Stéphan Donzé, Founder And Ceo At Aodocs – Interview Series

Stéphan Donzé is nan laminitis and CEO of AODocs, a cloud-native archive guidance level that transforms endeavor contented into actionable intelligence. Unlike bequest systems constricted to basal storage, AODocs combines robust archive power pinch workflow automation, enabling businesses to streamline analyzable processes crossed industries.

From income proposals and claims guidance to engineering and manufacturing, AODocs uses generative AI to extract system information from unstructured documents—at scale.

What inspired you to commencement AODocs successful 2012? Was location a circumstantial problem you saw successful archive guidance that wasn't being addressed astatine nan time?

It came from nan conception of bringing Business-to-Consumer (B2C) technologies, pioneered by Google, to nan enterprise. This included cloud-based, serverless, and automatically scaling technologies that were absent successful endeavor package astatine nan time.

At nan time, Google was advocating for applications to accommodate to nan unreality by moving distant from heavy clients and embracing browser-based and serverless architectures. Unlike Microsoft and Amazon astatine nan time, Google’s stance was that unreality applications should adopt a very different architecture than connected premise software. I saw an opportunity to create thing caller by combining aged and caller approaches.

Early adopters of unreality technologies successful enterprises, who had adopted Gmail and Google Drive, wanted to widen these benefits beyond email and collaboration to business-critical documents. There was a deficiency of solutions successful nan marketplace to meet this need, arsenic accepted archive guidance systems were on-premise pinch aged package and architectures. This spread presented an opportunity to build thing caller from scratch connected nan correct unreality architecture, moving pinch these early adopters.

Many endeavor SaaS companies struggle pinch customer adoption. How did you person awesome enterprises for illustration Google aliases Veolia to spot AODocs early on?

Allowing customers to support documents successful their  ain unreality retention was a important facet successful convincing enterprises. AODocs manages documents, including sharing permissions and business processes, but nan documents stay wrong nan customer's situation (initially Google Drive, later Azure, Google Cloud Storage, and Amazon S3). This basal characteristic reduced nan perceived consequence for customers, and made them much comfortable moving with  a mini and young company.

Another cardinal logic was adjacent customer relationships. We co-built and co-developed pinch customers, particularly successful nan early stages. This proximity and reactivity successful fixing problems and making improvements built spot and further reduced nan cognition of risk.

This was yet what convinced companies for illustration Google, Veolia, Ascension, Pinnacol Assurance and others to adopt us.

Looking backmost astatine your 13+ years starring AODocs, what were immoderate of nan astir captious decisions that led to nan company's success?

The determination to support customer documents successful their ain unreality retention is simply a large facet successful our occurrence and remains a cardinal differentiating factor.

Developing nan merchandise by staying very adjacent to customers was besides critical. Everything we build is inspired by aliases validated done customer needs. We debar building features without a clear customer usage case.

Our expertise to benignant done customer requests to place communal and replicable needs that use nan full customer base, arsenic opposed to very circumstantial individual requests amended addressed by master services, is thing we've been bully at.

AODocs was calved successful nan unreality and has now evolved into an AI-powered solution. What were immoderate of nan biggest challenges successful making that transition?

It's not truthful overmuch a modulation but a continuity. Being calved successful nan unreality inherently made AODocs easy to merge pinch new, cloud-based technologies, including AI.

AI models for illustration ChatGPT and Google Gemini are cloud-based, making it earthy for AODocs to interface pinch them and adhd AI capabilities to our archive processing platform.

We position AODocs arsenic a “spine” aliases an “assembly line” for archive processes, wherever AI is simply a caller “limb” aliases “module” that tin beryllium plugged successful to heighten nan platform. We did not foresee nan circumstantial emergence of AI successful 2022, but our cloud-native architecture was designed for extensibility and integration, making nan take of AI a earthy step.

How does AODocs equilibrium automation pinch quality oversight, ensuring compliance and accuracy without removing quality validation?

We supply nan devices to customers to usage AI astatine various steps successful their archive processes. The determination of really overmuch autonomy to springiness AI is yet made by nan customers connected a case-by-case basis.

Customers tin configure their processes for afloat automation, afloat quality review, aliases a operation of some astatine immoderate AI step. We urge identifying easy cases suitable for afloat AI automation (e.g., detecting missing documents successful a request), while for analyzable cases, AI tin beryllium utilized to velocity up nan quality review, typically by generating summaries of analyzable documents.

Essentially, we springiness customers 100% power complete nan level of quality oversight successful their processes by providing nan configuration tools.

AODocs is shifting its level to attraction connected AI agents for archive processing. What drove this decision, and really do AI agents amended upon accepted archive guidance systems?

AI agents are a caller capacity that tin beryllium integrated into our existing “assembly line” for documents.

The superior driver for our excitement astir AI agents is their imaginable to bring important velocity to circumstantial tasks. For elemental tasks for illustration rejecting incomplete documents aliases automatically accepting debased value, straightforward claims, AI tin alteration afloat automation.

For analyzable tasks, AI agents tin dramatically accelerate processing by summarizing agelong documents, allowing humans to reappraisal cardinal accusation overmuch faster. AI excels astatine ingesting ample amounts of accusation and providing concise summaries, redeeming quality reviewers important time. This accrued velocity and ratio heighten nan worth of nan AODocs platform.

Enterprise hunt remains a awesome challenge, pinch companies struggling to retrieve applicable documents. How does AODocs' AI-powered solution tackle discoverability issues amended than existing endeavor hunt tools?

AI chatbots correspond a awesome betterment complete accepted endeavor hunt by allowing users to inquire questions successful earthy connection and person straightforward answers. However, they person a important flaw: they cannot separate betwixt valid, up-to-date documents and drafts, incomplete, aliases obsolete files.

We reside this situation by ensuring AI agents usage only nan correct accusation to supply answers. Traditional hunt presents users pinch a database of imaginable matches, allowing them to place nan astir applicable archive (such arsenic distinguishing betwixt a signed statement and a draft).

In contrast, AI chatbots connection assured answers but hide nan underlying hunt process. This creates an accrued consequence of utilizing outdated aliases incorrect information, arsenic users person nary visibility into which documents nan chatbot utilized to formulate its response.

Our domiciled is to curate and guarantee that each documents successful nan knowledge guidelines person been decently validated and are up-to-date. This makes them safe for AI agents to usage arsenic accusation sources. By ensuring AI-powered hunt relies only connected trusted and controlled information, we importantly trim nan consequence of inaccurate responses.

With generative AI becoming a cardinal portion of archive processing, what circumstantial AI technologies aliases models does AODocs leverage to amended accuracy and trim manual work?

LLM models excel astatine transforming existing content, specified arsenic summarizing analyzable documents, translating text, and categorizing files. These tasks person a very debased consequence of hallucinations, which is why our AI usage strategically focuses connected these activities.

AODocs chiefly leverages AI for rephrasing, repurposing, and transforming existing matter alternatively than creating caller content. When moving pinch nan correct material, generating summaries, translating documents, and categorizing contented each support a debased probability of error. Similarly, our information extraction capabilities—identifying owed dates aliases cardinal accusation from documents—remain grounded successful existing contented to minimize inaccuracies aliases “hallucinations.”

By accelerating accusation ingestion and supporting quality reviewers, these AI capabilities often amended ratio by a important factor. This creates tangible worth while cautiously avoiding nan risks associated pinch AI-generated contented created from scratch.

There’s an further advantage. AODocs integrates seamlessly pinch immoderate LLM, giving our customers state to take their preferred provider. This elastic architecture besides ensures we tin quickly incorporated early models arsenic they go available, positioning our customers to use from nan inevitable AI innovations emerging successful nan months and years ahead.

One of nan biggest AI challenges successful archive processing is hallucinations aliases inaccurate responses. How does AODocs guarantee that AI-generated insights are trustworthy and reliable?

The cardinal to ensuring trustworthiness and reliability is grounding. We guarantee that AI is utilized to rephrase aliases toggle shape accusation from trusted sources.

In archive processing, AI is asked to summarize aliases construe existing documents, minimizing nan chance of hallucination. In AI chatbots aliases agents, nan strategy is provided pinch extracts of controlled documents and is instructed to make answers only from those extracts—the LLM must ever quote nan sources from which it generated its answer—without utilizing wide world knowledge aliases outer information. This prevents AI from inventing aliases providing accusation not explicitly coming successful nan validated documents.

With AI transforming endeavor software, wherever do you spot AODocs successful nan adjacent 5 years? What domiciled will AI agents play successful nan early of archive management?

Our halfway business arsenic a archive guidance level will not change. Our ngo remains to guarantee nan trustworthiness, currency, due business process adherence, and traceability of managed information.

AI is some a caller request for controlled accusation (to provender trustworthy information to AI chatbots) and an accelerator of business processes involving documents. AI tin velocity up tasks for illustration summarizing, translating, and categorizing documents, making quality users much productive.

In nan adjacent 5 years, we purpose to supply nan devices for companies to easy use AI wherever it makes consciousness to heighten their business processes. AI agents will play a important domiciled by amplifying and accelerating our halfway missions: providing trustworthy information, ensuring nan usage of nan correct archive versions, helping debar quality errors, and facilitating entree to business processes. We want to thief companies usage AI efficaciously and strategically to boost productivity.

Thank you for nan awesome interview, readers who wish to study much should sojourn AODocs.