How XR Extreme Reach Hit 80% Straight-Through Processing with 99.8% Accuracy and Zero Manual Touch-Ups

What does straight-through processing actually mean in practice?

Jace: Hyperbots claims to deliver straight through processing. What does that actually mean in practice?

John: Yeah, thanks for having me. It’s a great question. For every PO-based invoice, our copilots handle every single step - from email discovery, extraction, geocoding, and three-way match to ERP posting - without a handoff. It’s all the way through.

It’s not just about taking invoicing or doing one piece of the process; it’s true straight-through processing. We’re running about 80% STP today. Most vendors in our NetSuite instance never hit an AP inbox.

Competitive OCR tools top out at 40–50% because they stop after data capture and push the rest to humans. Their accuracy is poor, requiring manual review on nearly every invoice. We used to be at around 20% STP with prior tools, but Hyperbots has exceeded that by a huge margin.

How does Hyperbots discover and triage invoices in a messy shared inbox?

Jace: And how does Hyperbots discover and triage invoices in a messy shared inbox, for instance?

John: It’s designed for that. The AI agent goes through your full email - even spam, shipping docs, and noisy folders - and only pushes true invoices into the processing pipeline.

False positives are nearly zero because of that deep checking. Other solutions can’t weed out the noise and often have 20–50% false positives, which still end up clogging AP review queues.

How are duplicate invoices handled?

Jace: Duplicate invoices are a silent cash leak. How is that handled for you guys today?

John: The copilot checks every header and line field, cross-referencing with the PO and historical invoices. That blocks any duplicate freight bills that older software might have paid twice.

If an issue ever occurs - say a vendor resubmits something - Hyperbots catches it automatically, so duplicates never get through.

What extraction accuracy are you seeing?

Jace: Extraction accuracy is everyone’s buzzword these days. What numbers do you guys see?

John: Significant improvement. We’ve moved from around 20% to nearly 100% accuracy.

Hyperbots uses field-specific models trained on over 35 million invoice fields, which has been a massive step up. Typical OCR tools average only 50–60%.

How does Hyperbots handle validation and anomaly checks?

Jace: Validation often slows down auto posting. What’s different here?

John: The biggest difference is the agent itself. It runs mathematical checks, detects anomalies, and provides plain-English reasoning when something looks off.

Competitors either don’t do validation or just flag “Error 104” - leaving AP to puzzle it out.

What happens when data is missing?

Jace: If data is missing, invoices often get kicked out in other systems. What happens here?

John: Hyperbots doesn’t just kick it out - it augments missing data.

For example, if the “Ship-To” field is blank, it infers it from the PO or vendor defaults. Our straight-through rate actually rose from 65% to 80% after enabling this augmentation feature.

Can you tailor matching rules?

Jace: And can you guys tailor the matching rules?

John: Yes, absolutely. We can configure 3-way, 2-way, or even no-match rules.

We can also apply different tolerance thresholds per vendor or category - for instance, set direct materials at 0% variance but allow freight up to 2%.

Most competing tools are hardwired to global rules, but Hyperbots applies AI-driven tolerance for precision.

How many fields can be matched?

Jace: How many fields can actually be matched?

John: Up to 140 header and line fields - quantities, units, descriptions, even custom UDFs.

Legacy tools match maybe 4–5 numeric fields and stop there, so this level of granularity is rare.

What do business users see in notifications?

Jace: When it comes to notifications, what do your business users see today?

John: Real-time alerts. For instance, “Invoice 1458 auto-posted” or “Invoice 1460 has a price variance.”

It’s all real time and continuous - not a giant morning report. Teams used to spend 2–3 days a month combing through those reports; now they see everything as it happens.

How granular is the audit trail?

Jace: Auditability is huge. How granular is your trail now with Hyperbots?

John: Every AI or human action is timestamped, with old and new values hashed.

During our last audit, we exported a ledger in minutes. The auditors called it “the best evidence trail they’d ever seen.”

Because it’s all straight through, we don’t need to pull endless samples - it’s all traceable instantly.

How accurate is geocoding?

Jace: Geocoding accuracy often breaks straight-through processing. How does the recommender perform for you?

John: After two weeks of learning, we reached 99% accuracy on coding lines - far better than our manual 80%.

If AP overrides a code, the model learns instantly, so there’s no guessing. Competitors don’t even offer automated geocoding, so this alone is a major edge.

Are there any manual journal entry touch-ups?

Jace: And speaking of ERP posting, do you still have any manual journal entry touch-ups?

John: None. Hyperbots writes the voucher, reads it back, confirms entries, and posts automatically.

Old RPA bots often failed mid-process or left items unreconciled until month-end - that problem’s gone entirely.

How does it handle custom UDFs?

Jace: We use custom UDFs on invoices. Does that ever trip you up?

John: Not at all. The connector auto-discovers UDFs, extracts them, and maps them directly.

We recently added a “Project Phase” field - no code changes required. It was mapped instantly through the UI.

How fast do models start performing accurately?

Jace: How fast did the models start working accurately for you?

John: Day one. Hyperbots is pre-trained on millions of invoices, so accuracy was at 99% immediately.

Competing vendors usually take six months or more to train and build logic. Here, we went live with perfect extraction from day one - no historical uploads needed.

Can it handle large multi-page invoices?

Jace: If you had multi-page invoices - say, 100-line healthcare invoices - can the system handle those?

John: Absolutely. We process invoices with thousands of lines.

Hyperbots scrolls through every page, extracting and splitting lines correctly into GL codes. Legacy OCRs typically just total the page and skip the detail.

What happens when multiple invoices come in one PDF?

Jace: Sometimes vendors send multiple invoices in one PDF. What happens then?

John: The copilot automatically detects and splits them.

Even if it’s a combined statement, it identifies individual invoices, assigns document numbers, and processes them independently - no manual prep needed.

How is sales tax verified?

Jace: Do you verify sales tax codes on each line item?

John: Yes. The Sales Tax Copilot cross-checks origin, destination, and line descriptions with tax code dictionaries.

It’s helped us eliminate overpayments and automatically trigger vendor credits where applicable.

How do vendors track invoice status?

Jace: How do vendors track the status?

John: Vendors now use a self-service web portal to track invoices and payment status in real time.

That’s reduced inquiry emails by about 60% since they no longer need to chase statements or remittances.

How does Hyperbots handle multiple ERPs?

Jace: We run two ERPs. How does the copilot juggle that?

John: It can connect to multiple ERPs simultaneously.

For each invoice, it knows where to post - even across subsidiaries. This consolidated view was one of the main reasons we chose Hyperbots.

How does Hyperbots differ from competitors overall?

Jace: Bottom line - how does this differ from typical competition?

John: Competitors capture data. Hyperbots captures, reasons, and acts.

That’s how we reached 99.8% accuracy and 80% productivity gains. It executes the actions AP teams normally handle - validation, posting, duplication checks all in real time.

No duplicate paths, a single audit trail, and zero delays none of which were possible before.

Closing thoughts

Jace: Awesome. That’s all I have for you. Thanks again, John, for your time - we look forward to connecting again.

John: Sounds good. Appreciate it!