Reconcile.Remember.Resolve.
AI agents that do the first pass. Memory that remembers across runs.
Two agent modes. Cross-run exception memory. Counterparty pattern intelligence. Built for modern finance teams.
Sign up for early access — 1 month of Growth Tier free upon launch.
5 reconciliations to start · No credit card required · Built for Indian marketplace and gateway formats
Press play, or pick a flow.
Eight branches. One to three minutes each. Narrated by Ruhaan, recorded on the real product — Amazon settlement audit with four parallel checks, AI translating a French invoice, ACRE closing a payroll exception, Tally bank rec end-to-end. No mock UI, no sizzle reel.
Do the first pass
Column mapping, rule drafting, exception suggestions — proposed by AI, signed off by you.
Remembers across runs
Late settlements auto-link to the original break. Counterparty patterns surface on cycle three.
Probabilistic, not binary
Every match gets a 0–100 confidence score. Deterministic, reproducible, auditable.
Eight flows. Pick the one that matches your day.
- 01~3 min
Marketplace Settlement Audit
Amazon, Flipkart, Meesho, Razorpay — auto-detected. Four parallel audits in one cycle: bank-payout match, fee-leak detection, GMV waterfall, GST input credit. Built for Amazon Settlement Report V2 (GST-aware commission), Flipkart's four-fee structure, Meesho's price-base nuance.
- 02~3.5 min
Run a reconciliation + close a break
Two files, AI-drafted column mappings and rules, ACRE's 6-stage probabilistic matching, then one exception worked end-to-end — investigate, assign, accept, close. The exception is remembered for next cycle.
- 03~3 min
Document Intelligence
PDFs, scanned invoices, vendor bills in French or Italian, phone photos. Textract for structured docs, Vision AI for the rest — non-English invoices come out as English-tagged structured rows with side-by-side preview and editable extraction.
- 04~2 min
Tally Ledger Cockpit
Separate product for Tally customers. Five workflows auto-detected from your Tally exports: customer settlement, vendor settlement, bank reconciliation, party statement, ledger transfer. Deterministic accounting first; bookkeeper-ready output.
- 05~45 sec
Reports & Audit Lineage
Cross-cycle artefacts your statutory auditor expects: lineage from raw upload to final disposition, SOX evidence pack, immutable audit log. Hand it over as-is, no follow-up requests.
- 06~1.5 min
Workspace Settings
Per-organisation defaults that travel with the team. Four matching strictness modes — Lenient, Balanced, Strict, Auditor — to tune ACRE's pairing aggression to your scenario. Every change is logged.
- 07~1 min
Plans & Billing
Free to start, no card required. Paid tiers from ₹3,999/month. Change anytime, prorated, no sales call required for tier changes.
- 08~45 sec
Share Feedback
Direct line to the product team. No support queue, no chatbot. Every note is read within a working day.
Spreadsheets miss the settlement issues that erode margin.
Manual reconciliation makes it hard to trace commission variance, missing remittances, COD delays, and unmatched bank credits across multiple marketplaces and gateways.
Minutes, not hours
Review exceptions and fee variance in one workflow instead of stitching together CSVs by hand.
Audit-ready output
Give finance, operations, and CAs a clear discrepancy trail with source and target context.
Bank credits, settlement reports, rate-card checks, and COD follow-up move into a single review surface instead of separate manual checks.
Reconciliation is a stateful problem solved with stateless tools.
The same exceptions recur every cycle against the same counterparties. Schemas drift. Settlements split and aggregate. Most tools rebuild everything from scratch each run. ReconPe pairs probabilistic matching with cross-run memory so the system compounds over time instead of plateauing.
Settlement reports, payment gateways, bank credits, and internal rate cards all need to line up.
Commission, shipping recovery, ads, and other deductions vary by category, marketplace, and cycle.
Manual workflows often show issues after the settlement cycle has moved on.
Where stateless reconciliation breaks down
Concrete examples from our active test market — Indian marketplace and gateway settlements. The underlying failure modes apply anywhere multi-source reconciliation runs on a cycle.
Formats change faster than spreadsheet logic
Column names, report layouts, and deduction fields shift across marketplaces and gateways. Manual mappings break quietly.
ReconPe auto-detects marketplace formats and maps them into a consistent reconciliation workflow.
Fee validation is hard to maintain manually
Commission logic, shipping recovery, and category-based deductions are difficult to verify at line level inside spreadsheets.
ReconPe audits settlement lines against configured rate cards and flags likely variance with line-level context.
Remittance follow-up gets buried across reports
COD and settlement timelines stretch across marketplaces, gateways, and bank credits, so delays are easy to miss.
Timeline-based tracking shows what was expected, what arrived, and what still needs follow-up.
Meet the team
Four named agents. One reconciliation stack.
ReconPe is not a black-box AI. It’s a team of focused agents — each with its own job, its own voice, its own place in your workflow — so you always know who just did what, and why.
Hermes
Your co-pilot.
The chat front door. Ask what's going on, start a reconciliation, draft a rule set — Hermes has typed access to every tool and keeps the session context so follow-ups never start over.
Argus
Many eyes on every run.
Runs the Ask and Investigate loops, reviews every exception, and proposes fuzzy matches for unmatched records. When confidence is high, Argus resolves and logs its reasoning — when it's not, it hands you the evidence.
Iris
Reads your files. Drafts your rules.
From messy spreadsheets and scanned PDFs to a clean canonical dataset. Iris detects the domain, maps the columns, reads invoice images, and drafts the first working rule set. You approve, she ships it.
Clio
Writes the story. Remembers the history.
Risk narratives, recurring-pattern banners, counterparty intelligence. Clio keeps a history of every run so this month's settlement is always read in the context of last quarter's.
The matching decision itself is always ACRE’s — deterministic, auditable, and reproducible. The team never makes the match. They operate downstream: Iris proposes mappings and rules, Argus investigates and resolves exceptions, Clio narrates risk and remembers history, Hermes runs the room. LLMs reason. ACRE matches. The audit trail stays intact.
Stateful memory
Reconciliation that remembers across runs.
Most reconciliation tools treat every run as an island. ReconPe carries exceptions forward, correlates late settlements against earlier open items, and learns which dispositions your team confirms versus rejects — so the same exception never gets re-triaged from scratch.
The scenario stateless tools can’t close
Every eligible exception gets a deterministic fingerprint (counterparty + amount + direction). When a prior-run open item matches a new exception, ReconPe surfaces the candidate — one click closes both.
After three confirmed settlements for a counterparty, ReconPe tags the pattern — "late settler, avg lag 5 days" — and shows it as an amber banner on future exceptions for that entity.
Users who say "no, that's not the same event" teach the system to stop proposing that pair again. Persisted per-organisation, so different tenants can disagree and both be right.
Cross-run matching is a single indexed database lookup on a deterministic fingerprint, not an AI inference. No black box. Any suggested link shows exactly which fields matched. Open items age out after 180 days — searchable for audit, no longer active candidates. The chat agent can query the pool and pattern library directly via typed tools.
Multi-format ingestion
Messy inputs. One reconciliation.
The hardest part of reconciliation isn’t matching — it’s that your source data arrives as five spreadsheets, two PDFs, and a phone photo of a receipt. ReconPe takes the pile, unifies it with AI, lets you sign off, and hands your matching engine one clean dataset.
What you upload
AI unify
Document Intelligence on PDFs & images.
Schema unification on every column.
What reconciliation sees
Document Intelligence on every format
CSV and Excel flow straight through. PDFs and images route through Document Intelligence on upload — AWS Textract for structured documents like bank statements and invoices, Vision AI for handwritten and multilingual receipts. Extracted rows are indistinguishable from a CSV on the other side. Included on every plan — one trial document on Free, scaling to 1,500 pages a month on Pro.
AI proposes. You sign off.
The model reads every source's schema and samples, collapses them into one canonical set of fields, and surfaces the full mapping for review. Edit anything, preview 10 sample rows, then commit.
A single source for recon
Committing materialises one unified dataset. It shows up in your reconciliation picker like any other file — because downstream, it is one. The matching engine never knew the input was five files in five formats.
The reconciliation engine below never sees the messy side. ACRE matches one clean dataset, not five.
Proprietary Algorithm
ACRE. Adaptive Cascade Reconciliation Engine.
Most reconciliation tools do exact-key lookup and call it matching. ACRE runs a 6-stage probabilistic pipeline — profiling your data, cascading through blocking levels, scoring every candidate pair with Bayesian confidence, and enforcing domain invariants like conservation law before a single exception is raised.
6-stage pipeline
Exact key → relaxed fuzzy → LSH nearest-neighbour. Catches matches even when order IDs have typos, date formats differ, or amounts are off by rounding.
Every candidate pair gets a Fellegi-Sunter confidence score from 0–100. No more matched / unmatched — ACRE tells you how confident it is and why.
ACRE learns Fellegi-Sunter field weights from your organisation's feedback. The more you use it, the more accurately it matches your specific settlement patterns.
ACRE runs deterministically — no LLM calls, no latency, no hallucinations. It produces confidence scores, per-field agreement levels, and explainability metadata for every match decision. AI then consumes that output to answer plain-language questions like “why was this exception raised?”, suggest resolutions, and surface risk narratives — grounded entirely in what ACRE computed, not inference.
Closing the books, not the marketplace?
ReconPe FinanceOps brings subset-sum matching to AR-to-GL and bank-to-GL tie-out. Built for controllers, FinanceOps leads, and CAs — same engine, ledger-shaped voice.
Reconciliation is a team sport.
ReconPe is not a single-user tool. Invite your team, assign exceptions to the right person, track disposition through approval chains, and keep a full audit trail of who decided what and when.
Invite your team
Email-invite flow, role-based access (Analyst, Finance, Compliance, Admin), seat caps per plan.
Assign exceptions
Route each exception to the right owner. “Assigned to me” view focuses individual queues. Every assignment logged.
Approval chains
Configure approval rules for high-value dispositions. Reviewer and approver roles separate. Full audit trail.
₹1,185 shortfall · category rate drift vs rate card
Assigned by Amit · 2h ago
Approval chain
Why teams move off spreadsheets to ReconPe
Matching, fee audit, and follow-up inside one workflow instead of separate tools.
| Feature | Excel | Cointab | Paxcom | ReconPe |
|---|---|---|---|---|
| Price | Free | Rs 5,000+/mo | Rs 10,000+/mo | Rs 0-Rs 3,499/mo |
| Hybrid matching | No | Rules only | Rules only | Rules + AI scoring |
| Commission audit | Manual | Basic | Basic | Rate-card aware |
| Risk scoring | No | No | No | AI risk score + tier per run |
| Workflow | Fully manual | Rule-based, analyst-driven | Rule-based, analyst-driven | AI first pass; analyst reviews and closes |
| Marketplace auto-detection | Manual | Limited | Moderate | 7 Indian formats (deep audit on Amazon/Flipkart/Meesho/Razorpay) |
| Cross-run exception memory | No | No | No | Yes — org-wide pool, fingerprint, pattern intelligence |
| AI agent modes | No | No | No | Two — Investigate (fixed pipeline) + Ask (ReAct) |
What the product actually ships
AI agents, stateful memory, probabilistic matching, and per-marketplace depth in one workflow.
Two AI agent modes
Agent Investigate runs a fixed-pipeline root-cause analysis on a reconciliation and streams its trace live. Ask Agent is a ReAct planner loop with typed tool access to the exception pool, pattern library, and risk analytics.
Stateful cross-run memory
Exceptions survive beyond the run that produced them. A deterministic fingerprint (counterparty + amount + direction) correlates today's exception against prior open items, so late-arriving settlements link automatically.
Counterparty pattern intelligence
After three confirmed settlements for a counterparty, ReconPe tags the pattern ("late settler, avg lag 5 days") and surfaces it as an amber banner on future exceptions for that entity.
ACRE probabilistic matching
6-stage cascade: profile, block, Fellegi-Sunter score, Hungarian assignment, domain validation, exception classification. Deterministic, auditable, and adaptive per organisation.
AI-suggested mappings and rules
Upload a new settlement file and AI proposes canonical column mappings and matching rules for human approval. Schema drift gets absorbed without manual remapping.
AI exception resolution
Every flagged exception gets a plain-language explanation and a resolution suggestion grounded in reconciliation state, historical pattern library, and counterparty memory. Analyst confirms.
Commission variance audit
Category-aware rate-card comparison. Real depth for Amazon Settlement Report V2 GST structure, Flipkart four-fee breakdown, and Meesho price-base semantics.
Risk scoring + narrative
Every reconciliation gets a 0–100 risk score, a tier (Low / Medium / High / Critical), and an AI-generated plain-language narrative for finance, ops, and audit stakeholders.
Team collaboration + assignment
Invite your team by email, assign exceptions to specific members, filter to an "Assigned to me" queue, configure approval chain rules for high-value dispositions. Role-based access (Analyst, Finance, Compliance, Admin) with full audit trail.
Multi-provider AI
Route to Anthropic Claude, OpenAI, DeepSeek, or Google Gemini — configurable per-tenant. Pick based on data-residency, cost, or compliance needs.
Explore by marketplace, topic, or comparison
Deep-dive pages for each supported marketplace, head-to-head comparisons, a reconciliation glossary, and writing on how reconciliation workflows scale.
Marketplaces
Compare
From the blog
Fractional bookkeepers for Indian e-commerce: when software stops being enough
When does an Indian e-commerce founder hire a fractional bookkeeper instead of a full-time accountant or another tool? A decision framework — order volume, GST complexity, marketplace count, dispute load — and what a vetted fractional bookkeeper actually does that software does not.
✦ AI-Powered · FAQ
How ReconPe AI handles your settlement reconciliation
From column mapping to exception resolution — AI takes the front seat.
AI handles the heavy work at every stage. When you upload a settlement file, AI identifies the format and maps the columns automatically — no manual field matching. It then suggests the matching rules based on your data. After the reconciliation runs, AI does a second pass on every unmatched record to find potential matches that rule-based logic missed. It auto-generates resolution suggestions for exceptions, scores each reconciliation by financial risk, and writes a plain-language narrative explaining what happened. Your team reviews decisions. AI does the rest.
When you upload a settlement report from Amazon India, Flipkart, Meesho, Razorpay, Cashfree, PhonePe, or PayU, ReconPe's AI analyses the column headers, data types, and value patterns to identify what each field represents. It suggests column-to-column mappings between your source files so you can review and confirm rather than build the mapping from scratch. Most standard marketplace and gateway formats are recognised immediately.
After column mapping, AI inspects your data and proposes a rule set — which fields to match on, what tolerance to apply for numeric differences, and how to handle format variations like date styles or currency symbols. You review the suggested rules before running the reconciliation. This replaces the manual process of writing matching logic for each new file format or settlement cycle.
For every exception flagged — whether a missing record, value mismatch, or commission variance — AI generates a resolution suggestion grounded in the actual data. It explains why the record was flagged, what the likely cause is, and what action the team should consider. Exceptions are also scored by severity and financial impact so your team knows which ones to review first. AI suggestions are generated automatically in batches when a reconciliation completes.
Yes. After the primary rule-based matching pass, AI runs a second pass on all unmatched records. It uses fuzzy comparison and contextual reasoning to identify POTENTIAL_MATCH cases — records that are likely the same transaction despite small differences in reference numbers, amounts, or dates. These are surfaced as a separate exception type for your team to confirm, rather than being silently dropped.
Every completed reconciliation gets an AI-generated risk score based on three factors: match rate (how many records were cleanly matched), exception severity (the mix of critical, high, and medium issues), and exception type (missing records carry more weight than minor value mismatches). The score runs from 0–100 and maps to a risk tier — Low, Medium, High, or Critical. AI also writes a short narrative explaining the score in plain language, so your finance team or CA understands the risk without reading every line.
ACRE stands for Adaptive Cascade Reconciliation Engine — ReconPe's proprietary 6-stage matching pipeline. Most reconciliation tools do a simple key lookup: if the order ID matches, it's a match. ACRE runs a probabilistic pipeline instead: it profiles your data to understand field types and cardinality, cascades through three blocking levels (exact key → relaxed fuzzy → LSH nearest-neighbour) to find candidates even when IDs have typos, scores every candidate pair with a Bayesian confidence score from 0–100 using Fellegi-Sunter weights, resolves N:M key groups using the Hungarian optimal assignment algorithm, validates domain invariants like conservation law (∑source = ∑target) and temporal ordering, then classifies exceptions by type and severity. The result is a traceable, confidence-ranked output — not a flat matched/unmatched list.
Yes. The 'Adaptive' in ACRE refers to its learned field weights. ACRE records Fellegi-Sunter match and unmatch probabilities per field, per domain, and per organisation. As your team confirms or overrides matches, ACRE updates these weights so that fields your data treats as highly discriminative (like a specific settlement reference format) are weighted more heavily in future reconciliations. Each organisation's ACRE instance adapts to their specific marketplace mix and settlement patterns over time.
Settlement reconciliation is the process of matching payment settlements received from marketplaces or payment gateways against your order records, bank credits, and expected fee deductions. It confirms that every rupee credited matches what was expected — and flags discrepancies like commission overcharges, missing remittances, or delayed COD payments before they become write-offs. ReconPe uses AI to automate this process end-to-end.
ReconPe auto-detects settlement export formats from Amazon India, Flipkart, Meesho, Razorpay, Cashfree, PhonePe, and PayU. Upload your settlement report and ReconPe's AI identifies the format automatically — no manual column mapping required.
ReconPe tracks cash-on-delivery collections from the order level through to bank credit. Each COD cycle is monitored across dispatch, delivery confirmation, remittance expectation, and actual bank receipt — so your team can see what is pending, what arrived, and what is overdue, without building a manual tracker in spreadsheets.
Yes. ReconPe's free plan includes 5 reconciliations to try the product, with core AI matching and a downloadable review report — no credit card required. Paid plans start at Rs 3,999/month for teams that need higher volume, more marketplace connections, or advanced features like AI exception review and risk analytics.
GST reconciliation is available on the Enterprise tier. It lets you match GSTR-2A / GSTR-2B data against purchase invoices and settlement deductions so your finance team or CA can close GST filings with confidence.
Excel is the most common starting point for Indian marketplace reconciliation, but it breaks down quickly with multiple marketplaces or high order volumes. Dedicated tools include Cointab (Rs 5,000+/month, rules-based), Paxcom (Rs 10,000+/month), and ReconPe (free tier available, AI-powered). ReconPe differs from both in using probabilistic matching via the ACRE engine rather than rigid rule lookups, which handles the format inconsistencies common in Indian marketplace exports. It also adds AI exception review and risk scoring that spreadsheet workflows cannot replicate.
Cointab is a reconciliation tool used by some Indian commerce teams, priced at Rs 5,000 or more per month. Alternatives include: (1) ReconPe — AI-native with a free tier, probabilistic ACRE matching, and risk scoring, faster at approximately 3 minutes to first review vs Cointab's 30 minutes; (2) Paxcom — more expensive at Rs 10,000+/month, rules-based; (3) Excel — free but fully manual, scales poorly. ReconPe's key differentiators over Cointab are AI exception review, Bayesian confidence scoring per match, and adaptive field weight learning per organisation.