Line extraction
Were descriptions, quantities, units, rates, totals, notes, exclusions, and qualifications captured with correct source references?
AI Pilot Experience · Quote Comparison
Convert differently formatted subcontractor quotes into a source-linked comparison of inclusions, exclusions, quantities, units, qualifications, dates, and unresolved scope for authorized commercial review.
Pilot experience
Subcontractor quotes arrive as PDFs, spreadsheets, schedules, letters, and attachments using different descriptions, units, quantities, qualifications, inclusions, exclusions, and pricing structures. Direct total-price comparison can hide material scope differences.
The workflow identifies every quote and revision, extracts source-linked line items and commercial qualifications, maps them against an approved package structure, and exposes missing, excluded, provisional, differently measured, or unresolved scope.
An estimator or commercial reviewer opens each cited source, corrects mappings, records clarifications, and accepts the comparison table. The workflow stops before supplier selection, negotiation, approval, award, contractual interpretation, or commitment.
Visual walkthrough
Each view shows how source records, system outputs, exceptions, corrections, and human decisions remain connected during the pilot.
Source quotes remain connected to extracted lines, scope mappings, gaps, reviewer corrections, and the accepted comparison.
Each bidder column retains quote identity, revision, date, inclusion, exclusion, qualification, and source reference.
Different wording and units map into an approved package structure without erasing the original quote language.
Missing items, different units, exclusions, provisional values, date conflicts, and clarification questions receive owners and states.
The reviewed comparison supports a separate human commercial decision without making it automatically.
User journey
The interface can feel simple while the pilot preserves document identity, revision, permission, retrieval, uncertainty, and review evidence behind every result.
Record package ID, approved scope reference, project, estimator, bidders, purpose, confidentiality, deadline, and authority.
Fingerprint each quote and attachment with bidder, quote ID, revision, date, currency, validity, source, and status.
Apply OCR, layout, table, spreadsheet, and text processing to candidate descriptions, quantities, units, rates, totals, and qualifications.
Relate each candidate line to the package structure while preserving original wording, source coordinates, and mapping uncertainty.
Show inclusions, exclusions, missing items, provisional values, alternative units, quantity differences, dates, and qualifications.
An estimator or commercial reviewer corrects mappings, opens source evidence, records questions, and assigns unresolved scope.
Freeze the reviewed table with quote versions, corrections, open items, reviewer, timestamp, and permitted use.
Provide the comparison to separate clarification, negotiation, supplier selection, approval, and award processes under human authority.
Technology options
The final selection records exact versions, licences, data handling, cost, limitations, replacement options, and the evidence required before use.
| Component | Examples | Pilot role | Boundary |
|---|---|---|---|
| OCR and layout | Docling, approved OCR, table and document-layout services | Recover quote text, line-item tables, notes, totals, schedules, and qualification structure. | OCR and table extraction can merge rows, lose units, misread totals, or disconnect notes; the source quote remains controlling. |
| Claude Sonnet | Exact available Claude Sonnet model selected at pilot start | Prepare source-linked candidate quote lines, scope mappings, differences, qualifications, and clarification questions. | Model output does not select a supplier, interpret a contract, approve commercial terms, or authorize award. |
| GPT-5.6 tiers | Luna, Terra, or Sol with exact model ID and version recorded | Prepare source-linked candidate quote lines, scope mappings, differences, qualifications, and clarification questions. | Model output does not select a supplier, interpret a contract, approve commercial terms, or authorize award. |
| Qwen3.7-Plus | Exact Qwen3.7-Plus endpoint and version recorded | Prepare source-linked candidate quote lines, scope mappings, differences, qualifications, and clarification questions. | Model output does not select a supplier, interpret a contract, approve commercial terms, or authorize award. |
| Deterministic controls | Package registry, schema, date rules, required-field checks, source citations, permissions | Enforce identity, source, revision, field, citation, status, and output conditions outside the model. | Rules depend on correct source records, tested logic, exceptions, monitoring, and ownership. |
| Review interface | Client-owned comparison table, source viewer, scope matrix, and clarification queue | Show every compared field beside original quote evidence, mapping, correction, and unresolved state. | Accepting the table confirms comparison preparation only and does not approve a bidder or commercial decision. |
Client-owned pilot outputs
Evaluation evidence
A fluent answer is not the acceptance unit. The pilot tests whether a reviewer can reach the controlling evidence efficiently and detect important failure.
Were descriptions, quantities, units, rates, totals, notes, exclusions, and qualifications captured with correct source references?
Did the comparison expose covered, missing, excluded, provisional, alternative, duplicated, and unresolved package items?
Did differently worded quote lines map to the approved package structure without flattening meaningful differences?
Were currency, tax, validity, dates, units, alternates, qualifications, and arithmetic presented without unsupported normalization?
Measure time to inspect, correct, remap, clarify, reject, and accept the comparison.
Measure preparation, OCR, extraction, model use, storage, review, correction, monitoring, support, and reprocessing per package.
Cost drivers
Scaling factors
Use durable package, bidder, quote, attachment, line, mapping, clarification, and review IDs with queues and retries.
Reprocess affected bidder fields while preserving prior quote, mapping, correction, clarification, and decision history.
Version scope structures, descriptions, units, quantities, inclusions, exclusions, qualifications, and review roles by package.
Preserve opportunity, client, project, user, document, purpose, retention, and export permissions across every route.
Keep documents, schemas, citations, corrections, tests, and accepted records independent of any provider.
Require separate security, integration, monitoring, recovery, service, training, ownership, and cost approval.
Cross-industry reuse
The reusable capability is permission-aware multilingual retrieval across versioned documents with citations. Each industry keeps its own source authority, terminology, consequence, retention, and qualified review.
Client tender packages, addenda, specifications, drawings, schedules, return forms, and subcontractor quotes.
Review: Preconstruction, estimating, commercial, design, legal, and authorized bid leadership.
Trade invitations, scopes, drawings, specifications, bills, schedules, supplier and subcontractor quotations.
Review: Estimator, commercial lead, operations, procurement, engineering, and authorized management.
Consultant appointments, contractor tenders, work packages, development requirements, and commercial submissions.
Review: Development, project, design, procurement, cost, legal, and commercial authority.
Technical invitations, scopes, schedules, qualifications, deliverables, consultant and supplier proposals.
Review: Technical lead, bid manager, commercial, legal, quality, and authorized signatory.
Design invitations, briefs, deliverable schedules, consultant scopes, forms, and fee submissions.
Review: Practice leadership, bid lead, design, commercial, legal, and quality owner.
RFQs, technical specifications, drawings, bills, supplier quotations, delivery schedules, and qualifications.
Review: Procurement, engineering, production, quality, commercial, and legal owner.
Service tenders, maintenance scopes, asset schedules, compliance requirements, and vendor quotations.
Review: Property, facilities, procurement, compliance, finance, legal, and technical owner.
Service requests, asset lists, response requirements, rate schedules, and subcontractor or supplier quotes.
Review: Service manager, operations, procurement, commercial, safety, and finance owner.
Authority boundaries
Related paths
Continue into the operating, technical, security, and readiness guidance connected to this workflow.
Review pilot scope, evaluation, price range, duration, and possible decisions.
Open pageConnect approved requirements, quote review, clarifications, award controls, and learning.
Open pageReview provider-independent procurement planning, tender, award, and learning controls.
Open pagePrepare cases, expected outcomes, acceptance criteria, critical failures, and monitoring.
Open pageStart with one approved collection
Use general workflow information during the assessment. Do not submit confidential documents, passwords, API keys, authentication codes, or unrestricted system access.
Plain-language route guide
Choose the smallest route that answers your next decision.
The formal service names remain useful for scope and contracts. The plain-language labels explain what each route actually does. These are alternatives, not four mandatory stages.
Operating-Layer Blueprint
Buyer question answered
What should we build, and where should the boundary be?
Typical input
One priority workflow, a named owner, current systems, representative records or files, and known failure points.
Output
A client-owned current-state map, target design, source inventory, implementation boundary, timeline, and fixed quote.
Bounded AI Agent Pilot
Buyer question answered
Can one specific AI-assisted task work reliably enough to justify more?
Typical input
One named task, approved sources and tools, representative cases, a human reviewer, and explicit stop conditions.
Output
A working pilot, evaluation evidence, cost and failure findings, review requirements, and a proceed, revise, or stop recommendation.
Single Workflow Implementation
Buyer question answered
How do we put one recurring workflow into controlled production?
Typical input
A defined trigger and completion point, accountable owners, approximately three core systems, rules, approvals, and test cases.
Output
Connected records, an operating view, integrations, a dashboard, acceptance testing, training, documentation, and handover.
Complete Operating Layer Implementation
Buyer question answered
How do we connect shared data and decisions across teams?
Typical input
Two to five related workflows, shared records, several departments or systems, an executive sponsor, and named operating owners.
Output
A phased operating layer with shared records, permissions, interfaces, integrations, reporting, controlled automation, training, and handover.
Still unsure which route fits?
Describe one broken workflow. The free assessment may recommend a Blueprint, pilot, implementation, a smaller discovery step, or no engagement.