# Fragment

> Fragment (https://fragment.ai) builds AI agents that autonomously resolve procurement exceptions – invoice exceptions, three-way match failures, GL coding, change orders, and AP/P2P work – on top of a company's existing systems (SAP, Ariba, Coupa) with no rip and replace. Agents investigate each case against the company's own records (contracts, receipts, resolution history), clear what the evidence supports, and escalate judgment calls to people with the investigation already done.

## Product

- [Workflows](https://fragment.ai/workflows): The procurement and AP exception workflows Fragment's agents run end to end
- [Autonomous Context Engine](https://fragment.ai/autonomous-context-engine): The reliability layer that lets agents reason over enterprise systems and tribal knowledge
- [Request a demo](https://fragment.ai/demo): See Fragment's agents on your own exception queue
- [About](https://fragment.ai/about): Company background and principles

## Use Cases

> Examples of the source-to-pay workflows Fragment automates end to end (https://fragment.ai/use-cases), each with the manual process it replaces, how agents run it, impact figures, and an FAQ.

- [Use Cases hub](https://fragment.ai/use-cases): Example source-to-pay automation use cases across procurement and AP
- [Requisition-to-PO Conversion](https://fragment.ai/use-cases/requisition-to-po-conversion): Approved PRs converted into compliant POs automatically, with pricing validated before submission
- [Change Order Management](https://fragment.ai/use-cases/change-order-management): Agents read demand signals, update POs, confirm with suppliers, and post back to the ERP
- [RFQ Process Automation](https://fragment.ai/use-cases/rfq-process-automation): Quote packages assembled, responses chased, quotes normalized for side-by-side comparison
- [Supplier Master Data](https://fragment.ai/use-cases/supplier-master-data): Duplicates, stale details, and expired certifications fixed before they cause exceptions
- [Invoice Exception Resolution](https://fragment.ai/use-cases/invoice-exception-resolution): 3-way match exceptions triaged, investigated, and cleared end to end
- [Duplicate Invoice Detection](https://fragment.ai/use-cases/duplicate-invoice-detection): Fuzzy matching across ERPs catches what invoice-number checks miss
- [Credit & Rebate Management](https://fragment.ai/use-cases/credit-rebate-management): Credits matched before payment, rebate entitlements computed from contract terms
- [GL Coding & Cost Allocation](https://fragment.ai/use-cases/gl-coding-automation): 90%+ first-pass auto-coding with anomalies caught before posting
- [Maverick Spend Detection](https://fragment.ai/use-cases/maverick-spend-detection): Off-contract spend caught before payment, in real time
- [Tax Hold Resolution](https://fragment.ai/use-cases/tax-hold-resolution): Tax holds cleared by cross-referencing data you already have

## The Procurement & AP Wiki

> A reference library (https://fragment.ai/wiki) that defines source-to-pay terms for a world where AI agents do the operational work and people keep the judgment calls. Each article gives a standalone definition, a concrete worked example, and what changes when agents run the process.

### Invoice exceptions

- [Coding and approval exceptions: the judgment calls agents learn](https://fragment.ai/wiki/coding-and-approval-exceptions): The two exception families where the invoice is fine and the internal record failed.
- [The cost of invoice exceptions: a ledger agents change line by line](https://fragment.ai/wiki/cost-of-invoice-exceptions): One AP operation's exception costs priced line by line, and the lines agents move.
- [Duplicate invoices: how AI agents catch and clear them](https://fragment.ai/wiki/duplicate-invoices): How duplicates are born, why exact matching misses them, and what an agent verifies first.
- [Invoice exception rate benchmarks, and how agents change the math](https://fragment.ai/wiki/invoice-exception-rate-benchmarks): Verified benchmark figures, the counting rules behind them, and what agents change.
- [What is an invoice exception, when AI agents do the work?](https://fragment.ai/wiki/invoice-exceptions): What an invoice exception is, and how an AI agent investigates and clears one.
- [Manual vs automated exception resolution: analyst, rules engine, AI agent](https://fragment.ai/wiki/manual-vs-automated-exception-resolution): One shorted delivery cleared three ways: analyst, rules engine, and AI agent.
- [Non-PO invoices: how AI agents validate them without a match](https://fragment.ai/wiki/non-po-invoices): How agents validate invoices with no PO from contracts, approvals, and spend history.
- [Price variance exceptions: how an AI agent clears them](https://fragment.ai/wiki/price-variance-exceptions): The six causes an agent tests in order, and the credit request when overbilling is real.
- [Quantity and receipt mismatches: what an agent checks first](https://fragment.ai/wiki/quantity-and-receipt-mismatches): Why the goods receipt is the weakest match document, and the order an agent checks it.
- [Tax and freight discrepancies: how AI agents resolve them](https://fragment.ai/wiki/tax-and-freight-discrepancies): When invoiced tax or freight disagrees with what the company's systems computed.
- [Types of invoice exceptions: what each one asks an agent to prove](https://fragment.ai/wiki/types-of-invoice-exceptions): Six proof families, where each answer lives, and how an agent clears them.
- [Why invoice exceptions happen: the upstream causes agents trace](https://fragment.ai/wiki/why-invoice-exceptions-happen): Five upstream causes of invoice exceptions, traced along one purchase order's life.

### Three-way matching

- [How three-way matching works, and where agents pick up](https://fragment.ai/wiki/how-three-way-matching-works): The match pipeline in order – capture, pairing, tolerance checks, and what a hold means.
- [Match tolerances and thresholds: the dials agents let you retune](https://fragment.ai/wiki/match-tolerances-and-thresholds): Every tolerance dial trades risk for labor; cheap agent investigation moves the settings.
- [What is three-way matching, when AI agents run the match?](https://fragment.ai/wiki/three-way-matching): The PO-receipt-invoice control: what a pass proves, what a fail hides, where agents work.
- [Two-way vs three-way vs four-way matching, when agents work the failures](https://fragment.ai/wiki/two-way-vs-three-way-vs-four-way-matching): What each match level proves, what it costs, and how to pick the level per category.
- [Where three-way matching breaks: five assumptions agents restore](https://fragment.ai/wiki/where-three-way-matching-breaks): The five unstated assumptions behind the match, and how agents work each break.

### Exception resolution

- [What is autonomous exception resolution?](https://fragment.ai/wiki/autonomous-exception-resolution): A strict definition, plus a four-rung autonomy ladder for placing any vendor claim.
- [The context problem: the wall agents cross in exception resolution](https://fragment.ai/wiki/context-problem-exception-resolution): Six facts, four systems, two written down nowhere: the wall every automation wave hit.
- [Exception aging: what a hold costs before an agent picks it up](https://fragment.ai/wiki/exception-aging): A hold is free until day 10; then discount, late-fee, and write-off thresholds switch on.
- [Exception escalation: what should reach people when agents run the queue](https://fragment.ai/wiki/exception-escalation-best-practices): Four criteria for what escalates, and the anatomy of a handoff a reviewer can act on.
- [Exception routing: why holds land on the wrong desk, and what agents change](https://fragment.ai/wiki/exception-routing): Routing tables bet on the hold code; agents diagnose first and route only the residue.

### Economics of exceptions

- [Outsourcing invoice exceptions: what moves offshore, what agents change](https://fragment.ai/wiki/outsourcing-invoice-exceptions): Outsourcing moves labor and procedures; the context that clears hard cases stays home.
- [What a point of touchless rate is worth, and how agents move it](https://fragment.ai/wiki/touchless-rate-value): One worked model prices a point of touchless rate and why agents move the hard points.
- [Why exceptions never go to zero, even with agents on the queue](https://fragment.ai/wiki/why-exceptions-never-go-to-zero): Four structural reasons zero is unreachable, and the two numbers to manage instead.

### GL coding

- [Automated GL coding: rules, models, and agents compared](https://fragment.ai/wiki/automated-gl-coding): Rules decay, models need reviewers, agents post from precedent: an honest comparison.
- [GL coding for invoices: where agents find each line's code](https://fragment.ai/wiki/gl-coding-for-invoices): How PO inheritance, splits, freight, credits, and accruals decide each line's code.
- [GL coding in complex operations: the dialects agents learn](https://fragment.ai/wiki/gl-coding-in-complex-operations): Scale gives one company many coding conventions; agents learn them per entity.
- [What is GL coding, when AI agents keep the books?](https://fragment.ai/wiki/gl-coding): Translating purchases into ledger language, and why the hard cases run on convention.

### Procure-to-pay

- [P2P bottlenecks: where the process queues, and what agents drain](https://fragment.ai/wiki/p2p-bottlenecks): Every P2P queue mapped stage by stage, with a verdict on which ones agents can drain.
- [The P2P process end to end: one order, every record agents keep true](https://fragment.ai/wiki/p2p-process-end-to-end): One order walked from requisition to remittance, naming every record the chain creates.
- [P2P vs S2P vs source-to-settle: the scopes agents work across](https://fragment.ai/wiki/p2p-vs-s2p-vs-source-to-settle): What each scope covers, and why the boundary you pick decides what you can fix.
- [What is procure-to-pay (P2P), when AI agents run the process?](https://fragment.ai/wiki/procure-to-pay): The requisition-to-payment cycle, read as a chain of records agents keep true.
- [Touchless invoice processing: the pipeline, and where agents extend it](https://fragment.ai/wiki/touchless-invoice-processing): The pipeline an invoice must survive to stay untouched, and where agents extend it.

### AP automation

- [AP automation ROI: how to measure it honestly, agents included](https://fragment.ai/wiki/ap-automation-roi): Baseline first, verify every gain, and avoid the four double-counting traps.
- [The AP automation stack: four layers, and the one agents add](https://fragment.ai/wiki/ap-automation-stack): Anatomy of the four AP automation layers, and the resolution layer agents add on top.
- [What is AP automation, when agents do the resolving?](https://fragment.ai/wiki/ap-automation): The history of AP automation, the residue every generation left, and what agents change.
- [OCR and invoice capture: what it solves, and what agents assume](https://fragment.ai/wiki/ocr-invoice-capture): Where invoice capture fails, and why agents treat extracted data as a hypothesis
- [Where AP automation stalls, and what an agent restarts](https://fragment.ai/wiki/where-ap-automation-stalls): The four stall points every AP automation program hits, and which ones an agent restarts.

### Change orders & PO amendments

- [Automating change orders: how agents run signal to confirmed PO](https://fragment.ai/wiki/automating-change-orders): Why automating change orders means covering all five stages, from signal to record sync.
- [Why change order SLAs mislead, and what agents measure instead](https://fragment.ai/wiki/change-order-slas): SLA clocks start at record creation; agents measure reality-to-record lag instead.
- [Why change orders overwhelm shared services, and what agents absorb](https://fragment.ai/wiki/change-orders-overwhelm-shared-services): The workload physics that break shared services teams, and the layer agents absorb.
- [PO amendments: how orders drift from reality, and how agents catch it](https://fragment.ai/wiki/po-amendments-drift-from-reality): The four channels POs drift through, how to measure your drift, and how agents catch it.
- [What is a change order in procurement, when agents do the work?](https://fragment.ai/wiki/what-is-a-change-order): Most purchase changes never become change orders; agents close the event-to-record gap.

### Procurement automation

- [AI in procurement: what actually works today](https://fragment.ai/wiki/ai-in-procurement): A maturity survey of AI in procurement: production, supervised, and oversold tiers.
- [The business case for procurement automation, built to survive finance](https://fragment.ai/wiki/procurement-automation-business-case): Build a case a CFO can probe: baseline, verifiable benefits, honest costs, evidence gates.
- [What is procurement automation, when agents join the team?](https://fragment.ai/wiki/procurement-automation): Automation creates records well and resolves them badly; agents change the second half.
- [RPA vs agentic AI in procurement: what each can carry](https://fragment.ai/wiki/rpa-vs-agentic-ai-in-procurement): Does the task require new reading each time? The test that splits scripts from agents.
- [Which procurement processes to automate first, now that agents exist](https://fragment.ai/wiki/which-procurement-processes-to-automate-first): Rank by volume, cost per touch, and context; agents put exception resolution first.

### Source-to-pay

- [Exceptions across source-to-pay: the full map agents work](https://fragment.ai/wiki/exceptions-across-source-to-pay): The stage-by-stage map of every source-to-pay exception family and what resolves each one.
- [The PO-first culture problem: why buyers work around the process, and how agents fix it](https://fragment.ai/wiki/po-first-culture-problem): Workarounds are rational when the PO path is slow; agents make compliance the fast path.
- [Requisition-to-PO conversion: the compliance step agents make fast](https://fragment.ai/wiki/requisition-to-po-conversion): The anatomy of the conversion pipeline, why latency sets compliance, what agents run.
- [What is source-to-pay (S2P), when AI agents work the process?](https://fragment.ai/wiki/source-to-pay): S2P as two layers with a seam, and how agents keep negotiated value from leaking
- [Where source-to-pay automation breaks down, and what agents bridge](https://fragment.ai/wiki/where-source-to-pay-automation-breaks-down): S2P breaks at the seams between modules; agents read both sides and bridge them.

### The CPO agenda

- [Agentic AI in procurement: what changes for the CPO](https://fragment.ai/wiki/agentic-ai-in-procurement-what-changes-for-the-cpo): A working agenda for the CPO: operating model, metrics, governance, and year one.
- [How to evaluate agentic AI vendors for procurement](https://fragment.ai/wiki/how-to-evaluate-agentic-ai-vendors-for-procurement): The five-check test protocol that separates agentic substance from agentic labeling.
- [Procurement digital transformation: where programs stall, and what agents change](https://fragment.ai/wiki/procurement-digital-transformation-where-programs-stall): Four stall points of multi-year transformation programs, and what avoiding them takes.
- [Procurement KPIs the board actually reads, in the agent era](https://fragment.ai/wiki/procurement-kpis-the-board-actually-reads): The four questions boards ask procurement, and the KPI that answers each credibly.
- [What is autonomous procurement?](https://fragment.ai/wiki/what-is-autonomous-procurement): Autonomy is a per-process claim: the honest S2P matrix and how a CPO verifies it.