Stéphane Donzé is the Founder and CEO of AODocs, with more than 20 years of experience in the enterprise content management industry.
The release of ChatGPT a few years ago felt like a game changer in applying generative AI (GenAI) to large-scale enterprise content management. It opened up new possibilities for categorizing and processing information. But no less transformative, though less publicized, is how AI processing costs have dropped dramatically—up to 1000x since the release of GPT-3. Once a steeply priced luxury, limited to selected data, it’s now an affordable commodity, enabling businesses to apply AI across much more of their enterprise content.
And so the buzz around GenAI in DMS may be justified sooner than you think. Companies could leverage the technology’s diminishing price tag to complete complex tasks faster and boost productivity.
Here’s how these breakthroughs can translate into real-world impact.
From Prohibitive To Profitable
Two years ago, you would have processed this RFP using the first version of GPT-3 dated July 2022. Assuming you were to use 400 GenAI “tokens” per page, it would have cost you about 2.4 USD in AI processing fees. Moreover, to process the RFP, you’d have to split it into multiple pieces and feed GPT-3 with separate prompts. That was due to GPT-3 only being able to process about five pages simultaneously.
Fast-forward (only) two-and-a-half years to today’s technological landscape. Nowadays, you can get the same processing quality as GPT-3 with a small LLM, like the 3 billion parameters version of LLama 3.2, which costs around $0.05 per million tokens. This translates into processing your 100-page RFP for $0.002. That’s 1,000x cheaper than the 2022 GPT-3!
Moreover, you can cover the entire 100-page document in one single prompt. That’s because the LLM’s context size, or the number of tokens a language model can process at once, is much more significant nowadays.
There you have it: Using GenAI is a complete game changer in tackling complex content and 1,000 times cheaper in processing costs.
Looking at a single RFP, one dollar or one cent doesn’t make much of a difference in a company’s bottom line.
Fair enough. But what if you have millions or billions of documents to process? Many companies have faced the following challenge: A team needs to categorize every file out of a pool of millions as “confidential” or “non-confidential” based on the information they contain.
Let me grab my calculator once more.
If you wanted to process a million documents of 10 pages each, on average, with the 2022 version of GPT-3, you’d have to shell out $240,000 in AI processing costs.
No way. It just wasn’t affordable to apply AI at such a scope.
With LLama 3.2, you can process the same set of documents for about $200, which is 1,000x cheaper. Other small models, like GPT-4o mini or Gemini 1.5 flash, will also cost you a few hundred dollars per million documents. The cost savings are staggering, making it finally possible to apply GenAI at scale.
The Full Picture
It may seem like a no-brainer. But should you drop everything—including staff members—and go on a GenAI shopping spree? Reports suggest some major tech players, such as Salesforce, may opt for hiring freezes. And “78 percent of hiring managers anticipate that AI will lead to layoffs of recent college graduates at their company within the next year,” a survey found.
Yet GenAI’s promise is not in ditching human talent but exactly the opposite: clearing time for employees to focus on high-value tasks. Overcoming skepticism and resistance about human agency—which are understandable—requires clear planning, honest engagement with stakeholders and commitment to reskilling.
Implementing GenAI for DMS also introduces technological obstacles.
First, there’s a limit to what “narrow,” older AI models can do. Those excel at processing structured data you can find in a PDF or a spreadsheet. But when deployed on more complex formats, like images or design plans, “classic” tools might falter, leading to misinterpretations, data loss or costly errors. Modern GenAI better handles unfamiliar formats.
Security and permissions could be another blind spot. Without robust control and auditing, organizations risk over-relying on autonomous tools. In one case, misconfigured access permissions led to confidential files popping up in Microsoft’s AI Copilot search results. The researchers who’ve exposed this vulnerability titled their paper“ConfusedPilot.” Bloggers described a nightmare scenario with “employees casually stumbling upon their CEO’s emails or private HR documents.”
The lessons? Rigorously test security systems and access rights before you deploy your shiny new AI. And invest in the collaboration between IT, developers and business units.
As AI becomes integral to document management, several best practices can help organizations maximize benefits while minimizing risks:
Document Accuracy: Many organizations mistakenly assume AI will always pull the most relevant documents, which can lead to costly errors—such as outdated pricing or contract details. Businesses should implement a DMS with strict versioning and approval processes to ensure AI only accesses the most up-to-date, verified documents.
Workflow Integration: Integrating AI into existing workflows is another major challenge. Take a phased approach by starting with simple tasks like document sorting and data extraction. As the system matures, you can scale. AI must complement human oversight rather than replace it, empowering employees to handle exceptions and ensure quality control.
Security Protocols: AI systems must be configured to safeguard sensitive data with robust access controls and audit trails. This ensures that only authorized personnel can modify critical information and prevents security breaches.
The ‘And With’ Age
Thinking that you couldn’t apply GenAI to up to billions of documents without getting an emergency cash injection for your business is no longer relevant. We’re experiencing a profound shift in favor of applying GenAI tools for enterprise document management at scale in ways that were unthinkable even a few months ago.
It’s not just about cutting costs or getting more quality work done in less time. The “either/or” dichotomy in DMS is gone; we’re now in the “and/with” age.
Companies capitalizing on such opportunities might be best placed to unleash new growth and innovation potential.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
