Antons Brains Bārdiņš — ABB

Founder & CEO, ObsidianBK. Building an EU-focused ERP and accounting automation platform — plus EasyWallet fintech. ↓ Portfolio

[Antons Brains Bārdiņš — ABB @ABB_EZ]

@Hexagon_Money Le risque TPE n’est pas seulement le format. Si l’entité, le centre de coût ou l’approbateur manque à l’arrivée, on digitalise l’erreur. Il faut une file d’exceptions et une piste de correction. Avant septembre, testeriez-vous 20 factures réelles de bout en bout ?

[Antons Brains Bārdiņš — ABB @ABB_EZ]

Accounting AI loses trust when one invoice gets different account suggestions.

Today I tested an ObsidianBK fix: a human correction now replaces the older supplier rule. One active rule, version history, audit trail.

What inconsistency bothers you most?

https://t.co/9Ef2ACAaqc

[Antons Brains Bārdiņš — ABB @ABB_EZ]

@DigitalEU For accounting AI, transparency should go beyond a label. Users need to see which fields came from extraction, the confidence, every human correction and the final approver. That audit trail turns “AI-assisted” from a black box into a reviewable workflow.

[Antons Brains Bārdiņš — ABB @ABB_EZ]

@DigitalEU @EU_Commission For SMEs, a useful Digital Decade metric is not only adoption, but administrative friction removed: fewer duplicate invoice entries, fewer documents chased across inboxes, and a clearer human review trail. Are workflow-level outcomes part of the programme review?

[Antons Brains Bārdiņš — ABB @ABB_EZ]

@DigitalEU Transparency matters where AI touches regulated records. In SME finance, can a user see the source document, suggestion, corrections and human approval before posting? EU policy becomes practical when small teams can build that evidence trail without enterprise budgets.

[Antons Brains Bārdiņš — ABB @ABB_EZ]

Accounting AI should never silently turn a confidence score into a journal entry.

A useful loop is: source document → suggestion → human review → correction → auditable posting.

Faster is good. Defensible is better. What exception costs your team the most time?

[Antons Brains Bārdiņš — ABB @ABB_EZ]

@BojanRadojici10 The bottleneck in accounting is trust, not generation. I’m most curious about exception review: source document → suggested entry → human correction → auditable posting. Which of the 100 have teams kept using after month one, rather than demoed once?

[Antons Brains Bārdiņš — ABB @ABB_EZ]

@jasonlk This is where the one-person-company idea gets dangerous. Building is compressed; support, accounting controls, security and recovery are not. The durable advantage is a narrow promise you can still support at 2am. Which function do solo founders underestimate most?

[Antons Brains Bārdiņš — ABB @ABB_EZ]

Uploading the same invoice twice should not create two accounting realities. If AI changes its suggestion, show why and route uncertainty to review. Tenant context, consistency and audit trail matter more than a confident-looking answer. Which field causes the most corrections?

[Antons Brains Bārdiņš — ABB @ABB_EZ]

@SolutionFiscal The receiving mandate is the hard part: entity data and approval ownership need to be clean before day one, not after invoices begin arriving. For SMEs preparing now, which gap appears most often—directory connectivity, field mapping, or internal review?

[Antons Brains Bārdiņš — ABB @ABB_EZ]

Accounting automation starts too late. The first bottleneck is missing context: a PDF in chat, supplier mismatches, or the wrong legal entity. At ObsidianBK, AI suggests; a person reviews before posting. EU accountants: which exception creates rework?

https://t.co/9Ef2ACAaqc

[Antons Brains Bārdiņš — ABB @ABB_EZ]

EU accountants: when an owner forwards an invoice, which missing detail creates most rework? A) counterparty identity B) business purpose C) VAT treatment D) payment match E) approval. We are shaping onboarding around evidence, not guessing. Latvia-first; EU discovery open.

[Antons Brains Bārdiņš — ABB @ABB_EZ]

@Techstars In Latvia, the gap is not only capital; it is repeated founder–customer contact that makes learning visible. Communities built around useful introductions compound faster than those built around events alone. What ritual have you seen turn an audience into community?

[Antons Brains Bārdiņš — ABB @ABB_EZ]

@EEN_EU @samsuntso @EU_EISMEA @EU_Growth @BlackseaEen Useful proof that partner access can matter more than generic visibility. For a Latvia-based B2B software startup validating accounting workflows across the EU, which EEN entry point would you recommend first: local diagnostic, matchmaking or partner search?

[Antons Brains Bārdiņš — ABB @ABB_EZ]

@LIAALatvija Labi redzēt lielo projektu koordināciju vienā līmenī. Būtu vērtīgi regulāri publicēt arī īsu statusu: kas konkrēti ir nākamais solis, termiņš un atbildīgā puse. Tas palīdzētu arī mazākiem Latvijas uzņēmumiem saprast, kur veidojas piegāžu un sadarbības iespējas.

[Antons Brains Bārdiņš — ABB @ABB_EZ]

Most accounting AI demos begin after the data is clean. Real SME work begins earlier: the invoice is in a chat, context is missing, and the accountant has to chase it. ObsidianBK is testing the opposite flow: Telegram → review queue → human approval. https://t.co/9Ef2ACAaqc

[Antons Brains Bārdiņš — ABB @ABB_EZ]

@DigitalEU Secure scaling needs an SME path, not only frontier infrastructure. In financial workflows, accountability means visible evidence, a named reviewer and an audit trail. How will the action plan help small firms test systems safely before connecting them to real records?

[Antons Brains Bārdiņš — ABB @ABB_EZ]

@BojanRadojici10 Coverage is not safe adoption. I’d score each finance use case by reversibility, evidence quality and correction cost—not only time saved. Drafting with a named reviewer is a different risk class from autonomous posting. Which use case had the strongest production evidence?

[Antons Brains Bārdiņš — ABB @ABB_EZ]

Accounting AI should know when to stop.

The same invoice can need different treatment depending on buyer identity, purpose, VAT setup and history.

Our rule at ObsidianBK: extract → preview → human review → post. No silent posting.

https://t.co/9Ef2ACAaqc

[Antons Brains Bārdiņš — ABB @ABB_EZ]

@BojanRadojici10 Coverage is not readiness. In accounting, I separate draft/summarize, deterministic calculation and actions that change the ledger. That last category needs provenance, confidence thresholds and human approval. Which use cases have survived a real close with an audit trail?