AI Isn’t Killing SaaS. Misunderstanding It Is. A Conversation with Jana Eggers, CEO of Nara Logics

Headlines proclaim software is dead. Public software stocks are down roughly 40% from their peak. Private equity firms are quietly writing down portfolio companies acquired at COVID-era valuations (perish the thought). And every board meeting Lancer Group attends opens with the same anxious question: did AI just gut our investment thesis?

So Lancer Group called Jana Eggers — someone with 35 years in artificial intelligence and a knack for cutting through hype — and asked her what she’d tell a room full of SaaS CEOs about AI strategy (particularly those with PE backing whose valuations have fallen hard in the last 120 days). She didn’t reach for a framework from McKinsey or Gartner. She reached for breakfast.

Chicken, Eggs, and Bacon

Everyone wants to have an answer for their AI strategy.

“I call it the chicken, eggs, and bacon of AI,” Eggers says. “The chicken is the algorithm. The eggs are the data — think many, lots of data, many eggs. And the bacon? How does it make us money?”

Most companies, she observes, are over-indexed on the algorithm — chasing the latest model or tool. Some are starting to understand the data. Very few have clarity on the business case (the bacon).

“If you don’t know your business case, you have no idea — you may have a great algorithm and set of data, and yet there really isn’t a business problem of value for that algorithm and that data,” Eggers explains. “The bacon is the thing that makes our customers happy. That’s why they’re going to pay us money. Why are they going to hire us to do something? It’s the bacon.”

It's a disarmingly simple framework from someone with the credentials to make it complicated. Jana Eggers has been working in artificial intelligence since the 1990s — long before it became a boardroom panic. She started at Los Alamos National Laboratory, building algorithms for computational chemistry on supercomputers. She went on to lead business and product organizations at Intuit (Nasdaq: INTU) and Blackbaud (Nasdaq: BLKB, then a ~$327M nonprofit software business serving 22,000+ customers, formerly a Hellman & Friedman and JMI Equity portfolio company), ran a European e-commerce company as CEO (backed by Accel Partners), and for the last decade has led Nara Logics, a neuroscience-based AI company in Boston. She also sits on the board of Amadeus IT Group, the global travel technology group (market cap ~$29 billion).

Lancer Group has known Jana since placing her as SVP of Products & Marketing at Blackbaud in 2010, where she ran a 900-person team across 90 products. We caught up with her from Madrid, between board meetings.

Her response, after 35 years in the field, was characteristically blunt:

“The early Q1 market overreaction was the, ‘Oh, AI is going to replace all of software.’ And that just makes no sense. It (the market reaction against SaaS valuations) really is that bad. It’s not just a little bad — the market is dramatically overreacting.”

What follows are her perspectives on what actually matters for SaaS executives and the investors behind them.

Most Companies Are Looking in the Wrong Place

When SaaS CEOs ask where to begin with AI, the conversation usually jumps to tooling, org charts, or pilot programs. Eggers consistently pulls it back to first principles.

“Who are you solving a problem for?” she asks. “I’m going to go back to the basics and say, who are your customers? How well do you know their pains? There are plenty of companies that are systems of record that don’t really know their customers.”

It’s a question that sounds obvious, but in practice it’s where many software companies drift. As organizations scale, they optimize around features, roadmaps, and internal priorities — and slowly lose clarity on the core problems of their customers.

The constraint isn’t how to build the technology. It’s clarity on the customer. AI doesn’t fix that gap. If anything, it makes the gap more visible. Without a deep understanding of the underlying problem, teams default to surface-level use cases that sound compelling but don’t meaningfully change outcomes.

The Real Advantage Is Data, Not Models

Much of the AI conversation focuses on models and capability. Eggers reframes it around something more fundamental.

“As a software company, am I a system of record for my customers?” she asks. “If the answer to that is no, ask, ‘What data do I have access to that I have rights to act on for my customers?’ Because data — it’s the blood of AI. It’s what circulates around and actually makes actions intelligent.”

She’s careful to note that “system of record” doesn’t require owning the data outright. Many companies can act on data on behalf of their customers, and that access is itself a form of strategic advantage.

What may change more quickly, she adds, is how users interact with software. “I think we’re going to have a lot more interfaces into data.”

Her example is Intuit, where she previously ran QuickBase and built the innovation lab.

“I know their stock got pummeled, and I think that’s wrong. They are the system of record for 85% of the small businesses in the US, and they do have access to that data. They’re still in a great position and thus have great opportunities because they know and understand its importance and how it’s used better than anyone else.”

For PE-backed vertical software companies — the kind Lancer Group works with daily — this framing is directly relevant. A specialized CRM for nonprofits, an ERP for mid-market manufacturers, a clinical workflow system for healthcare networks: these companies often underestimate how much latent value sits in the data they already touch. AI doesn’t change that position. If anything, it reinforces it.

Why “AI Efficiency” Won’t Shrink the Market

There’s a widespread assumption that AI-driven efficiency will compress the software industry — fewer engineers needed, fewer seats sold, lower total spend. Eggers counters with an economic concept that deserves more attention from PE operators: Jevons Paradox.

“It occurs when technological progress reaches efficiency with the resources used, but now the new demand makes up for that efficiency,” she explains. “So total consumption doesn’t decrease.”

The classic example dates to 1865, when British economist William Stanley Jevons observed that as steam engines became more efficient, coal consumption didn’t fall — it rose, because efficiency made coal economical to deploy in far more places. The paradox has been validated repeatedly across industries: cheaper electricity led to more electrified appliances. Faster cars led to longer commutes. Cheaper computing led to vastly more software.

Eggers offers a more recent example from her own field.

“Way back in 2016, one of the deep learning folks said, ‘Oh, there’ll be no radiologists going forward.’ The need for radiologists has increased, up over 20% from that time, because we can do more radiology exams. AI has made radiology more efficient.”

The data supports her observation. The Harvey L. Neiman Health Policy Institute and the Association of American Medical Colleges both project a persistent radiologist shortage through 2055. Average radiologist workload nearly doubled between 2008 and 2018. Imaging volume continues to outpace residency growth.

She sees the same dynamic playing out in software. AI won’t eliminate the need for engineers and products — it will expand the surface area of what’s possible to build, creating demand that didn’t previously exist.

“There are very few software companies, if any, that believe they’ve solved every problem that their customers have,” she says. “That’s where I see AI being more productive. And maybe not decreasing costs at all.”

For PE investors modeling the impact of AI on portfolio companies, this reframing matters. The question isn’t whether AI reduces headcount. It’s whether it expands the addressable problem set.

The AI Productivity Question Most Boards Aren’t Asking

If Jevons Paradox is the macro argument, Eggers has a micro one drawn from her own engineering team.

“My team was very active early on in using AI for code development, and people were so happy. They said, ‘Oh my gosh, this makes it so much easier.’ And, yes, I was the pointy-haired manager who said, ‘Cool. What metrics can you tell me?’ Them, ‘Well, it’s just easier.’ Me, ‘How do you know that?’”

When they measured, the results were sobering.

“It turns out that our pull requests (PRs) — which are requests for review of software code — had 30% more code. And then we looked at the time to review, and the time to review dramatically increased. More code can be great if it is solving more of a problem, but we likely didn’t actually save any development time.”

The problem wasn’t just volume.

“The code was more complicated because it didn’t follow our standards. Now we need to update our standards to include generated code. We won’t stop using AI, but we learn to work differently to take advantage of the benefits.”

She reports hearing the same pattern elsewhere. “A friend of mine was at a conference this past week, and they said the amount of code that’s being shipped from AI has increased dramatically — and so has the issue count.”

Her takeaway isn’t that AI coding tools are useless. It’s that the productivity gains haven’t been proven in any rigorous way — and the downstream costs (review time, incidents, inconsistency) are real and measurable.

“I’m not saying it won’t get better,” she says. “But we’re not there yet, and we all need to learn and adjust how we work.”

For boards evaluating management team claims about AI-driven engineering efficiency, this is a useful sanity check. Ask for the full pipeline metrics, not just output volume.

What Agents Mean for Your Business Model

While most of the AI conversation focuses on product and engineering, Eggers thinks the more disruptive shift is economic. As AI agents begin acting on behalf of users, the entire consumption layer of software changes.

“When and if — and I would say it’s just when, not if — there are agents involved in the process of interacting with your system of record data,” she warns. “If an agent goes and performs actions, is that a seat in software? We don’t know a lot of these things.”

She reaches for concrete examples.

“Simple example: So I have an agent that goes to Google for me and finds something for me. That wasn’t me. I never saw the ads. How does Google monetize that?”

And at a more mundane level: “A restaurant has agents calling for reservations. That may quadruple the number of calls they get. How are they going to handle that? Are they going to go to agents too? Are they going to have a special line for agents?”

This isn’t theoretical. Every SaaS company with a per-seat or per-user pricing model needs to think through what happens when a meaningful share of interactions shifts from humans to agents. Volume goes up, but the value-per-interaction may go down.

As Eggers puts it:

“People should be stepping through the process and say, ‘We’re now going to have agents doing this. What does that mean?’”

She draws a parallel to the early cloud cost panic — instructive for the current moment of token-cost anxiety.

“I had a total flashback to the early days of cloud. Everybody shifted to the cloud and then went through this process of, ‘Oh my gosh, my cloud costs are completely out of control; this isn’t cheaper.’ Companies like CloudHealth came up with a solution, but then Amazon built it in. So there are some opportunities that crop up and then may end up going away because the tool providers change their offers.”

The Insight Most People Will Miss: Text, Like Talk, Is Cheap

Here’s where Eggers gets genuinely technical — and where SaaS operators should pay close attention.

Large language models process information linearly. That’s fine for text. But most enterprise software runs on structured, tabular data — rows and columns where position carries meaning. As Eggers explains:

“Large language models are not great with tabular structured data. And I don’t just mean by what’s displayed on the screen, but just the way that data is — because the columns mean something and the rows mean something. LLMs don’t have an idea that a column may have a meaning and that actually impacts what’s in that row.”

This is a fundamental limitation that the current hype cycle largely ignores. The AI tools getting all the attention — ChatGPT, Claude, and GitHub Copilot — are built for unstructured text and code. They’re not natively suited to the kind of structured data that powers CRMs, ERPs, financial systems, and most vertical software.

But a different branch of AI research is emerging to address exactly this gap.

“There’s some deep learning guys that are working on this and they call the models tabular foundation models,” Eggers says. “There are a few advantages: it’s smaller data usually, and it has more patterns because there’s a reason why we’re putting it into tables. That works really well for people like Blackbaud and Intuit, and workflow process software companies like that.”

For companies like those — and by extension, the entire universe of PE-backed vertical software — this may be the more relevant AI development to track.

“It’s probably less LLMs where people think that LLMs are what we’re moving to,” she says. “There are more types of AI that can be applied and maybe apply very well in those cases.”

The implication for investors: when a portfolio company says “we’re implementing AI,” the first questions should be which kind and on what data structure. Not all AI is the same, and the most commercially valuable applications for vertical software may come from approaches that aren’t yet making headlines.

Back to the Bacon

This is why Eggers’ breakfast framework matters more than it might first appear. The chicken (algorithm) and the eggs (data) get all the attention. But the bacon — the business case, the thing that makes customers happy enough to pay — is where most AI initiatives fall apart. Without it, you’re running experiments, not building value. And most likely, you are adopting an AI that doesn’t fit your business case.

Find Enablers, Not Gatekeepers

On the talent question — who should own AI within an organization — Eggers is less interested in org chart design than in disposition.

“Is your data organization a gatekeeper or an enabler? That’s the big question,” she says. “If you’re going to fully leverage AI, you need to make sure you have enablers rather than gatekeepers. I saw people getting into a lot of trouble with regulations around data because they would have a data team that all they were focused on was stopping people from doing things with the data. That’s not how you do business.”

She frames the hiring signal vividly: “Did they build their empire where they were building up fortress walls, or did they build lots of bridges?”

The person you want leading AI efforts, in her view, is someone who “knows how to build bridges with the right guardrails, and get sh*t done.”

What This Means

Jana Eggers isn’t telling SaaS leaders to ignore AI. She’s telling them to stop panicking about the wrong things. The models will keep improving. The tools will get cheaper. What won’t change is the importance of understanding your data, your customer, and your business case — in that order.

For PE-backed software companies navigating this environment, the strategic questions are more specific than the hype suggests:

• What data do you actually have rights to act on?
• How does agent interaction change your pricing model?
• Are you measuring AI productivity gains rigorously, or celebrating output without tracking downstream costs?
• And are you paying attention to the AI developments — like tabular foundation models — that are specifically relevant to structured enterprise data?

These aren’t questions that get answered by hiring a Chief AI Officer or announcing a partnership with OpenAI. They get answered by the kind of disciplined, customer-first thinking that has always separated great operators from average ones.

Jana Eggers is CEO of Nara Logics and serves on the board of Amadeus IT Group (market cap ~$29B). The Lancer Group placed Jana as SVP of Products & Marketing at Blackbaud in 2010.