ABBYY

"The weakest link in your AI pipeline is almost never the model. It is the document."
Here's the problem almost no one is talking about: the damage happens before your model reads a single word.

When a typical PDF passes through raw extraction, things quietly fall apart:
> Reading order collapses
> Tables flatten into meaningless strings
> Multi-column layouts interleave into noise
> Structural hierarchies disappear

By the time your AI sees the document, it's already corrupted. The model isn't learning your domain knowledge. It's learning the noise.

ABBYY research puts a number on it: when document parsing accuracy drops below 70% to 80%, most downstream natural language tasks become unreliable. That's where higher hallucination rates and weaker RAG retrieval come from.

So here's the shift in thinking. Structure-preserving extraction isn't a nice-to-have enhancement. It's a prerequisite for reliable enterprise AI.
You can invest in better prompts and bigger models all day. If the input is broken, the output will be too.

Maxime Vermeir, VP of AI Strategy at ABBYY, unpacks why your documents are corrupting before AI ever reads them, and what to do about it.

Is your AI failing because of your model, or because of your data foundation? Read the article: https://hubs.li/Q04yc_lX0

For every $1 of fraud loss, financial services absorbs $5.75 in total costs. And the documents generating that loss are becoming harder to distinguish from legitimate ones.

GenAI has changed what a forgery looks like. Resistant AI 's forensics catches tampering the human eye cannot detect, and Naviant brings deep Know Your Customer (KYC) expertise to help institutions modernize onboarding and compliance workflows. ABBYY has spent 35 years converting unstructured documents into trusted data and this October, we're bringing both capabilities to the table.

ABBYY Circles: Two invitation-only sessions for senior leaders in banking, insurance, and fintech:

📍 Charlotte · Wednesday, Oct. 7, 2026 · 12:30–3:30 PM · Capital Grille
📍 Toronto · Thursday, Oct. 8, 2026 · 4:00–7:00 PM · Riley's Toronto

These aren't presentations. They're working sessions. Walk in with your fraud and KYC challenges; walk out with a live forensics demo, a pKYC readiness assessment against AMLR and FATF, and a scored automation shortlist with potential project value exceeding $1M.

Select your city to request an invite:
Charlotte → https://hubs.li/Q04y6cWl0

Toronto → https://hubs.li/Q04y6cQr0

Which table will we see you at?

Can a document lie to you? In the age of GenAI, the answer is yes, and the fakes are getting harder to spot.

At ASPIRE Poland 2026: Man vs Machine, ABBYY and our partner Foresight are tackling exactly that challenge.

Join our session:

📌 Documents that lie, and how to catch them
📅 23 September 2026, 11:00–11:45
📍 The Old Tram Depot (Stara Zajezdnia), Kraków, Poland

Hear from our speakers Alina Sysko, Andrzej Błaszczyk, and Igor Ołdakowski as they show you:

• How fraudulent and manipulated documents slip past manual review
• What machines catch that the human eye cannot
• Where intelligent automation strengthens trust in your data

? When it comes to document fraud, the smartest teams put both to work.

Save your spot and see how to catch the documents that lie 👉 https://hubs.li/Q04xXC4c0

See you in Kraków✈️.

Enterprise software has spent decades designing for one user: people. What happens when the next user is an AI agent?

AI agents have become legitimate users of enterprise systems by retrieving information, orchestrating workflows, and triggering actions, all without a person standing by to compensate for the software's gaps.

Slavena Hristova, ABBYY Director of Document AI Strategy, asks what these changes mean for enterprise software usability, and offers documents as the proof point.

A person can glance at an invoice and infer meaning from layout alone. An AI agent needs that same signal made explicit, which is why document understanding is becoming the perception layer for enterprise AI.

Here's the part that surprises people. More AI doesn't mean less human input. It moves human expertise upstream. People stop executing routine work and start designing the rules that shape how work gets done.

Is your enterprise software ready for its second user?

Read the article to find out: https://hubs.li/Q04xXw3m0

Tomorrow, we go live. Are you in?

"You deployed FlexiCapture successfully. Now assure it." Join ABBYY and Reveille Software for a live session built for FlexiCapture customers.

Here's the incentive worth showing up for: attend live and you qualify for a complimentary Observability Assessment. Your FlexiCapture environment, scored against 90+ prebuilt tests.

That's real, actionable proof of how your environment is performing. Not a guess.

📅 September 23, 2026
🕙 10:00 am PT / 1:00 pm ET

Wayne Ford and Kieran Fox run it live on a real FlexiCapture environment.

Don't leave assurance to chance. Save your spot 👉 https://hubs.li/Q04xMhWf0

Singapore, that’s a wrap! 🇸🇬

ABBYY Ascend Singapore brought together AI leaders, technologists, customers, and partners for a day of conversations around one key theme: turning complexity into clarity, and AI potential into real-world results.

From exploring the role of purpose-built Document AI in agentic automation to sharing practical approaches for making GenAI production-ready, the day was packed with ideas and insights.

🔹 New perspectives on building safe, scalable AI with standardized document intelligence
🔹 Plenty of opportunities to connect, exchange ideas, and explore what’s next for enterprise AI

Thank you to everyone who joined us in Singapore for meaningful conversations and connections.

Want to keep exploring what’s possible with ABBYY’s purpose-built AI solutions? Share a bit about your project: https://hubs.li/Q04xSwpD0

What happens when AI moves from recommending an action to taking it?

Uber’s US$966 million fine is a stark reminder that AI governance can’t be treated as a compliance exercise after deployment.

In a new Intelligent CIO article, Maxime Vermeir, Vice President of AI Strategy at ABBYY, explores what the case means for organisations deploying agentic AI — and why meaningful human oversight needs to be designed into systems from the start.

The key question isn’t simply “What can this system automate?” but:

“What decisions should this system never be allowed to make unilaterally?”

Read the full article: https://hubs.li/Q04xNKdc0

On September 23rd, ABBYY and Reveille Software will host a webinar built specifically for FlexiCapture customers.

On a real ABBYY FlexiCapture environment, you'll see exactly what assurance looks like:

• End-to-end visibility across ingestion, batches, queues, and licenses
• Self-healing in action: a stalled process caught and restored before anyone opens a ticket
• An SLA report card on the processes your business actually depends on

This is the difference between hoping your environment is healthy and knowing it is.

Join us on 9/23 at 10:00 am PT / 1:00 pm ET
Save your spot 👉 https://hubs.li/Q04xMfx30