Innodata Inc.

programs with this customer we announced less than two weeks ago on April 24. We expect that these programs will ramp up over the next two months. While our customer agreements typically contain early termination upon notice provisions, we believe this customer is committed to a significant multi-year LLM strategy from which we stand to benefit. In fact, we are in discussions with this customer regarding potential new programs and expansions beyond what we have announced so far. We also signed two additional big tech companies, a large prominent generative AI company and a large prominent consumer facing ISV investing substantially in generative AI foundation models. As a result of these new wins, we now serve seven big tech customers. We believe we will continue to grow with these customer relationships in 2024 and that we may grow some of them, possibly quite substantially. For big tech customers, we provide a broad range of services to support their generative AI programs. This includes creating instruction data sets, which you can think of as the programming behind large language models. It also includes human preference data used in reinforcement learning and reward modeling to align models to human preferences and build guardrails against toxic biases and harmful responses. In a blog post last month that accompanied a major release, one of our large big tech customers stated that the quality of these instruction data sets has an outsized influence on the performance of their models and that some of their biggest improvements in model quality come from carefully crafted instruction data sets and multiple rounds of quality assurance. This statement crystallizes why we have become the partner of choice for such customers. We believe we are well positioned to anticipate big tech's changing needs and to grow with them. It is evident that big text aspirations extend beyond today's predominantly text-to-text English language models. We foresee expansion in terms of multimodal models, domain and task-specific models, models natively built in more than 30 different languages, and models capable of complex reasoning. All of these dimensions will require modeling with the kind of data that we create. We believe we are still in the early innings of this journey. I encourage you to read the latest quarterly earnings transcripts from the MAG7. Generative AI is a prevailing theme with promises of more Gen AI models, more Gen AI in products, and commitments to multi-year investment cycles and CapEx increases to support aggressive AI research and product development. We believe the emerging enterprise market which we call the middle layer, consisting of companies across verticals that seek to adopt generative AI technologies to be another important growth factor for InnoData, one that will ultimately dwarf the big tech market for us. In parallel with executing strategies to penetrate big tech, we're taking steps to prepare for what we foresee as a likely explosion in the enterprise space. We believe we are very well positioned due to our intimate knowledge of the GenAI roadmap of the large tech companies, which has enabled us to gain exceptional domain expertise in the future product needs of the enterprise market. We believe that enterprise adoption is about to enter a Cambrian period of explosive growth as a result primarily of three technology developments now underway. I'll explain and illustrate each with examples. Today, enterprise users are mostly using ChatGPT as a standalone application. We'll call this level zero use case. For example, if I'm an HR director at InnoData tasked with revising InnoData's employee handbook, I can prompt ChatGPT to write a first draft of the vacation policy. Companies are now shifting from this level zero to what we think of as level one. We think of level one systems as those based on retrieval augmented generation, or RAG, which we believe are likely to become better performing for reasons I will explain shortly. RAG systems couple search technology and prompt engineering. With such a level one system, an InnoData employee might prompt an InnoData HR chatbot with a request like, please summarize for me InnoData's vacation policy. A search engine working behind the scenes would then retrieve Vinodata's vacation policy from a large document repository and insert it into the prompt as context with an instruction to the LLM to answer the question primarily based on the inserted policy. RAG-based systems are about to become more useful as the latest crop of soon to be released LLMs offer significantly expanded context windows. A context window refers to the amount of retrieved information that can be included with a prompt. By including more context, the chatbot can become more consistent, relevant, and useful. One of the big tech companies is about to release a new model with a context window that is eight times larger than that of OpenAI's GPT-4 Turbo, enabling you to include, for example, 3,000 pages of documents in a single prompt. Today's expert or advanced expert augmentation systems are, for the most part, rag-based systems that combine generative AI with humans in the loop to deliver improved productivity. In a few minutes, I'll give you an example of such a system we started working on for a customer in the quarter. We believe a second technology development called agentic workflows will enable what we'll call level two systems. With an agentic system, rather than asking a question to a chat bot, you present a goal to a virtual agent. Your virtual agent then accesses multiple backend systems and LLMs talk to each other to accomplish your goal. Agentic workflows really open up the kinds of things you can ask computers to do with LLMs. With an agentic system, an InnoData employee might ask a virtual InnoData agent, please look up how many days off InnoData employees get. Check how many days off I have left and request a week off around my son's graduation so long as there are still available hotels in Boston. Imagine that. Now, while the full realization of agentic workflows may be years away, We believe incremental progress is being achieved and will likely accelerate. The third development that we believe will accelerate enterprise adoption is that the cost of training and serving models is likely to go down dramatically, making it possible for enterprises to train and serve models at scale. Once this happens, we believe that companies are likely to want to fine-tune their own models rather than relying on RAG-based architectures. We'll call these level three systems. Level three systems will support more complex use cases and enable sensitive information to be processed in private clouds or on premises rather than being served up as context to third party foundation models. We intend inundated to offer enterprise all of the services they require to navigate the journey from level zero to level three and beyond. This will include custom development, integration, and fine-tuning services, as well as managed services, services around data readiness and data governance, and industry-specific workflow platforms. We are not alone in our thinking that the enterprise market for generative AI is about to explode. In a report last year, Bloomberg estimated the market for generative AI focused IT services will grow to nearly $22 billion by 2027 and to nearly $86 billion by 2032, representing a 100 percent compounded annual growth rate for the 10-year period from 2022 to 2032. To position ourselves to drive enterprise growth, we are expanding our talent base, creating new accelerators, and winning new reference engagements. This quarter, One of the largest legal information companies in the world engaged us to develop a new LLM-based workflow system for their complex operational processes spanning legal and regulatory law in multiple European countries in multiple languages. This is an example of what I referred to a few minutes ago as an advanced level one system. Our implementation uses GPT. and a combination of several techniques including chain of density, prompt engineering, and a vector database with similarity matching. Our generative AI-enabled workflow system is expected to enable the customer to drive significant operational savings across high-cost processes that previously relied entirely on humans with language and legal expertise. This quarter, we also delivered a generative AI-powered tool that gathers on-the-fly insights from large-scale textual data, contextually analyzes the data for specific areas of interest, and performs language translations. The technology aims to increase organizations' efficiency by ensuring knowledge workers are equipped with the intelligence needed to make informed decisions. We built it into our Agility public relations platform, where we call it Intelligent Insights. We made intelligent insights generally available to agility customers in the quarter, and it has been well received. To build the solution, we utilized RAG-based prompt engineering. We recently demonstrated it as an accelerator to one of our large banking customers, and it inspired a POC that we are now executing. Agility revenue in the quarter increased 16.5% year over year. We have over 1,400 direct customers, and we're generating cash. We've been a leader in the industry rolling out cutting edge generative AI functionality that is bending the productivity curve for PR and communications professionals. We started early last year with the release of PR Copilot, a generative AI implementation that helps people write press releases and media pitches. In Q1, we announced the general availability of Intelligent Insights. We are planning another five significant generative AI feature releases over the course of the second half of this year and into the first quarter next year. As a result of what we believe is our generative AI leadership in the PR space, in the first quarter, We converted over 35% of demos to wins, up from less than 20% prior to implementation of our generative AI roadmap. Our customers told us that one of their biggest challenges was that they needed more hours in the day. With our generative AI innovations, we're making tactical PR a less labor-intensive process, giving our customers the time back they need for strategic thinking. For both the big tech market as well as the enterprise market, we see additional opportunities in model safety evaluation and responsible and ethical AI. We began working on trust and safety for one of our big tech customers in Q4 2023, providing model assessment and benchmarking services which help ensure that models meet performance, risk, and emerging regulatory requirements. We learned a lot from the work and development we've been doing on this engagement, so we decided to share our learnings, tools, and innovation with the market more broadly. Just a couple of weeks ago, we announced our release of an open-source LLM evaluation toolkit, together with a repository of 14 semi-synthetic and human-crafted evaluation datasets that enterprises can utilize for evaluating the safety of their large language models in the context of enterprise tasks. Using the toolkit and the data sets, data scientists can automatically test the safety of underlying LLMs across multiple harm categories simultaneously. Developers can understand how their AI systems respond to a variety of prompts and can identify remedial fine-tuning required to align the systems to the desired outcomes. We expect to release the commercial version of the toolkit and more extensive, continually updated benchmarking datasets later this year. In Q1, we won two additional engagements for LLM safety and evaluation, one for a hyperscaler's own foundation models and one for an enterprise customer of the hyperscaler through the white label program we have in place with the hyperscaler. In addition, in Q1 2024, We started pilots for a new customer and an existing customer around LLM trust and safety. I'll conclude with this. We believe we have an incredible opportunity in front of us. We believe we have the talent, capabilities, and scalability to support the world's leading companies' efforts to build AI models and services and to help enterprises advance AI and generative AI technologies. We believe we can drive best-in-class growth over the next several years and maintain our early leadership position in generative AI services. Moreover, we believe we can accomplish this without the need to raise equity, to incur debt, or to burn cash. This year, based on our current growth forecast, we intend to invest approximately 3.5 million in recruiting costs to scale our business and approximately another $3 million in new sales, marketing, and product development talent. The recruiting costs relate to the significant increase in revenues we expect this year and will not be incurred next year to support that revenue going forward. The investment in sales, marketing, and product development are incurred to continue our growth momentum, and we anticipate that they will yield revenue and profitability benefits primarily next year and beyond. We anticipate approximately 70 percent of the recruiting costs to be incurred in Q2, and most of the OPEX investment to be incurred in the second half of the year. We are making these investments while simultaneously driving year-over-year growth in adjusted EBITDA and building cash on our balance sheet. At the end of Q1, our cash balances were $19 million, up from $13.8 million at the end of Q4 2023, driven by positive cash flow from operations and tight working capital management. I'll now turn the call over to Maryse to go over the numbers, and then we'll open the line for some questions.

Thank you, Jack, and good afternoon, everyone. Let me briefly share with you our 2024 first quarter financial results. Revenue was $26.5 million, up 41% from $18.8 million in the same period last year. Net income was $1 million or $0.03 per basic and diluted share compared to a net loss of $2.1 million or $0.08 per basic and diluted share in the same period last year. The adjusted EBITDA was $3.8 million compared to adjusted EBITDA of $0.8 million in the same period last year. Our cash and cash equivalent and short-term investments were $19 million at March 31, 2024 and $13.8 million at December 31, 2023. We currently have an unused line of credit of $10 million with $9.2 million as borrowing base. So thank you, everyone. Paul, we're ready for questions.

Certainly. At this time, we'll be conducting a question and answer session. If you would like to ask a question, please press star 1 on your telephone keypad. A confirmation tone will indicate your line is in the question queue. You may press star two if you would like to remove your question from the queue. For participants using speaker equipment, it may be necessary to pick up your handset before pressing the star keys. Once again, please press star one on your phone at this time if you wish to ask a question. And please hold while we poll for questions. The first question today is coming from Aria Cole from Cole Capital. Aria, your line is live.