Innodata Inc. (INOD) Earnings Call Transcript & Summary

February 26, 2026

NASDAQ US Industrials Professional Services earnings 39 min

Earnings Call Speaker Segments

Amy Agress

executive
#1

Thanks, operator. Hello, everyone, and good afternoon. We appreciate you being with us today. Jack Abuhoff, our Chairman and CEO at Innodata, and Marissa Espineli, who is serving as Interim CFO, are the speakers you'll hear from. Aneesh Pendharkar, our Senior Vice President, Finance and Corporate Development, is also participating on this call. Rahul Singhal, our President and Chief Revenue Officer, couldn't make it today, though he's eager to be with us when we hold our next call. Jack will speak first and share his view of the business, after which Mariz will walk through a review of what we delivered in the fourth quarter and fiscal year 2025. Following that, we'll open things up for analyst questions. Before we begin, please be aware that forward-looking statements will be part of this call — that means predictions, projections, and other statements regarding future events; such statements rest on current expectations, assumptions, and estimates, and they carry risks and uncertainties. What actually happens could differ materially from what these forward-looking statements contemplate. You can find the factors that might cause such material differences in today's earnings press release, in the Risk Factors section of our Form 10-K, our Forms 10-Q, and other reports and filings made with the Securities and Exchange Commission; we have no obligation to update forward-looking information. Also, certain non-GAAP financial measures may come up during this call. In our earnings release filed with the SEC today, and in our other SEC filings, all of which are posted on our website, you'll find additional disclosures about these non-GAAP financial measures, including reconciliations of these measures with comparable GAAP measures. Thanks. Jack, over to you now.

Jack Abuhoff

executive
#2

Thanks, Amy, and hello everyone on the line this afternoon. Innodata delivered another robust quarter in Q4. Revenue came in at $72.4 million, which works out to 22% growth compared with the same period a year earlier; that pushed our total revenue for the full year to $251.7 million, equal to 48% growth year over year for 2025. On a consolidated basis, our adjusted gross margin for Q4 landed at 42%, which is above the 40% target we had shared publicly. Adjusted EBITDA came to $15.7 million, or 22% of revenue, and that also beat the analyst consensus figure by $1.2 million. In fact, we surpassed the analyst consensus on a whole set of key metrics — revenue, adjusted EBITDA, net income, and EPS among them; we closed out the year holding $82.2 million in cash, a sequential increase of roughly $8.4 million; all of this was accomplished while we made substantial growth-oriented investments in both COGS and SG&A; on the COGS side, we brought on capacity in advance of revenue ramp, and that decision repeatedly turned out to be correct. Within SG&A, we put money into engineers, data scientists, and account leadership focused on customers, and those investments likewise proved wise, producing innovation that has widened our opportunities. Our sense is that business momentum has never been stronger than it is now. Demand is strong throughout the whole generative AI lifecycle — covering development, evaluation, and continuous model optimization. We think we are building momentum with a large, broad, and diversified set of customers. Given market demand and our growing traction, we look forward to another year of potentially remarkable growth in 2026; as things stand, we estimate that our 2026 growth over the prior year could be roughly 35% or higher; that figure takes into account active programs, wins awarded recently, evaluations at a late stage, and opportunities where visibility is clear. It is still early in the year, and since LLM initiatives can spin up rapidly, we think there could potentially be considerable upside beyond that range. Our preference is to guide conservatively and then raise guidance as visibility improves. That said, because the programs we support are large and complex, variability in timing around customer ramp schedules, budget approvals, or changes in research priorities might affect how quickly revenue actually shows up. Built into our outlook is an assumption that spending from our biggest customer will rise somewhat this year, while the rest of the customer base grows at a quicker pace in aggregate. We expect that growth among other customers to arrive via a blend of the Mag Seven, domestic AI innovation labs, sovereign AI initiatives, and leading enterprises. We think this will add meaningfully to customer diversification. Our customers operate quickly, which drives shorter development cycles and faster responses to research breakthroughs; during 2025, we thrived in that environment largely because we followed the research, anticipated what customers would need, and pivoted wherever necessary. As an example, during the first quarter of this year, for our biggest customer, we deprecated a significant number of post-training workflows, which together amounted to roughly $20 million in annualized revenue run rate. We substituted a mix of new post-training workflows plus scaled pre-training programs, an area we have recently focused on and invested in. Looking at things through the lens of revenue run rate, the net effects came out on the positive side. In fact, our view is that ongoing innovation matters enormously if we're going to hit the ambitious plans we've set for 2026 and past that. What's genuinely thrilling is our belief that Innodata is on the cusp of a golden age of innovation, driven by the investments already made and those we plan going forward. Let me now take you through a few of our recent innovation initiatives. Because of competitive considerations, we'll remain suitably cautious, though what we tell you should provide a meaningful window into our thinking, our investment focus, the wins we're achieving, and our plans for seizing the opportunity that lies ahead. I'll quickly cover our recent innovation across three areas: generative AI model training, agentic AI, and physical AI. Before getting into that, I want to emphasize one theme that ties it all together. Each innovation I'm about to describe is at its core a data innovation. Whether the aim involves more capable LLMs, more dependable autonomous agents, or smarter physical AI systems, what sits at the center of it all is data quality, data composition, data validation, and data engineering at scale; those represent our core competencies. Let's begin with generative AI training. In the past, customers would tell us which kind of training data they were after. More and more, though, they want us to diagnose how a model is performing, design the appropriate training data sets, and show that those data sets will meaningfully improve outcomes. The way this works is as follows: we start by pinpointing performance gaps through our evaluation frameworks; next we engineer targeted data sets and confirm their efficacy by fine-tuning either the customer's model or a structurally similar proxy model. We scale only once we've measured and demonstrated the performance impact. This moves the conversation away from how much is the data toward how effective is the data; our belief is that two forces are driving this shift: the accelerating pace of AI research along with the cost and time incurred to train ever larger models. Discussions centered on data efficacy align directly with our strengths. We're also pushing forward methods for building data sets that enhance long context reasoning, which is an AI model's ability to absorb and reason over very large amounts of information at once. This stays one of the industry's most important technical challenge. To solve it, you need not only architectural improvements, but also advances in the creation at scale of very specific types of structured training data. Creating training data that enhances long context reasoning is a non-trivial problem, yet we have made meaningful progress on it and are continuing to do so. A second area of innovation concerns evaluating systems of autonomous agents and improving them through targeted data set creation. Our belief is that autonomous agents may represent the most significant business innovation opportunity since the advent of electricity. Companies soon find that many AI agents that performed impressively in controlled laboratory settings degrade in real-world production. The real world is chaotic. Edge cases, conflicting constraints, unpredictable user behavior, and adversarial conditions all shape it. Tackling this is fundamentally a data challenge. Continuous training and rigorous stress testing using realistic, varied, and intricate data sets are essential for agents. To address this, we've created a trio of hybrid offerings that work together effectively. The initial one is a platform for agent evaluation and observability. Throughout development, data scientists can leverage our platform to visualize and annotate agent trace data, construct LLM-as-a-Judge evaluators, develop evaluation rubrics aligned with business goals, produce golden data sets for regression testing, and generate test data at scale. After the agent goes live, our platform enables ongoing performance monitoring, root cause analysis of issues, and acquisition of mitigation datasets; we're excited to announce that we expect to soon begin a managed services partnership with a hyperscaler, where we'll utilize our platform to generate test data at scale, conduct automated evaluations, and pinpoint critical model vulnerabilities to enhance the performance of its customer-facing intelligent virtual assistant. The second offering is a managed agent optimization pipeline crafted to methodically train for and thus eliminate the unpredictability of real-world deployment at scale. This pipeline creates realistic test scenarios, automates evaluation, meticulously assesses constraint satisfaction, and yields reinforcement learning datasets. With this approach, we've shown gains of as much as 25 points in constraint satisfaction.

Jack Abuhoff

executive
#3

What's key here is that when agents are trained the usual way, their performance drops off a lot once tasks get more complex. Our pipeline, on the other hand, keeps agents performing well even as real-world difficulty ramps up. In the toughest situations, the gap between standard methods and what we've built stretches past 31 points. Right now, a number of AI innovation labs and enterprise customers are actively testing out this system. Our third offering built to enable enterprise agentic AI is an adversarial simulation system, which produces adversarial attacks that are high quality, semantically diverse, and scalable, in order to stress test agents. It covers the entire range of attack types — direct jailbreaks, indirect prompt injection through RAG pipelines, multi-turn social engineering steganographic payloads, plus compound attacks merging injection methods with domain-specific knowledge. After vulnerabilities surface, it creates mitigation datasets that are highly targeted to reinforce guardrails. Our belief is that this system delivers realistic adversarial attacks at scale in a way that matters, surpassing what else is out there. A lot of tools available today generate hostile content that's simplistic or template-based, missing the nuance and sophistication real threat actors bring, or they don't scale across varied scenarios, or they lean on generic tactics that models soon learn to expect and overfit to. Our framework, in contrast, is built to emulate attack patterns that are adaptive, multi-step, and strategically coherent, covering highly sophisticated model extraction, cybersecurity, cybercrime, and subprime threat scenarios that more accurately capture how advanced adversaries work and let our partners remain ahead of new threats. What comes out is adversarial training data that's scalable and durable, pushing models to generalize instead of memorize, which enables stronger real-world resilience. CISOs and security leaders at some of the top AI and Cybersecurity companies globally, along with relevant government experts, are taking interest in our work, and we've had early-stage engagements with a number of them. Given that the cyber industry is going through major disruption right now, these capabilities strengthen where we stand in the growing field of AI trust and safety — an area where our work with several hyperscalers is deepening in a meaningful way. Our view is that Innodata is well-positioned to become a leader in prompt layer security, safeguarding AI systems where interaction happens rather than depending only on conventional perimeter or endpoint defenses. All together, we think these solutions make us more than just a data supplier — a lifecycle partner in agent reliability; we also believe 2026 will bring an acceleration of physical AI, intelligent systems that perceive the physical world and interact with it. Robotics supplies the mechanical framework, while physical AI supplies the intelligence. Dataset quality and scale represent the main bottleneck here. Manual annotation and static QA sampling just can't scale to billion-sample corpora and environments that keep evolving. We've built a data engineering platform at scale that brings in structural validation, monitoring of distributions, checks for temporal consistency, and instrumentation with models in the loop; through it, we can spot and fix problems in datasets before they turn into performance failures; parts of this platform are already in use across the high-visibility engagements we announced recently with Palantir. Not long ago we won a major engagement to build foundational datasets for next-generation robotic datasets, egocentric data among them. With egocentric data, the world is captured from the robot's perspective — what it sees and experiences while moving. We're additionally collaborating with a leading robotics lab to produce affordance data at scale. Affordance data teaches a system which actions are possible within a given setting, going beyond recognizing objects to grasping how they may be used. Put together, egocentric data and affordance data create the cognitive scaffolding that lets machines behave intelligently in dynamic environments. This effort also puts us in position to aid the creation of what are called world models — internal simulations that let AI systems foresee outcomes, reason about cause and effect, and plan multiple steps ahead. World models call for richly structured datasets that capture interactions across time along with the consequences of actions, which is exactly the kind of data we're engineering now. Last, we recently created an AI model for detecting drones and other small objects that beats prior state-of-the-art benchmarks by 6.45%; in an area where gains are typically measured in fractions of a percentage point, an improvement of 6.45% counts as a material advance. The model boosts detection fidelity in real-world conditions, where small size, speed, cluttered backgrounds, and environmental noise render reliable perception extraordinarily hard. We think this advancement carries compelling dual-use implications that we're actively exploring with potential customers. I want to emphasize one of the key points I just raised. For decades, Innodata has focused on building high-quality complex datasets. These capabilities are now central to unlocking the next generation of AI systems. Advanced LLM reasoning, agent reliability in chaotic environments, and robotics perception in the physical world all rest on engineered data ecosystems. That's exactly where we play. Our innovations in LLM training, agentic AI, and physical AI aren't separate initiatives. Instead, they extend a single strategic advantage — our capacity to engineer data that measurably lifts model performance in real-world conditions. We believe our innovation pipeline will enhance margins as well as revenue; we expect adjusted gross margins in the 35%-40% range in early 2026 as we ramp up new programs, with normalization toward our target of 40% or better adjusted gross margins as new programs ramp up and as innovation-driven workflows scale. Automation, synthetic systems, and evaluation platforms all structurally raise our operating leverage. I'll now hand the call to Mariz, who will walk through the numbers.

Marissa Espineli

executive
#4

Thanks, Jack, and good afternoon to all of you. For Q4 2025, revenue came in at $72.4 million, which is a 22% rise compared with the same period a year earlier. Looking sequentially, we saw revenue grow 15.7% over the $62.6 million recorded in Q3. Adjusted gross profit in Q4 2025 stood at $30.1 million, marking a 6% year-over-year gain and a 9% sequential gain, with adjusted gross margin landing at 42%. Adjusted EBITDA came to $15.7 million, equal to 22% of revenue, while net income for the quarter was $8.8 million. To repeat that point, this figure is net of substantially increased data science and engineering investments, which are producing the kinds of innovations Jack just described; we closed the quarter holding $82.2 million in cash, which is up from $73.9 million at the close of the previous quarter and $46.9 million at the end of 2024; we made no drawdown on our $30 million credit facility with Wells Fargo. As Jack noted, given the momentum we currently have, we are now projecting revenue growth of 35% or more year-over-year for 2026. Thanks to all of you for being with us today. Operator, please open up the line for questions.

George Sutton

analyst
#5

Thanks. Jack, I feel as though I've just been in a sophisticated AI data science lecture, so I appreciate it. I'd like to pull back a bit, since I believe folks hold the notion that a portion of what's succeeding for you is rather short-lived. I think you've accomplished something intriguing by guiding us across prior quarters—beginning with post-training, moving to pre-training, and now we have striking additional use cases, such as robotics and autonomous agents. Could you discuss how wide-ranging the things you're observing are, and where you'd place us along this data science opportunity continuum for your company?

Jack Abuhoff

executive
#6

Certainly. Thanks, George. I appreciate the question. As we peer ahead, in the near term, 2026, we view ourselves as extremely well positioned thanks to the breakthroughs we put money into during 2025. That innovation output, we view it as a flywheel. We're improving, we're becoming more capable, and we're building solutions that address the very obstacles enterprises encounter when attempting to bring AI into their operations. Across the range of today's AI abilities and tomorrow's abilities in areas such as agentic systems, physical AI, robotics, all of it comes down to difficulties involving data engineering. Naturally, architectures will keep getting better. Models will grow larger. There will be more specialized models for domain-specific challenges. Fundamentally, when it comes to making systems dependable and safe at an enterprise level, it will hinge on breakthroughs like the ones we're unveiling today—datasets employed for evaluation, datasets employed for training and enhancing model safety and reliability. We believe we're only at the start, and our importance is in no way shrinking—it's only growing. That growth isn't limited to the foundation model builder level; it's plainly spreading throughout the enterprise. We're thrilled about our current position and about how the innovations we're developing are being adopted and will be adopted over the coming several years.

George Sutton

analyst
#7

That's excellent; now I have just one more question. Over the past 2 years, you began each one with a certain expectation, and then you wound up considerably surpassing what you'd initially anticipated. Is anything arranged differently as we head into 2026, whether in terms of what you're observing at your sites or what you're promising today?

Jack Abuhoff

executive
#8

No, not in the slightest. We're sticking to precisely the same approach. You know, we truly constrain our — or rather, we adopt a cautious stance when projecting growth, and we base it on possibilities where our visibility is extremely clear. If we can't forecast a closing rate, if we can't, you know, feel, you know, fairly sure that something will occur, you know, we simply won't factor that into the guidance we provide. Our goal is to astonish and to surpass what's expected. Looking at this year, I believe it will probably be yet another one in which we do just that. You know, we're witnessing tremendous opportunity with a far bigger group of clients. We believe that will translate into growth. I suspect we'll be raising guidance as the year progresses. I think it's going to be a year in which we achieve extremely significant diversification among our customers. Beyond that, as we've already covered, I think it's gonna be a year where, you know, we begin to observe increasingly hybrid solutions blending human and technological elements, which points to or offers the prospect, in my view, of, you know, greater recurring revenue. I think it holds the promise of better margins down the road, stronger stickiness, a great many things that will, you know, eventually, I believe, keep enhancing revenue quality too, on top of all the rest; regarding the work we perform with foundation model builders, you know, we're experiencing massive traction, not only with our biggest customer, but with others too. We're closely in step with what they're aiming to achieve in areas such as long context reasoning enhancements. We possess innovations that are adding to that. We're incredibly thrilled about where we stand at this moment.

George Sutton

analyst
#9

All right. Good stuff. Thanks, Jack.

Jack Abuhoff

executive
#10

Thank you.

Hamed Khorsand

analyst
#11

Hello there. Yeah, so regarding that initial question: earlier you mentioned expanding your operations while revenue, as you know, increases. At present, do you have sufficient staff? Do you anticipate needing to hire additional workers? And when do you expect gross margin to start rising from this point onward? Thanks.

Jack Abuhoff

executive
#12

Absolutely. Thank you, Hamed. I believe a lot hinges on the conditions we observe. If we start forecasting — and, as you know, internally — growth rates that are, as you know, quite substantial, then we will be putting money to work so that we can be sure we seize those growth rates. I genuinely believe that after absorbing some of those, you know, personnel-related expenditures we're incurring within COGS, which stem from the innovations we've been talking about, you know, various items along those lines, I truly believe that we are going to, you know, I truly believe that we will witness a shift, you know, returning toward our target gross margins as time goes on.

Hamed Khorsand

analyst
#13

Alright, so when it comes to that set of deals you mentioned — the ones involving customers besides your biggest one — is there any kind of timeline attached to them?

Jack Abuhoff

executive
#14

Pipelines do exist, yes—but the transactions I have in mind are mostly ones we are finalizing or already have finalized. It's not as though we're counting on guesses about future activity. These are matters currently in motion.

Hamed Khorsand

analyst
#15

All right. Thanks.

Allen Klee

analyst
#16

Hello there. Regarding 2025, my understanding is that you were guiding toward roughly a 23% margin. It's clear that reinvesting in the company matters to you for its well-being. What I'd like to know is whether there's any basis for expecting you to aim at an adjusted EBITDA margin above or below the one from 2025?

Jack Abuhoff

executive
#17

Capturing opportunity is our central priority at this moment. We think we're able to do so while remaining profitable. Beyond that, we feel that grabbing those opportunities and carrying out some of the initiatives we've been outlining — validating those innovations — matters more than chasing adjusted gross margin figures and holding to a particular percentage. As you know, we'll be putting money back into the company on an ongoing basis. The greater the number of opportunities we spot, the more we'll be reinvesting, up to a point; even so, we're confident that we can keep up profitability as we pursue extremely aggressive expansion and as we grow ever more essential to a bigger and increasingly broad group of clients.

Allen Klee

analyst
#18

All right. Among the bullet points on innovation, you mentioned a structural foundation for margin expansion via automation, synthetic data generation, and evaluation platforms. Could you clarify a bit what you mean, and which margin expansion you're talking about? Thank you.

Jack Abuhoff

executive
#19

Yes. What we mean is gross margin expansion over time. Many of the innovations we're currently working on and introducing to the market are hybridizations of software and human teams. My belief is that as time goes on, the gross margins tied to those capabilities will be maybe, you know, far above the gross margins we aim for today.

Allen Klee

analyst
#20

All right, understood. That clarifies things well. My final question concerns Q1 of 2026 — anything you'd like to flag that could really jump out, whether on the revenue side or on spending?

Jack Abuhoff

executive
#21

Sure, I won't claim it happens in the coming quarter for sure, but fairly soon, in my view, we'll post quarters whose top line exceeds the full-year revenue we booked 3 years back. That alone is quite encouraging. Over the course of this year, you should expect additional proof points, further evidence, and deeper involvement with a number of quite compelling businesses tied to the innovations we've been outlining. I believe we'll begin showing that we are in some measure shifting from being a supplier toward functioning as a foundational layer inside AI ecosystems. Turning into an entity capable of unlocking AI's promise across enterprise engagements. An organization that can assist enterprises in adopting sophisticated agents which plan, invoke tools, run intricate workflows, and generate substantial value. I expect we'll witness that. I expect proof of it in Q1. I expect that proof to keep appearing throughout the year.

Allen Klee

analyst
#22

Perhaps one final brief question. We were discussing your biggest client, and I'm not sure I completely follow. You brought up a figure of $20 million that might get substituted by a larger amount? Could you clarify the reason for that? Yeah.

Jack Abuhoff

executive
#23

Yeah. What I believe we were getting at is just how crucial innovation has become for our business right now, and it keeps growing in importance. You know, we finish certain projects while kicking off others. Through pursuing innovation — as Wayne Gretzky famously put it — by skating to where the puck is going, we can retire offerings that clients no longer need, while still being present for whatever new needs they develop. Once again, the new needs we're observing look more compelling in terms of business, revenue quality, and differentiation compared with what existed previously. The money we've put in is paying off. It's letting us expand and widen the scope of our engagements. It's helping us land new engagements and new clients, some of which we expect to be quite significant. They're really going to blossom this year. That will resolve the diversification concern. You know, looking ahead to 2026, we foresee an enormous year of growth. We think we'll probably be raising our guidance above what we're beginning the year with; we believe the solutions and the way we're integrated into workflows will become increasingly compelling and beneficial for both margin and revenue. It's shaping up to be an extraordinary year across every one of those dimensions.

Allen Klee

analyst
#24

That's great. Congratulations. Thank you.

Jack Abuhoff

executive
#25

Thank you.

Jack Abuhoff

executive
#26

Thank you, operator. So, in closing, 2025 turned out to be an excellent year. There's a real chance 2026 will be even stronger; during 2025, our top-line growth was robust. We beat expectations on the major financial metrics. Our margins got wider. Our balance sheet became stronger. We made successful investments ahead of demand. Those investments turned out to be, you know, extremely successful and positioned us nicely for 2026. You know, my view is that 2026 could well be an amazing year; we've provided guidance of roughly 35% growth, based on what we can see today. I think there could be quite substantial upside beyond that; we'll give you updates as the year goes on, just as we've done over the past couple of years. I also want to highlight, you know, our conviction that this year we may significantly diversify where our revenue comes from. We think expertly engineered data ecosystems are going to matter just as much as larger models and new architectures when it comes to pushing forward, you know, language models, media models, autonomous agents, robots, world models, and other forms of AI that nobody has even thought of yet. We're really excited about what's coming. We have a lot of confidence in where we stand; we're deeply committed to building one of the most important, and in our view, most capable AI enablement companies in the industry. It's going to be an exciting year. Thank you all for coming along on the journey with us. Looking forward to next time.

Operator

operator
#27

Everyone, that brings us to the end of today's call. We appreciate you joining us. You're free to hang up at this point.

Read the full transcript via the API

You're viewing the first half of this call. Get the complete Innodata Inc. transcript — plus 132,000+ transcripts from 7,000+ companies, speaker segments, AI summaries and full-text search — through the EarningsCalls.dev API.

Get the API View API docs →

This call discussed

For developers and AI pipelines

Programmatic access to Innodata Inc. earnings transcripts and 132,000+ others is available through the EarningsCalls.dev REST API. Plans from $39.99/month — full transcripts, speaker segments, full-text search, and the recently-added /api/v1/transcripts/recent polling endpoint for ETL pipelines.

More tools for stock research

OptionsBell: an unusual options activity scanner that tracks options flow across all US stocks and emails you when large trades appear in Innodata Inc. or any ticker on your watchlist.

StockMarketScan: a stock screener with alerts: 24 pre-built screens refreshed every 5 minutes, watchlists and email notifications. Both offer API and MCP access and fair monthly pricing.