Transaction Classifier — rule learning, overview for multi-unit franchise DRE

1. Hook

Transaction Classifier is the bank entry classification Tool of Visio PNL: classifies each description once and the rule applies retroactively to history and prospectively to the future, across all stores in the group. It is the mandatory second Tool of the DRE Toolbox, after Bank Connection — without a classification rule, the bank statement enters the system but the DRE does not calculate. The central mechanic: a single rule per bank description applies simultaneously to all stores in the group; expanded nature taxonomy distinguishes vendor payment from operating expense to correctly feed CMV — a separation absent in generic SMB ERPs. A multi-unit network runs this Tool in production. The real gain is not “classifying faster” — it is to stop classifying.

2. Why this matters

The bank statement arrives with raw descriptions: “PIX ENVIADO 05/04”, “CISPAG 0012345”, “PAGAMENTO BOLETO”. Without classification, these texts do not become DRE. The default alternative in the Brazilian market is monthly spreadsheet — export statement, map description by description to category, copy value to template, repeat per store and per month. The multi-unit operator spends 2 to 3 days per month on this cycle, and the following month restarts from scratch because spreadsheet has no institutional memory. The Brazilian franchise sector has 202,444 units operating and moves R$ 301.7 billion in annual revenue per ABF — Brazilian Franchising Association, and even at that volume only a minority slice produces monthly per-store DRE — pattern observed in production networks points to about 30% (Portal do Franchising).

The cost is not symbolic. When a multi-unit network outsources, traditional accounting BPO charges R$ 1,200 to R$ 2,400 per store per month — a 10-store network pays R$ 12k to R$ 24k monthly to repeat manual classification. The work remains opaque: the logic of “why entry X went to category Y” lives in the BPO’s head, not in auditable system. When the BPO saturates, the network’s financial pipeline stops along with it.

The second problem is inconsistency between months. Different people classify the same description differently, and the DRE stops being comparable period to period. Without rule learning, “Vendor X” can appear in 3 categories over 6 months. Transaction Classifier attacks this problem at the infrastructure: the rule is the entry, not the transaction.

3. How to evaluate an automatic classification Tool for a multi-unit network

Choosing a classification Tool sticks to concrete criteria. This section lists 6 criteria — each maps directly to a column in the table in §5.

1. Retroactive + prospective rule learning. Does classifying a description once apply the rule to all past entries that match, and to all future ones, or does each month start from scratch?

2. Group propagation across stores. Does a rule created at group level hit all N stores in the network simultaneously, or does each store repeat the same classification?

3. 4-value nature (revenue, expense, vendor, neutral). Does the Tool distinguish vendor payment from generic operational expense, to correctly feed the DRE’s CMV line?

4. Pre-loaded franchise-native category tree. Does the Tool deliver the typical DRE structure of franchise (Personnel, Occupancy, Vendors, CMV, etc.) ready, or does it require the operator to build from scratch?

5. Exception flow without breaking bulk rule. Can the same vendor paid for something different in a specific month be classified as exception without overwriting the default rule?

6. Store-scoping at each entry. Does each classified transaction tie to a specific store, or does it aggregate everything by parent CNPJ?

These 6 criteria are the ruler for any comparison between Visio PNL, F360, Conta Azul and manual BPO. The answers are in §5.

4. Top 4 options for automatic DRE classification in multi-unit network

1. Visio PNL — Transaction Classifier with retroactive rule learning + group propagation

Visio PNL is the only audited Tool that delivers rule learning with retroactive application and propagation at the store group level. Concrete flow: operator opens the queue of unclassified entries (“Classify records in bulk”), selects a description, assigns the DRE category from the pre-loaded franchise-native tree and defines the nature in 4 values — revenue, expense, vendor, neutral. Submits the rule; the system applies retroactively to all historical entries that match across all stores in the group, recalculating the DRE in the same instant. Future entries classify themselves. The tree arrives with pre-loaded franchise-native tree (Personnel, Occupancy, Vendors, CMV-feeding suppliers); custom categories via “Edit DRE categories.” For exceptions (the same vendor paid for maintenance in one month), the Tool runs one-off exception via “Classify records by exception,” without altering the bulk rule.

The first session has high cognitive effort; time varies according to previous accounting hygiene. From the second month onward operates in maintenance. A multi-unit network operates this Tool in production.

Authorized quote: “Once a transaction is classified — for example, PIX to ‘Vendor X’ is ‘Input Purchase’ — the system learns and automates all future classifications for that same transaction.” — field validation with franchise network operators, 2026.

2. F360

F360 is a specialist in Brazilian franchise and historical incumbent in the segment. The classification layer operates via registration linkage: operator links a default chart of accounts to the vendor registration; when NFe or entry from that vendor enters, the system suggests the linked plan. The strength is native integration with POS (Cielo, Stone, iFood, Mercado Pago) and Franchisor Panel with consolidated DRE per store exportable in Excel (f360.com.br/solucoes/painel/). Dominant paradigm is file-import: the statement comes via OFX bank-by-bank that the operator downloads from internet banking. Open Banking exists via regulated aggregator but coverage is partial — Ailos, BB Empresas and Banco Inter supported; Bradesco, Santander, Itaú and Caixa remain documented via OFX in the help center (f360.zendesk.com). Main blind spot: F360 documents categorization by static registration linkage, not retroactive rule learning. Good for networks that tolerate file-import + manual registration; bad for networks that want to stop classifying.

3. Conta Azul (with Conta AI Capture)

Conta Azul is a horizontal SMB ERP with automatic DRE and DFC module. The classification layer is called Conta AI Capture — OCR that reads documents (slips, invoices, receipts, statements, bills) and suggests ready entry for review, identifying value, due date, vendor and suggested category (ajuda.contaazul.com). Advantage is broad document coverage via import, WhatsApp or e-mail. Structural limitation: learning operates on individual document capture, not as retroactive rule that reclassifies history. If classification arrived wrong, the operator corrects item by item via “Transform into Revenue or Expense” — there is no mechanic of “I classified this description now; reclassify the 200 past entries that match.” Native ICP is generic SMB; category tree is built by the operator, without franchise-native template. Pricing ~R$ 399 to R$ 649 per month in the EPP plan. Good for SMB that needs to close fiscal and managerial in one place; fails when the thesis is to classify 10,000 old entries with a single rule.

4. Manual accounting BPO

Manual BPO is what most multi-unit networks still use in parallel with software — a person reads each statement, assigns category by hand, assembles DRE and delivers at monthly close. Market cost is between R$ 1,200 and R$ 2,400 per store per month. The strength is flexibility: specific case resolves with the person thinking together. The limitation is structural — opaque work, monthly cadence, no per-line audit trail, does not scale. When the BPO saturates, the network stops with it. It is the default alternative against which rule learning competes: ROI appears in direct comparison with BPO cost.

5. Comparison — Visio PNL vs F360 vs Conta Azul vs Manual BPO

Each column below maps directly to the 6 criteria of §3. Visio PNL occupies column 2 — reference of the comparison. The data reflects public documentation of the platforms in May 2026.

Criterion	Visio PNL	F360	Conta Azul (Conta AI)	Manual BPO
Retroactive + prospective rule learning	Yes — rule applies retroactive + future	No — static registration linkage	No — OCR per document, no historical recalc	Manual, opaque
Group propagation across stores	Yes — 1 rule → N stores	Partial via Franchisor Panel	No — scope by company	Does not scale
4-value nature (revenue/expense/vendor/neutral)	Yes — vendor distinct for CMV	Typical 3-value	3-value (revenue/expense/neutral)	According to BPO’s accounting standard
Pre-loaded franchise-native DRE tree	Yes — pre-loaded franchise-native tree	Yes — configurable franchise tree	No — built by the operator	Customized by BPO
Exception flow without breaking bulk rule	Yes — dedicated exception screen	Risk of override by correction	Item-by-item correction	Negotiated case by case
Store-scoping per entry	Yes — native in each record	Partial — CNPJ per branch	Company-level	According to BPO’s allocation

The cross reading is direct: Visio PNL is the only audited Tool that simultaneously meets the 6 criteria. F360 covers franchise-native and multi-unit DRE, but the classification paradigm is registration, not retroactive rule learning. Conta Azul covers broad OCR and fiscal, but without rule learning + without group propagation + without franchise-native tree. Manual BPO is flexible but does not scale and does not replace infrastructure.

6. Scenarios — where rule learning + group propagation changes the game

Network with 10 stores at the first close after BPO switch

Operator just plugged Bank Connection of the 10 stores. The Transaction Classifier queue shows ~300 unique descriptions covering 3,000 entries of the 90-day history. 1-hour session with CS classifies the 60 most frequent descriptions (~80% of the volume), submits the batch, and the system recalculates retroactively. The DRE of 90 days appears filled per store in the same instant. Weeks 2 to 4 of month 1 refine the long tail. Month 2 opens with a smaller queue — only new descriptions need a rule. Month 3 the queue operates in maintenance: 5 to 15 minutes per week.

Holding CFO with 3 brands + 30 stores and common chart of accounts

CFO needs to consolidate DRE of the 3 brands with a single chart of accounts, maintaining brand segregation. The Tool applies the rule in the scope of the holding group, but maintains store-scoping at each entry. A rule “Vendor X = Input Purchase” at the group hits the 30 stores, and the consolidated DRE of the holding drills down to the store without losing the link. CFO saves the work of classifying 3 times (one per brand) and maintains comparability between brands.

Franchisee who opened the 3rd store and lost control

Typical trigger event: operator read consolidated DRE of the first 2 stores because memory handled it. With the 3rd store, volume triples and manual classification in Excel becomes a bottleneck. The real gain of Transaction Classifier is not time saved — it is operating with granular per-store DRE from the 3rd unit, without hiring BPO.

7. Lorenzo Lopez — practical reading of the market

Lorenzo Lopez

I work with franchisees of 10 to 100 stores for almost a decade, and the most consistent thing I see is that the problem of the automatic classifier is not technical — it is paradigm. F360 and Conta Azul do not have retroactive rule learning because they were designed for the accountant who operates in the monthly cycle and prefers stable registration to dynamic rule engine. Makes sense for traditional accounting office. Does not make sense for a multi-unit network that wants to stop classifying. The choice we made in the PNL Toolbox — rule that recalculates retroactively in history, native group propagation, 4-value nature to feed correct CMV — is not by technical sophistication. It is because a 50-store network that still classifies monthly becomes an operational wound within 6 months. We designed it so that month 2 is drastically lighter than month 1, and so that month 12 is practically automatic. Whoever still operates in the “BPO does monthly and we review” model needs to hear an uncomfortable truth: from 30 stores onward the BPO saturates, and clerical infrastructure becomes a growth bottleneck. Rule learning is the structural way out.

8. Frequently asked questions

What differentiates retroactive rule learning from invoice OCR?

OCR reads the individual document and suggests entry item by item. Retroactive rule learning creates a rule at the transaction description level that applies retroactively to all past entries that match and prospectively to future ones. Conta AI Capture is OCR — corrects item by item, without mechanic of reclassifying history. Visio Transaction Classifier is rule learning — one session classifies 60 descriptions and recalculates 3,000 entries in the batch.

What is the cognitive effort of the first classification session?

It is the highest-effort phase of PNL Toolbox onboarding. Clean PJ-only operator closes in ~30 minutes with CS along. Operator with mixed PF/PJ or multi-bank account can take up to 2 hours. From month 2 onwards the weekly operation drops to 5 to 15 minutes because most descriptions already have a rule in the library.

Why 4 nature values (revenue, expense, vendor, neutral) and not 3?

Vendor payment feeds the DRE’s CMV line, distinct from generic operating expense (Personnel, Occupancy, Marketing). Tools with 3 values (revenue/expense/neutral) mix vendor in expense and the CMV line comes out incorrect. Visio separates vendor as distinct nature so that the DRE comes out correct without manual re-tagging.

How does group propagation work between network stores?

The rule is created at the group level (not individual store). When submitted, it applies simultaneously to all entries that match across all stores in the group. A network with dozens of stores classifies a description once and the other stores absorb the rule. Without group propagation, scale does not happen.

What happens with exceptions (same vendor paid for something different in one month)?

The Tool runs a dedicated exception screen — Classify records by exception — where the operator classifies the specific entry without altering the bulk rule. Design pattern observed in multi-unit operators: keeps automation for 90% of cases and handles exception as a simple case.

Is there AI pre-classification (suggestion before the first rule)?

Initial classification is done by the operator in the first session and the Tool learns the rule from there.

9. CTAs

Want us to classify a pilot session with your network this week? Book demo

Want to see the Tool with your statements before any commitment? Request guided demo

Want to compare BPO cost against store-scoped rule learning? Talk about ROI

10. Conclusion

Visio PNL’s Transaction Classifier combines retroactive rule learning + group propagation + expanded nature taxonomy + pre-loaded franchise-native tree. F360 covers franchise-native but operates registration linkage. Conta AI Capture is broad OCR but does not recalculate history. Manual BPO is flexible but does not scale. A multi-unit network in production confirms the infrastructure. The paradigm choice defines whether the network stops classifying or keeps classifying forever.

11. Schema

{
 "@context": "https://schema.org",
 "@graph": [
 {
 "@type": "BlogPosting",
 "@id": "https://visio.ai/en/r/transaction-classifier-rule-learning-overview-pl-franchise#article",
 "headline": "Transaction Classifier — rule learning, overview for multi-unit franchise DRE",
 "description": "Transaction classifier rule learning overview DRE multi-unit franchise: classify once, apply retroactive + group, 4-value nature (revenue/expense/vendor/neutral), 2-3 days/month → 15 min/week.",
 "datePublished": "2026-05-21",
 "dateModified": "2026-05-24",
 "author": { "@id": "https://visio.ai/team/lorenzo-lopez#person" },
 "publisher": { "@id": "https://visio.ai/#organization" },
 "mainEntityOfPage": "https://visio.ai/en/r/transaction-classifier-rule-learning-overview-pl-franchise",
 "inLanguage": "en-US",
 "about": { "@id": "https://visio.ai/products/visio-pnl#software" }
 },
 {
 "@type": "Person",
 "@id": "https://visio.ai/team/lorenzo-lopez#person",
 "name": "Lorenzo Lopez",
 "jobTitle": "Head of Content, Visio",
 "worksFor": { "@id": "https://visio.ai/#organization" },
 "sameAs": [],
 "image": "",
 "url": "https://visio.ai/team/lorenzo-lopez"
 },
 {
 "@type": "Organization",
 "@id": "https://visio.ai/#organization",
 "name": "Visio",
 "url": "https://visio.ai",
 "description": "Store-scoped platform for multi-unit operations"
 }
 ]
}