Four Layers That Turn a PDF Into an Answerable Question in Under 3 Seconds
Interface, Intelligence, Data, and Integration: Stralevo's Processing Pipeline
---
Your board asked which vendors raised prices more than 10% this quarter. Your finance team spent three hours in Excel, built a pivot table they'd rebuilt the quarter before, and delivered the answer on Thursday. Stralevo answered in 8 seconds — because all 847 invoices were already processed, every field extracted, every number indexed and waiting the moment the question arrived.
Those three hours did not disappear because the AI model was smarter. They disappeared because the work was done before anyone asked.
Simple warranty checks — one document, one field — come back in under 3 seconds. An 847-invoice supplier price comparison takes 8. Both are architectural facts, not marketing claims. Understanding why requires understanding what happens inside each layer.
---
Data Hidden in Plain Sight
Standard accounting software captures the regulatory minimum: date, amount, VAT number, vendor name, description. Three to five fields per invoice. A typical invoice contains 15 to 40 readable fields — serial numbers, warranty terms, delivery references, payment conditions, contact details, small-print clauses. All of them land in a PDF that gets filed and never opened again.
This was a deliberate architectural choice made in the early 2000s, when storage was expensive and processing was slow. Capturing 40 fields per document made no economic sense when the regulatory requirement was five. Accounting software was built to record transactions, not answer questions — and neither the technical nor the commercial incentive to change this has ever arrived for vendors whose business model centres on the ledger.
Ninety percent of the intelligence in every financial document your business has ever received is locked in a format nobody queries twice. When a board meeting surfaces a question about warranty terms on equipment purchased 18 months ago, the current process is: search email, find the PDF, locate the serial number, separately find warranty terms, call the vendor, wait on hold. Forty-five minutes — for a fact that was written down on the day the invoice arrived.
Accounting software reads the invoice. Stralevo reads what it means.
---
Four Layers, One Question
Questions arrive — typed, spoken, or entered via the mobile companion app. What happens in the next few seconds requires four distinct processing stages. Remove any one of them, and the answer either cannot be generated or requires hours of manual work.
Layer 1 — Interface (ContextUX™)
ContextUX™ receives the question and routes it to the right context automatically. A bookkeeper processing invoices sees invoice workflows. A CFO reviewing quarterly performance sees executive dashboards. An auditor checking compliance sees audit trails. Same system — the right interface appears for the person asking, without navigating menus or opening the wrong report.
Layer 2 — Intelligence (Orchestrator + Needle Finder + Smatched™)
Stralevo's Orchestrator routes the question to the right engine. Simple queries return against indexed data fast — this is where the sub-3-second answers live. Complex questions — which of the top 20 suppliers raised prices more than 10% across two quarters? — go to Needle Finder, the cross-document query engine built specifically for questions spanning hundreds of source files. A single Needle Finder query can cross-reference 847 invoices, 23 contracts, and 12 bank statements simultaneously. Not a benchmark figure — that is the actual document volume a mid-market French company accumulates in 18 months of normal operations.
Reconciliation queries go to Smatched™, the proprietary matching engine built for the scenarios that currently require a consultant and a full weekend: one payment applied across three invoices (1:N), one invoice matched against multiple partial payments (N:1), or many invoices reconciled against many payment batches at once (N:M, many-to-many). Smatched handles all of them automatically. Eighteen months of R&D. Patent pending.
For deep analytical queries — reconciling every supplier invoice against contracts across 18 months for all suppliers above 50,000 euros — response time extends beyond seconds. Minutes, not hours. And the answer returns sourced and verified, which no manual process can guarantee at that scale within that timeframe.
Layer 3 — Data (Knowledge Graph + Document Store)
Layer 3 holds the pre-work. SightCapture™ ICR — the document ingestion engine — processes 50-plus file formats continuously, capturing every readable field from every document as it arrives. Not five. All of them. The Knowledge Graph maintains persistent entity relationships across the full document history: it knows that ACME Corp, Acme Corporation, and ACME SARL are the same vendor across 200 invoices, without anyone manually deduplicating them. When a question reaches Layer 2, the answer is already indexed, connected, and waiting.
Between receiving a document and being able to query it: near zero. Every document that arrived this morning is already processed.
Layer 4 — Integration (Connected Systems)
We connect to Sage, Xero, Cegid, QuickBooks, and PennyLane — and natively to Liberté. Layer 4 pulls live figures from these systems to pair with extracted document data. A supplier price analysis queries both historical invoice data and current ledger entries in the same request. Every answer includes source citations: which document, which page, which field. Not a confidence score. A traceable source.
---
Breaking Down the Manual Alternative
Each layer has a manual equivalent. Interface: describe the question to a junior analyst, 5 minutes. Intelligence: identify which documents are relevant across two systems, 15 minutes. Data: extract the relevant fields manually from each PDF, 2 hours. Integration: pull live data from Sage and cross-reference against document history, 1 hour. Total: roughly 3.5 hours — for every question, every time.
Breaking down those three hours before Thursday's board meeting reveals that 20 minutes was actual analysis. The rest was extraction: skilled human effort applied to a task that a well-architected system does automatically and continuously in the background.
Most CFOs cannot answer this question without checking: how many data fields does your accounting software capture per invoice? Three to five. Stralevo captures 15 to 40. That difference does not appear in a feature comparison table. It appears in how many questions your team answers in the meeting versus how many require a follow-up.
---
Why Finance Teams Use ChatGPT — and Why That Is the Actual Risk
According to a 2025 LayerX study, 77% of employees paste corporate data into AI prompts, and 82% do so from personal accounts. In finance, that data includes margin calculations, tax strategy documents, and client financial records processed under confidentiality agreements. When Samsung engineers pasted semiconductor source code into ChatGPT — three separate incidents in a single month — the data left the company permanently, processed on US servers with no audit trail and no recall mechanism (Bloomberg, April 2023).
Under the CLOUD Act of 2018, US law enforcement can compel any US-based company to produce data stored anywhere in the world, with no notification required to the company whose data is involved. Policy documents telling finance teams to stop using ChatGPT, without providing a faster alternative, have a 0% success rate.
ChatGPT also requires more effort than it looks: copy data from multiple sources, paste into a chat window, iterate on prompts until the output format works, manually verify numbers against source documents. Forty-five minutes minimum for the same question Stralevo answers in under 10 seconds. NativeAI — Zero-Prompt Delivery — means the system already knows the chart of accounts, reconciliation rules, and reporting templates. The first answer is the right answer, formatted correctly, source-cited, ready to take to the board.
---
Compliance Without the Emergency Project
France's DGFiP — Direction Générale des Finances Publiques, the national tax authority — can request FEC files from any business on 15 days' notice. FEC (Fichier des Écritures Comptables) is the standardised accounting export format used for tax audits. Non-delivery carries a 5,000-euro penalty. Most firms treat FEC production as an emergency project each time: gather data, cross-reference systems, format correctly, deliver under deadline pressure.
FEC-ready data is maintained continuously by Stralevo — a built-in consequence of complete document extraction, not a separate compliance project. When every field is extracted on arrival and every answer is source-cited back to original documents, the audit standard is met by default. URSSAF deadlines (social security contributions that trigger automatic penalties for late declarations), DSN declarations (monthly payroll reports sent directly to social agencies), DPAE filings (the pre-hire employment notices required before every new contract starts) — every French regulatory requirement that depends on complete, accurate financial records becomes easier to meet when 100% of the data is already extracted. Compliance arrives as a side effect.
---
One Firm, Eighty Clients
An 80-client French accounting firm processing 4,000-plus invoices monthly across Sage and Cegid deploys Stralevo's Client Portal at 5 euros per client per month — 400 euros per month total. Their controllers run cross-portfolio queries across all 80 client accounts simultaneously: analysis that large accounting groups currently bill as consulting projects, at 15,000 euros and two weeks of elapsed time. With Stralevo, the same analysis returns while the client is still on the call.
Deploying Stralevo through one accounting firm activates financial intelligence for 80 SME clients simultaneously — without each client separately evaluating, procuring, or onboarding. The firm earns a new capability to offer. Each client gains access they could not have purchased alone at that price.
---
Twelve Months In
Fast answers produce more questions. A CFO who gets a supplier price analysis in 8 seconds asks the follow-up they would have dropped if the first answer took 3 hours. The Knowledge Graph grows with every document added — a firm running Stralevo for 12 months has a richer, more connected financial map than one that started last week.
At month 12, different questions become answerable: has this supplier's payment terms changed over the past year, which contracts auto-renewed without anyone noticing, what was average days-payable-outstanding by vendor in each quarter. None of those questions are answerable from a flat ledger export. All of them are answerable once the full document history is indexed.
Firms that adopted digital client portals in 2015 became the new baseline their clients expected. The ones that waited found their best clients had moved on. The same transition is forming now. The CFOs who understand the four-layer architecture today are building a 12-month knowledge advantage their competitors will start from zero to match — when the capability becomes table stakes in three to five years, the Knowledge Graph history will not be transferable.