as live
A superseded rate is worked as the current one.
Stage proposes
The build, the inputs and the dates they carry
03 · Tool / write
2 modes
- OU-02
Unsourced assumption passed on
A yield assumption moves with no source behind it.
- OU-05
Benchmark from another volume
A price from a different volume is compared.
Stage writes
Only where write access and approval policy allow it
04 · Output
1 mode
- OU-01
Estimate read as a price
A build leaves the file and is treated as a quote.
Stage returns
The estimate a buyer carries into a review
05 · Change / Version
1 mode
- OU-07
Silent assumption change
A yield assumption moves while the stored estimate keeps the old one.
Stage tracks
Model, prompt, build rules and input dates
- Sev-1 · an estimate quoted as a price
- Sev-2 · an aged input reaches a build
- Sev-3 · source degrades, estimate held
Affected slices
Volatile inputs absorb the rebuilds
A driver-level input figure can read clean while the most volatile commodity inputs are carrying nearly all of the rebuilding. Nestack reports the rebuild rate by slice, not only in total.
Slice performance — reported separately, not only in aggregate
Illustrative example| Slice | Failure rate | Lift | Lift vs. threshold | Status |
|---|---|---|---|---|
| Volatile commodity inputs | 7.5% | 3.7× | Review | |
| Low-volume machined parts | 5.4% | 2.7× | Review | |
| Multi-region labour bases | 3.3% | 1.6× | Watch | |
| Stable catalogue components | 1.8% | 0.9× | Normal |
Bar: rebuild-rate lift vs. stable-component baseline · scale 0–4.0× · tick marks the 2.0× review threshold
2 of 4 slices over threshold
Evidence-linked improvement
What an estimate taken as a price costs
A cycle shuts when the estimate quoted as a price is a regression case. That suite is what the next build issued is measured against.
Improvement cycle · five stages
Switchback — the path turns at Improve and returns at Learn
- 01
Detect
Rebuild rate rises on volatile commodity inputs.
- 02
Diagnose
The should-cost figure that walked into a negotiation as though it were a quote is worked backwards until one cause is left standing.
- 03
Improve
Changes ship numbered, with the estimates that drove them filed underneath.
- 04
Verify
Nothing releases while a single estimate case is still red.
- 05
Learn
The case stays on, and the build rules are amended in the same commit.
Learn → Detect
The return edge. The next estimate is measured against a suite one case longer.
Typical build scope
Twelve workstreams across six weeks
The build scope read against the delivery timeline. Week structure follows the six-week plan — discovery, sources, estimate logic, evaluation, integration, then production validation and handover.
Workstream
Week 1Week 2Week 3Week 4Week 5Week 6
- 01Cost-model discovery and automation-boundary design.Week 1
- 02Index, rate and quotation source review.Week 1, Week 2
- 03Input-to-estimate and benchmark-comparability mapping.Week 1, Week 2
- 04Input ingestion and normalisation.Week 2
- 05Build logic and assumption binding.Week 2, Week 3
- 06Confidence scoring and review routing.Week 3
- 07Cost analyst review workflow.Week 3, Week 4
- 08Index-and-quote integration.Week 5
- 09Build and input cases.Week 4, Week 5
- 10Guardrails and publication controls.Week 4
- 11Estimate-trail instrumentation.Week 5, Week 6
- 12Deployment, documentation and Agent Care handover.Week 6
12 workstreams · 6 weeks · bar shows the weeks a workstream is active — several run in parallel
Final scope and sequence confirmed in discovery
Engagement tiers
What each tier includes
Rows are the capabilities named in each tier's scope. Higher tiers include everything below them.
| Capability✓ in scope · — not at this tier | PilotOne category, one cycle | ProductionProduction estimating workflow | AdvancedMultiple categories / regions |
|---|---|---|---|
| Introduced at Pilot | |||
| Estimate builds to your cost model | ✓ | ✓ | ✓ |
| Named cost analyst acceptance | ✓ | ✓ | ✓ |
| Cost-model baseline | ✓ | ✓ | ✓ |
| Introduced at Production | |||
| Reporting by cost driver | — | ✓ | ✓ |
| Senior review workflow in your systems | — | ✓ | ✓ |
| Approved write-back | — | ✓ | ✓ |
| Index-and-quote integration | — | ✓ | ✓ |
| Introduced at Advanced | |||
| Cross-region input libraries | — | — | ✓ |
| Multi-part estimate packs | — | — | ✓ |
| Large quotation histories | — | — | ✓ |
| Multi-region input controls | — | — | ✓ |
| Build price | From $5,000 | From $8,000 | Custom quote |
- Final build price
- Confirmed after discovery based on integrations, workflow complexity, input volume, approval controls and deployment requirements.
- Separate from build
- Build pricing is separate from recurring Agent Care, which covers managed monitoring, evaluations, incidents and verified improvements after launch.
What we need from you
What you bring, and what we build with it
Each input maps to a piece of build scope and a week in the delivery timeline.
You bring
We build with it
01Your live cost model and the drivers it names
→Cost-model capture and assumption versioning
Week 102Representative drawings, rate tables and quotations
→Source binding, build logic and the estimate baseline
Week 203Your estimating calendar and the analysts it names
→Cost-model mapping, input binding and the automation boundary
Week 104Access to relevant APIs, feeds or exports
→Index, rate and quotation source assessment, then integration setup
Week 205Estimates you would not want quoted
→Input cases and failure-mode testing
Week 406What no should-cost figure may fix
→Confidence scoring, review routing, guardrails and release controls
Week 307A named cost analyst who accepts the estimate
→Release to the cost analyst, then pilot and production validation
Weeks 5–6
Nothing else is required
Deployment, documentation and Agent Care handover are ours.
Delivery timeline
Four phases across six weeks
Where these bands overlap the phases overlap; the fifth week is a measurement, not a compromise.
Phase
W1
W2
W3
W4
W5
W6
- DiscoveryW1
- BuildW2 – W3
- EvaluateW4 – W5
- Pilot & LaunchW5 – W6
Week focus
W1Cost-model discovery, driver mapping and the automation boundary
W2Index, rate and quotation integration and the estimate baseline
W3Build logic, confidence scoring and release controls
W4Evaluation suite, input cases and failure-mode testing
W5Estimating-system integration, pilot builds and targeted corrections
W6One estimating cycle run under the cost analyst, then Agent Care handover
- Reading the band
- Each bar covers only the weeks its own work is named for, and week five is shared by design.
- At the end of W6
- Once the build record validates, Agent Care takes the agent on.
- Duration
- Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.
Next step · Procurement AI agent
Build a should-cost agent around the estimate your last negotiation treated as a quote.
Show us one part you estimate every year and the build behind the last figure. Not a number to open at. A build a supplier can argue with, line by line, with a date on each input. If those inputs are older than the last quarter, the figure is not an estimate of anything.
- Typical delivery
- 4–6 weeks
- Deployment
- Your drawings and inputs
- Cost analyst
- Build visible
- After launch
- Agent Care available
Nestack Agents · Should-cost
AGT-PR-10 · Agent Care available after launch