> ## Documentation Index
> Fetch the complete documentation index at: https://docs.neoflo.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# Human-in-the-Loop: Expert Oversight for Every Exception
> Neoflo combines AI automation with expert human review so edge cases and exceptions are always handled correctly — not just the easy ones.
AI alone breaks on edge cases. Unusual document formats, ambiguous data, policy exceptions, high-value transactions that need a second opinion — these are the scenarios where automation-only platforms fail and where real operational risk accumulates. Neoflo is built on a different model: AI handles the volume, and a dedicated team of human specialists handles everything the AI can't resolve with confidence. This isn't a fallback — it's a deliberate design choice that makes Neoflo's SLA guarantees possible. Every task that enters the platform gets completed, no matter how complex.
## The AI and Human Collaboration Model
Neoflo's human-in-the-loop model is not a queue of last resort. It's a structured collaboration where AI and human specialists each handle what they do best. The [AI processes the high-volume, rules-based work](/platform/ai-workflows) — typically 95% or more of all tasks — at machine speed. Human specialists take over the moment a task exceeds the AI's confidence threshold or triggers an exception rule.
The human team doesn't just fix errors. They own the outcome. When a Neoflo specialist resolves an exception, they're accountable for the result — not just for reviewing a suggestion. That accountability is what underpins Neoflo's SLA commitments and financial guarantee.
## When Humans Get Involved
Neoflo routes tasks to human review based on configurable confidence thresholds and exception rules. Common triggers include:
* **Unusual document formats** — vendor invoices that don't match any known template and fall below the AI's confidence threshold for reliable extraction
* **Data discrepancies** — amounts, quantities, or dates that don't match the corresponding PO or contract terms
* **Policy exceptions** — transactions that exceed approval thresholds or require sign-off under your internal controls
* **High-value transactions** — invoices or payments above a dollar threshold you define, routed for mandatory human review regardless of AI confidence
* **Ambiguous or missing data** — fields the AI could not extract reliably, or documents with conflicting information
* **Duplicate detection** — potential duplicate invoices flagged for human confirmation before rejection
You control which conditions trigger human review. Thresholds and routing rules are configured during your [pilot setup](/guides/pilot-design) and can be adjusted at any time.
## The Escalation Flow
The AI identifies a task it cannot process with sufficient confidence or that violates a business rule. It flags the exception, attaches the extracted data and the reason for escalation, and removes the task from the auto-processing queue.
The exception is assigned to a Neoflo specialist with the relevant domain expertise — AP exceptions go to AP specialists, support escalations go to support specialists. Routing happens automatically based on task type and urgency.
The specialist reviews the original document, the AI's extraction, the flagged discrepancy, and any relevant history (prior invoices from the same vendor, contract terms, previous exceptions). All context is surfaced in a single view.
The specialist resolves the exception — correcting the data, contacting the vendor, applying a policy override, or rejecting the document — and completes the task. Every action is logged with a timestamp and reason code.
The resolved task is processed and delivered to your system of record (ERP, CRM, or other destination) within the SLA window, regardless of how complex the exception was.
The specialist's resolution is fed back into Neoflo's AI model as a labeled training example. The model updates its understanding of how to handle similar cases, reducing the likelihood of the same exception recurring.
## Types of Exceptions Neoflo Handles
Price mismatches, quantity differences, missing PO references, incorrect tax amounts, and line-item disputes — Neoflo specialists investigate and resolve with the vendor or internal stakeholders as needed.
Transactions that fall outside standard approval workflows, require policy interpretation, or involve unusual circumstances that can't be handled by a binary rule. Specialists apply judgment aligned with your policies.
Multi-line, multi-currency, or multi-entity reconciliations where automated matching breaks down. Specialists trace discrepancies across systems and resolve them to a confirmed match or documented exception.
Support tickets and customer issues that require empathy, context, or authority beyond what an AI should handle. Neoflo support specialists resolve escalations within SLA and log outcomes for CSAT tracking.
## How Human Review Improves Accuracy Over Time
Every exception resolved by a specialist becomes a data point for Neoflo's self-improving AI. Over time, the model learns your vendor base, your policy interpretations, and your tolerance for edge cases. Exceptions that required human review in month one are handled automatically by month three.
This is why Neoflo's accuracy improves without any effort on your part — the human-in-the-loop isn't a workaround for AI limitations, it's the mechanism that continuously raises the AI's capabilities for your specific environment.
Neoflo guarantees 100% task completion — every item that enters the platform gets resolved. Nothing sits unprocessed, and nothing falls through the cracks. If the AI can't handle it, a specialist will.