FynTwin
Croatian financial analysis platform that turns an exported accounting journal into GFI-POD statements, 40 ratios, forensic testing and going-concern models, with AI commentary citing the applicable HSFI or MSFI standard for auditors, accountants and CFOs.
What is FynTwin?
FynTwin presents itself as a digitalni financijski stručnjak, a digital financial expert built for the Croatian accounting framework. Its starting point is the posting journal, the dnevnik knjiženja, exported as CSV, XLSX or XML from whatever bookkeeping software a firm already runs. From that one file the platform walks a complete chain: import, automatic account-to-AOP mapping, validation, statutory statements, analysis, forensics and export.
The reporting module rebuilds the regulatory statements in GFI-POD format across 188 AOP positions: Bilanca, RDG (around 70 AOP) and Bruto bilanca (200+ AOP), with automatic balance checks and a link from any figure back to the entry that produced it. On top sit 40 ratios in eight named categories - liquidity, leverage, profitability, activity, economy, structure, growth and valuation - a direct-method cash flow across three activities, and a what-if simulator driven by 11 variables.
Forenzika examines the entire journal rather than a sample, across nine detection domains: Benfordov zakon, round numbers, duplicates, related parties, purchase anomalies, cash anomalies, a period-end heatmap, revenue recognition and backdating. Findings aggregate into a risk score from 0 to 100, and each one traces back to the originating transaction. A going-concern module cross-checks three insolvency models - Altman Z-score (1968), Taffler (1977) and Springate (1978) - on the same data, projects cash 18 months out in three scenarios and monitors seven liquidity signals. A separate procjena vrijednosti module values the business by DCF with an explicit five-year projection, sector market multiples and the net asset method, using a sector WACC.
The claimed differentiator is the AI layer. Every comment on a position cites the standard and paragraph it rests on - HSFI 11 for receivables, for instance - and each of the 188 positions carries one of four statuses, so nothing arrives as a generic answer. An integrated Akademija adds lessons and learning paths, exercises, a sandbox, capstone projects, a formula glossary and certificates.
It runs in the browser with nothing to install, and the publisher advertises the whole run as taking about the time of one coffee. FynTwin d.o.o. of Zagreb, incorporated on 26 March 2026 and led by its two co-founders, sells it as a licensed product with a limited number of seats, with access arranged individually.
What it does
- Import an accounting journal exported from any ERP as CSV, XLSX or XML
- Produce Bilanca, RDG and Bruto bilanca in the Croatian GFI-POD format, across 188 AOP positions
- Read 40 financial ratios in 8 categories, with horizontal and vertical analysis and a direct-method cash flow
- Run forensic testing over the entire journal: 9 detection domains and an aggregate 0-100 risk score
- Assess going concern with three insolvency models and project cash 18 months ahead in three scenarios
- Simulate what-if scenarios on 11 drivers and value the business by DCF, sector multiples or net assets
- Question the AI assistant about your own figures and receive answers referenced to the applicable standard
When to use FynTwin / When not to
A quick filter to help you decide if FynTwin is the right fit.
When to use FynTwin
- Audit firms that want full-population screening and a pre-mission risk score before fieldwork starts, plus a pre-filled MSRevS checklist and white-label PDF output.
- Accounting practices handling many client files, where one saved account-to-AOP mapping is reused month after month, quarter after quarter.
- CFOs and financial controllers running monthly close, executive review, what-if simulation and forecasting without rebuilding spreadsheet models.
- Tax advisers who want exposure identified with a pointer to the applicable Croatian tax article (ZPDV, ZPD) rather than a generic flag.
- Company directors and owners with no accounting training, who need their own numbers explained without jargon.
When not to use FynTwin
- Organisations reporting outside the Croatian framework: GFI-POD, HSFI/MSFI, MSRevS and the OIB check digit shape the entire product, so its value drops sharply beyond Croatia.
- Anyone who does not read Croatian. The interface and all published content are Croatian only, and the site has no English version - the publisher does offer the sales conversation and onboarding in Croatian or English, but that is not an English product.
- Teams shopping for a bookkeeping system or an ERP replacement. FynTwin sits downstream of what you already run, and the ERP stays the source of truth.
- Buyers who want to sign up, see a price and get started alone. There is no online registration, no published rate card and no unattended trial.
- Practitioners hoping to hand professional judgement over to software, or IT teams expecting an API. The terms leave interpretation with the user, and no API or software integration is announced.
How to use FynTwin
A typical end-to-end flow, from setup to results.
- Submit the access request form on the contact page. There is no self-service sign-up, and the publisher first checks that the product fits your practice.
- Expect a reply generally within one working day. The conversation is announced as commitment-free, no payment card is asked for, an NDA is available on request, and it can be held in Croatian or English.
- During that conversation, the publisher runs Bilanca, RDG and AI commentary live on a CSV file you supply, rather than on a canned demo.
- Once onboarded, drag in the CSV or XLSX journal for the period you want to analyse - the import step is advertised at around 30 seconds.
- Review the automatic mapping: columns and the account-to-AOP link are detected for you, and anything wrong can be corrected by hand.
- Save the mapping as a template so the next period, quarter or year loads without repeating the work.
- Clear the 12 validation checks - debit/credit balance, date formats, OIB check digit, chart-of-accounts membership, locked period, currency conversion, duplicates, empty codes, decimal separator, UTF-8 encoding, row length and VAT counter-entry - each returning an explicit message with a suggested fix.
- Load the file. Large journals take seconds, and an import can be cancelled if something looks wrong.
- Work through the six analysis types by tab, then the forensics, going-concern and valuation modules, using the four statuses to triage the commented positions.
- Export a branded white-label PDF, an Excel file with live formulas, or JSON.
Pros & Cons
Pros
- Deep regulatory anchoring rather than a localisation layer: GFI-POD, 188 AOP positions, HSFI/MSFI, MSRevS 240/315/570 and the OIB check digit are built into the product.
- Every AI comment is referenced to a standard and paragraph, and every finding traces back to the source journal entry - traceability is claimed end to end.
- Forensic testing covers the whole journal rather than a sample, which is unusual for a screening tool at this level.
- Works with any bookkeeping software that exports CSV, XLSX or XML: nothing to install, and nothing to change in the existing ledger.
- Unusually broad functional coverage for such a young product - statutory statements, ratios, forensics, going concern, valuation and training in one place.
- Above-average privacy transparency: subprocessors named individually with their processing location, a DPA supplied on request, an identifier filter applied before anything reaches the AI, an EU database in Frankfurt, per-client isolation, an immutable activity log and cookie-free analytics with no banner.
- The founding team covers both sides of the problem: an engineer and a licensed tax adviser with an audit background.
Cons
- No public pricing of any kind: no rate card, no range, no plan names anywhere on the site, the sitemap or the llms.txt file, and /cjenik simply redirects to the contact form.
- No self-service sign-up. It is a licensed product with a limited number of seats, sold case by case, with no free trial and no free plan on offer.
- Site and product are Croatian only, with no English version of the site and no alternative locale, and the regulatory scope travels poorly beyond Croatia.
- No API and no named software integration: everything moves through files. There is no mobile application either.
- The generative AI analysis is executed by OpenAI in the United States; the move to EU processing is announced as being in preparation, without a date, and Cohere's processing region is described as still being confirmed.
- Model training on customer data is never addressed - no commitment and no opt-out - and no security certification (ISO 27001, SOC 2, ISAE) is claimed.
- Signs of a very young operation: no named customer or published testimonial, a LinkedIn link repeated on all 20 pages that returns a dead page, terms (clause 8.1) pointing at prices that do not exist, a 99% availability target with no SLA behind it, and no refund of amounts already paid on cancellation.
Pricing & Plans
No price is published. FynTwin operates no rate card: there is no pricing page anywhere on the site, in the sitemap or in the llms.txt file. Both /cjenik and /demo redirect to the contact form, while /cijene, /pricing, /plans and /paketi return genuine 404 pages. The publisher describes the offer as a licensed product with a limited number of seats, access and onboarding being negotiated individually. No free plan and no free trial is offered. The terms of service (clause 8) describe an auto-renewing subscription, cancellable at any time with effect at the end of the current billing period, with amounts already paid non-refundable and prices subject to change without prior notice, changes being notified at renewal. One inconsistency is worth raising with the vendor: clause 8.1 states that prices are available on the website, yet no pricing page exists. The pre-sales conversation is announced as commitment-free and requires no payment card. Consequently no entry price and no currency can be stated for this tool.
Data, GDPR & hosting
A consolidated view of how FynTwin handles your data.
GDPR overview
GDPR implementation is concrete rather than decorative. FynTwin d.o.o., Rudeška cesta 238, 10000 Zagreb, OIB 48151066729, is named as controller. Six data subject rights are listed - access, rectification, erasure, restriction, portability and objection - all exercised through a single channel, info@fyntwin.com, with AZOP (Agencija za zaštitu podataka) named as the supervisory authority for complaints. Legal bases are stated per purpose, a DPIA was performed for the AI analysis, processors are named individually with their processing location, and a data processing agreement is supplied on request. The GDPR is referenced in the Impressum, in the terms (clause 10) and throughout the privacy policy, all three carrying a last-amended date of May 2026. Two gaps: no data protection officer is appointed, and no security certification is claimed anywhere. No Article 27 representative is named, consistent with a controller established in the EU.
Who owns the data?
Clause 6 of the terms is explicit: the outputs generated from your data - the Izlaz, meaning reports and analyses - are your property and you may use them for your own purposes. Three restrictions apply: you may not repackage them into a competing service, distribute other users' outputs, or publish them outside your own client documentation without authorisation. Everything belonging to the platform itself - code, algorithms, design, trademarks, content - remains the publisher's copyright. The security page states it plainly: your data is yours, neither sold, shared, nor used for other purposes. A complete account snapshot is exported on request, and clause 10 allows deletion or export at any time.
Reuse rights
Customers may reuse the reports and analyses produced from their own data without asking permission, for their own needs, subject only to the three restrictions in clause 6 of the terms. On the publisher's side, the purposes are narrow and stated one by one. Website data covers the contact form (name, email, optional company, message) and the newsletter (email), both on consent, plus cookie-free anonymous analytics and abuse prevention on legitimate interest; there is no cookie banner and no cross-site tracking. The privacy policy names five processors with their location: Supabase for the database in Frankfurt, Resend for email, Cloudflare for hosting, security and analytics, Plausible Analytics, and Sentry for error tracking, all declared in the EU. The security page adds two AI processes: OpenAI in the United States for generative analysis, and Cohere for knowledge-base search with its processing region described as still being confirmed. Only limited context is sent, never the database, and a filter strips structured identifiers - OIB, email, IBAN, tokens - along with the company name, while the free-text description of a journal entry is not sent by default. Client data is strictly isolated, a DPIA was carried out for the AI analysis, and a processing register is maintained.
Data retention & training
Hosting summary
The database, authentication and files are held in the EU: Supabase running on AWS infrastructure in Frankfurt, Germany. All five subprocessors named in the privacy policy - Supabase, Resend, Cloudflare, Plausible Analytics and Sentry - are declared as EU-located. Two processes leave that perimeter, both documented on the security page: OpenAI in the United States performs the generative analysis, and Cohere handles knowledge-base search with a processing region stated as still being confirmed. What travels is limited: only the context needed for the answer in progress, never the database, with a filter removing structured identifiers (OIB, email, IBAN, tokens) and the company name, and the free-text description of an entry not sent by default. A migration to European AI processing is announced as being in preparation, without a date. On the platform side, client data is strictly isolated, encrypted at rest and in transit, and backed up in encrypted form with a restore procedure the publisher says it has tested. The site itself is served through Cloudflare; the resolved IP address is anycast and identifies no hosting country.
Where FynTwin works
Country-level availability.
Available in
Things to keep in mind
Risks and trade-offs to weigh before adopting FynTwin.
- The generative AI analysis leaves the EU: it is executed by OpenAI in the United States, and the move to European processing is announced as being in preparation, with no deadline attached to it.
- Cohere is named as a second AI process in the security page FAQ, with a processing region described as still being confirmed, but it does not appear in the list of five subprocessors in the privacy policy. That is a discrepancy between two of the publisher's own documents, and it is worth asking about before signing.
- Neither model training nor certification is addressed: the site never says whether customer data feeds model training and documents no way to opt out, and no security certification (ISO 27001, SOC 2) is claimed, so the security statements remain declarative and unattested by a third party.
- The publisher's own documents do not always agree with each other. Clause 8.1 of the terms points at prices available on the website when no pricing page exists, and the audit standard appears as MRevS 570 on the home page but as MSRevS 570 in the terms and on the forensics and auditors pages.
- Commercial terms deserve a careful read: the subscription renews automatically, amounts already paid are not refunded on cancellation, and the 99% availability figure is presented as an objective rather than as a contractual commitment.
- The terms place the accuracy of imported data and the interpretation of results squarely on the user (clauses 5 and 7.2). The real risk in daily use is treating an AI comment as a conclusion: this is support for professional judgement, not a replacement for it, and a reviewer who stops reading because the status says Dobro has misused the tool.
- The publisher is five months old, with EUR 3,000 of share capital, no named customer and no published testimonial, and the LinkedIn link repeated across the whole site leads nowhere. Ordinary commercial due diligence is advisable.
Setup & Integrations
Technical difficulty
Technically undemanding. FynTwin is a web application: nothing to install, no development work, no change to existing bookkeeping. The only prerequisite is that the accounting software can export the journal as CSV, XLSX or XML. Columns and the account-to-AOP mapping are detected automatically, can be corrected by hand, and are saved as a reusable template. Twelve validation checks return explicit errors with suggested fixes, and an import can be cancelled. Announced timings: about 30 seconds to import, 4 minutes to statements. The real friction is commercial, not technical: access requires a request, qualification by the publisher, then a practice-specific onboarding.
Deployment
Supported languages
Behind FynTwin
Resources
All the official URLs gathered for verification and reference.
Frequently asked questions
Which accounting software does FynTwin work with?
Does FynTwin replace my ERP or my bookkeeping software?
Where is my data held, and what leaves the EU?
Can I get all my data back out?
How is one client's data kept apart from another's?
Do I need a data processing agreement?
Does the AI cite actual standards, or does it improvise?
What does the forensic module actually test?
Which insolvency models are used, and over what horizon?
How do I get access?
Should you pick FynTwin?
FynTwin is a deliberately vertical product: it addresses the accounting and audit profession in Croatia and makes no attempt to be anything else. What stands out is the functional depth for a publisher incorporated on 26 March 2026. Statutory GFI-POD statements across 188 AOP positions, 40 ratios, a nine-domain forensic engine, three cross-checked insolvency models, business valuation and an in-product academy would be a wide spread for a far older vendor.
The central argument holds up on the evidence published: the AI is not meant to answer generically. Each comment names the standard and paragraph it rests on, and each finding traces back to the originating journal entry. Privacy and security transparency also sit above the market average, with subprocessors named one by one alongside their processing location, a DPA supplied on request, and an identifier filter applied before anything reaches the AI.
The reservations are real and worth weighing. No price is published anywhere and there is no free trial, so an evaluation starts with a sales conversation rather than with the product. The site and the product are Croatian only. No API and no integration is announced, so data moves as files. The generative AI analysis runs at OpenAI in the United States, with an EU migration described only as being in preparation. No security certification is claimed, and no customer is named. The LinkedIn link published across the entire site leads to a page that does not exist - a small thing, but the kind of small thing that marks a young operation still tidying up after itself.
For a Croatian auditor, accountant, controller or tax adviser, FynTwin is worth the conversation. For anyone reporting outside the Croatian framework, or expecting to sign up and start alone, it is not the tool.
- Choosing a selection results in a full page refresh.
- Opens in a new window.