Practical AI That Works: High-ROI Use Cases & Guardrails from Poseidon AI Lab

There is no shortage of enthusiasm about AI in the enterprise. There is, however, a growing shortage of results. Across the market, organizations are spending aggressively on AI initiatives — but industry-wide, many of those projects are stalling, caught in the gap between what leadership expects AI to deliver and what it can actually do today. Bridging that gap is the central mission of Poseidon, Altum Strategy Group’s AI Lab.

“When we launched Poseidon in 2024, the goal was to help organizations move from theoretical AI concepts to meaningful day-to-day execution,” says Matthew Gantner, Founder and CEO of Altum Strategy Group. “Two years in, that mission hasn’t changed — but the technology has advanced significantly. We’ve implemented our five-step governance framework with clients, and we’re now in a position to deliver practical AI solutions that generate real, measurable ROI. The challenge for the broader market is translating excitement into implementation.”

What “practical” actually means

The word “practical” is doing important work in Poseidon’s positioning, and Andy Pojuner, Managing Director of Intelligence, Data & Technology at Altum, is precise about what it means.

“Practical AI should function as a deterministic tool, not a purely probabilistic one,” Pojuner explains. “The most valuable applications we build involve using AI to generate code that can systematically analyze data or perform specific tasks — not relying on AI’s predictive capabilities alone. When you need consistent, auditable, repeatable results, you need determinism. That’s what we design for.”

The distinction matters because it shapes what Poseidon recommends to clients and — just as importantly — what it advises against. Probabilistic AI has its place, but when the use case involves financial data, compliance evidence, or operational processes where consistency is non-negotiable, the approach has to be engineered for reliability rather than novelty.

The use cases delivering ROI right now

Pojuner points to well-defined business processes as the highest-ROI territory for AI deployment today, particularly accounts payable and accounts receivable automation.

“AP and AR invoice processing is a strong example,” he says. “You’re dealing with unformatted invoices that would otherwise require manual work — someone reading the document, extracting the data, and entering it into a system. AI handles that efficiently, accurately, and at scale. The ROI is immediate and measurable.”

The second category is data integration — using AI to analyze structured data across data warehouses or data lakes and surface insights that would take analysts significantly longer to find manually. When AI has access to well-organized data and a clear set of questions, it can compress what used to be weeks of analysis into hours.

These aren’t futuristic use cases. They’re running in production, and the operational impact is tangible. Gantner identifies three shifts that clients experience once these deployments are in place: 1) team members move from manual data entry to more strategic work, 2) staffing costs decrease as automation handles volume, and 3) the culture of the organization begins to change as people see what’s possible.

“That cultural shift is the one people underestimate,” Gantner notes. “When a finance team goes from spending 60% of its time on manual processing to spending that time on analysis and decision support, the way people think about their roles changes. That’s a transformation in itself.”

The data foundation comes first

Both Gantner and Pojuner are emphatic on one point: AI amplifies whatever it sits on top of, including dysfunction. If the underlying data is dirty, fragmented, or ungoverned, AI will produce results that reflect those problems — faster.

“Some CEOs believe AI can solve their data problems independently,” Gantner says. “The reality-based leaders understand that human business processes still need to address data challenges before AI can be effective. That’s what Poseidon focuses on first — getting the data foundation right.”

This echoes the argument Altum has made in its Forbes Business Council piece on the connected enterprise: decision rights, governance, and a single source of truth have to precede AI deployment. In Poseidon’s experience, the AI initiative itself often becomes the catalyst for addressing those foundational issues. Companies come to Altum wanting AI and discover that the real work is cleaning up the data and processes that AI needs to function. The AI project becomes the forcing function for a broader transformation — one that delivers value regardless of how far the AI deployment ultimately goes.

Governance as a design principle

AI that touches financial statements, compliance processes, or customer-facing operations needs governance that’s at least as rigorous as any traditional system control. Gantner is direct about this: the probabilistic nature of AI models makes it inherently challenging to achieve consistent outcomes, so the governance framework must account for monitoring, tracing, and validating what the AI is actually doing.

“Traditional controls still apply,” he says. “If an AI model is generating outputs that affect your financial reporting, you need the same level of auditability and oversight that you’d apply to any other system. The challenge is that AI models can produce different results from the same inputs, and that requires a governance framework that allows you to trace what happened and why.”

Altum’s responsible AI framework — originally published as a five-step process in its AI white paper — is deployed based on client maturity. Pojuner explains that some organizations need a gradual introduction, starting with basic governance principles and building toward more sophisticated monitoring, while others can move through the framework more quickly. The approach is situational, not one-size-fits-all.

Where AI doesn’t belong — yet

Not every problem is an AI problem, and both Gantner and Pojuner are willing to tell clients when the technology isn’t the right answer.

Gantner identifies two primary disqualifiers. The first is when the decision genuinely requires human judgment — situations involving nuance, context, or stakeholder relationships that AI cannot replicate. The second is when the organization’s leadership and cultural readiness aren’t in place. “If the leadership team isn’t mature enough to govern AI properly, or if the culture isn’t ready to adapt to how AI changes workflows, the deployment will struggle regardless of how good the technology is,” he says.

This raises the question of whether operating model changes must precede AI deployment entirely. Gantner’s answer is nuanced: targeted pilots can work alongside existing operating models, but larger-scale AI transformations require a more comprehensive redesign of how the organization operates. The pilot approach lets companies learn and build confidence, but at some point, the operating model has to catch up.

The human in the loop

Human oversight in a Poseidon deployment is not ceremonial. Pojuner describes it as both a technical and a process requirement.

“Human oversight is necessary to validate and veto AI actions,” he says. “AI systems can introduce security vulnerabilities. They can bypass established processes. The human layer exists to catch those things — and it needs to be designed into the workflow, not bolted on afterward.”

This is consistent with the position Altum has taken across its responsible transformation and cybersecurity work: governance and oversight are accelerants when they’re built in from the start, and liabilities when they’re added after something goes wrong.

Measuring what matters

For CEOs trying to quantify the return on their AI investment, Gantner’s advice is to resist the instinct to frame ROI purely as cost savings.

“Focus on revenue generation and customer delight,” he says. “Cost savings matter, but they’re a lagging indicator. The leading indicators are whether AI is helping you serve customers better, make faster decisions, and create capacity for your people to do higher-value work. That’s where the real return compounds.”

What’s next for Poseidon

Looking ahead, Pojuner outlines two priorities for Poseidon’s next phase. The first is testing larger AI models chained together to produce better, more reliable outputs — an approach that leverages the strengths of different models for different tasks within a single workflow. The second is embedding AI tools more deeply into day-to-day business processes, moving beyond standalone deployments toward AI that’s woven into how work actually gets done.

Gantner adds that the approach will remain grounded in what Altum has always prioritized: data quality, practical experience, and a willingness to test different models to determine their optimal use cases rather than defaulting to whichever technology is generating the most headlines.

“The companies that will get the most from AI over the next 18 months aren’t the ones spending the most,” Gantner says. “They’re the ones who’ve done the foundational work — clean data, sound processes, proper governance — and are now in a position to deploy AI on a platform that can actually support it.”

For more on Poseidon AI Lab and Altum’s approach to responsible AI implementation, visit altumstrategy.com/insights