Let’s say we’re an AI company, but we focus on documents, not data. That’s changing over time, but our focus has always been the information hidden inside the vast corpus of documents people write and then forget they wrote.

It’s information to knowledge. And of course there is this well-known data, information, knowledge, and wisdom pyramid. So we’re working our way from information to knowledge, and eventually to wisdom.

That it helps them find information lost in their file systems, and it does it easily and quickly. We’ve had customers complete tasks in 30 minutes that could take them a month using old technology. We have built in taxonomies and knowledge bases with over 2 million clues that consistently tag and manage information across the enterprise.  The i2k AI platform quickly and consistently classifies content and processes it in place so that users can easily find information through a GIS map, a faceted search interface, and a chat window that work together.

Our focus has always been on leveraging domain knowledge to figure out what’s in documents based on curated, automatically learned knowledge, not on statistically measuring whatever the system was trained on.

That’s a good question. We used to think the value of a system like ours was that it helped you find the answers faster. Today, the value is that it provides the answers by eliminating the need to read a collection of documents and ferret out the information you want. Our technology can go straight to the answer you’re looking for and provide it in a useful way with citations to the original sources, so you can confirm it’s correct.

The energy industry runs on a lot of content, reporting, and interpretation of data from measurements, processes, and operations. That data isn’t useful until domain experts interpret and verify it. And there’s so much of it. Our industry can’t always leverage the data for future work because, over time, it can be lost without a way to find it and determine whether it’s relevant to another problem. It’s a tax on the business if they can’t reuse the effort spent making those reports and distilling the information, knowledge, and wisdom in them.

Definitely an enhancement. If you’re not using AI, somebody else is, and they’re going to do a better, faster, more complete, more confident, and verifiable job.

The energy industry has been tarred with the brush of being old-fashioned for years. In fact, it’s one of the most high-tech businesses there is. I’ve been in the business since the early ’80s, and I’ve seen it change from being software-technology-averse and very heavily focused on iron, to being extremely advanced, data-driven, automated, and integrated with new technology.

And i2k is a company made up of professionals with over 200 years of oil and gas experience. Domain knowledge helps us work with customers to find solutions quickly, prove the technology’s value in small experiments, and then scale to very large projects.

With our AI technology, there’s a dramatic shift in what every individual can do today. You can now get an analysis, graphs, and data files processed quickly and easily.

With generic AI, people want to know why this stuff is so expensive. Everybody’s talking about how much they spend on tokens.

When our team meets with companies, they find they spend a lot on tokens and the time finding the information that feeds their processes. You’ve got maybe terabytes or petabytes of information stored from decades past, and you would like to use it. But if you simply task Claude to go “please find this for me,” it just won’t. It’ll take forever. It knows how to write code and make graphs, but it doesn’t know your structure or how to find things in your environment.

Our platform has been doing that for more than a decade now. And by plugging it in and making it available as a tool to AI agents, you’re getting insights from your entire corporate knowledge base, and not just a few files you can feed it one at a time.

The i2k AI Platform forms a symbiotic relationship with generic AI tools because it provides the right data and information. It’s like fodder for the generic agents. You can’t give generic AI tools all the data at once because there are terabytes of data. The i2k platform finds data relevant to the task.

We have a completely offline version that lets you bring in knowledge from an external source onto your laptop, then disconnect and do important work locally. It’s effectively edge, right?

For the phone, we are building solutions that work on very small models. Today it’s still LLM-based for our chat product, as anybody’s would be, but it can use records and archives processed ahead of time and downloaded onto the device.

What we’ve learned from working with customers is that GenAI is becoming standard for modern workflows, but it’s hard to get the most power from it because it is expensive and often slow. With our platform, domain expertise is built in, so we can pre-process and prepare your data from your corporate archives and new workflows to make your agents, chat solutions, and everything else as capable as possible.

The challenge will still be how to harvest the knowledge that has been stored and never really surfaced. Gen AI won’t automatically know what to do with your data unless you describe it in full detail. So you will either spend your time sifting through your corporate archive to figure out what you’ve got and how to use it, or you can use i2k technology built for that purpose.

In the future, what won’t change is that you’ll benefit from bringing in the knowledge of your corporation’s past, and the experiences you’ve had and documented but aren’t being surfaced.

What I personally enjoy is that we’re a team of problem solvers. We’re not trying to sell the same old solution for years and years and years, just hoping people get locked in. We’re always watching what the actual need is, and we know customers need solutions that actually work. I know what we’re building and providing is increasing their velocity and ROI, and we can provide evidence of that. Typical cost savings of 70% to 80% on large language model tokens when processing documents across the enterprise are a significant benefit.

A couple of things.

Number one is the people. We have a great group of smart, customer-focused people.

The second thing is we get pulled into some really interesting challenges, things that other people haven’t necessarily solved, where customers are having issues. For example, they’re using AI but finding the costs are too high, not finding the content they need, and that a solution that works for five documents doesn’t work for millions of documents.

I want prospective customers to know the i2k AI Platform solves real-world problems.

I have to be careful how I answer because we work in industries where people might be able to identify a company that considers our solution part of their competitive advantage. Here are some published examples.

The i2k AI Platform powers the SPE Research Portal for over 100,000 upstream oil and gas professionals. We also embedded our technology as the AI layer for slb’s DELFI Cognitive E&P Environment, enabling shared workspaces for data, models, and interpretations. Our Oilfield Places® capability recognizes location references and auto-classifies structured and unstructured content using a geopolitical/geological taxonomy. The platform consolidated millions of documents from siloed databases when two major oil companies merged, and during a divestiture it split a multi-million-document library while finding and removing licensed data, including images and material embedded in reports and slide decks. Outside energy, it serves as an enterprise layer for contract management and for navigating ESG and compliance documentation.

Our Oil & Gas Agentic Chat™ capability can serve as a virtual member of the organization’s expert community, delivering the right level of support and insight to managers, geologists, and engineers when needed. Accessing historical data from multiple disparate sources frees up more high‑value geoscience and engineering time for analysis and decision-making.

We see our platform contributing more on the information-finding side, ensuring decisions are based on all available information. Finding all the information de-risks decisions by preventing you from missing critical information that would have prompted a different decision.

We also complement generic AI tools by accessing information very rapidly. Quickly identifying where all relevant documents are is more efficient and effective than relying on institutional knowledge to know where the right documents might be.

We have a fully vertically integrated solution, and our platform is also callable by MCP.

MCP, or Model Context Protocol, is an open standard originally proposed by Anthropic that lets AI models and applications plug into external tools and data sources through one common interface, so any model can call any compliant tool without custom integration work for each pairing.

For example, it’s what lets Claude call the i2k AI Platform to find oil and gas documents, and lets the i2k AI Platform call a reservoir simulation tool.

Let’s say you have a Claude chat agent, and that’s what you want to deploy. With MCP, you can use Claude to reach out to our platform to find oil and gas documents it wouldn’t be very good at finding. Claude has some form of search, but certainly not within your internal corpus of documents.

We can also go the other way because our platform can access data output from a reservoir simulation technology that has MCP.

Our tools are very specialized tools for oil and gas because, for example, they understand where you’re going to find measured depth and deviation information to create a well path plot or to plot a production history and updated forecast. The tools are callable, and the platform understands how you use them to get an answer that makes sense to a geologist, geophysicist, or petroleum engineer.

Agentic chat is more than asking a question and getting an answer because it lets you pull in specialized tools that let the LLM decide what will solve the problem and answer the questions. At its core, you’re using a reasoning model that allows the LLM to become a little bit closer to being a peer, or maybe a more junior staff member, that actually can reason across, “How would I go about getting that answer?” Agentic chat can use tools to give detailed answers or create a summary answer if it anticipates you need one, even if you didn’t specifically ask for a summary.

The basic approach is similar, but our tools are oil- and gas- and geology-specific. While other AI systems and agentic chat systems can plot, our Oil & Gas Agentic Chat tool understands that when you’re creating a cross-section, it needs to calculate TVD because using measured depth in a horizontal well won’t give you an accurate answer.

And so if you’re creating a cross-section, it’s got to understand what that looks like, create something that actually makes sense to a geologist, be able to take the wells and plot them on a map and say, “Okay, well, if I draw a cross-section between these two wells, what’s the projection of these other wells?” These are applications you can ask ChatGPT or Claude about, but they still require too much domain knowledge to answer accurately.

The second-order effect on the business will depend on the application. One operator was taking 40-plus hours to find the information they needed before they could even start the analysis to determine, “Is this an opportunity we want to go explore?” And by explore, I don’t mean drilling exploration. The operator needed a commercial and financial analysis, and our platform took it down to about 40 minutes.

We believe the initial return would definitely be north of 50%, and it would improve over time depending on the type and amount of data you have. You can begin using the system from day one because you don’t have to ingest all of your data. You’re not changing one system to another because our system doesn’t move any of your documents. We find them where they exist.

People will use the system and say, “Hey, I could do that. I found that. I was able to do this analysis quickly, and I got it to do that.” From the start, users get real answers relatively quickly, even if you haven’t finished the implementation project. That is a key difference between most IT projects and using a system like ours. And it’s quite a fundamental change.