As the AI tools marketplace grows more crowded, enterprise customers want solutions that excel across several dimensions.
Organizations want systems that deliver practical value-generating AI that users will actually embrace.
They want systems that treat trust, governance, and explainability as first-class requirements rather than retroactive additions.
VP of Engineering at iManage.
How can organizations best address these requirements – and in doing so, help deliver AI capabilities users and organizations will not be able to live without?
Start with the right use cases
Most organizations are beyond their initial phase of adoption with AI, yet many are still making the same foundational mistake: deploying AI without a clear, carefully defined use case. This is a pattern that plagued earlier data science and machine learning projects, and it remains just as damaging a misstep today.
A disciplined focus on appropriate use cases substantially improves the odds that the AI system will create real, tangible value – which in turn drives user adoption.
Consider an AI system that tells professionals information they already know or automatically alerts them about situations they’re fully on top of. Take, for example, an AI system that tells insurance adjusters that a claim that has been open for two years is a problematic claim. The professionals already know that.
Now, compare that to a system that extracts key fields from a claim form and auto-populates them into a claim acknowledgement letter, eliminating tedium and reducing risk.
The former use case is persistent, not of any real value, and likely to create apathy towards AI initiatives among end users. The latter use case is borderline transformative in its real-world practical impact. Defining impact and user-perceived value up front is a big part of what can put AI systems on the track to success.
Think deeply about implementation and user experience
Once the use case is defined, the next critical step is a rigorous consideration of implementation. Are you choosing an AI product from a vendor that brought the product, marketing, infoSec, platform, backend, and API teams into the same room to ensure user experience and security were accounted for, from the outset? The same considerations apply for any enterprise building its own AI system in-house – it can’t be developed in a vacuum by the engineering team alone.
Avoiding blind spots in the development process becomes critical when you consider the regulatory landscape that AI products must now operate within. The EU AI Act’s high-risk system obligations – effective as of February 3, 2026 – create clear requirements around safety, transparency, and oversight for AI vendors that systems must plan to meet, if not exceed, if they are operating in the EU. Any AI vendor or enterprise team operating in the EU must plan to meet those requirements, and the time to design for compliance is at the architecture stage, not after launch.
Geography introduces another consideration: is the large language model that the enterprise intends to use available in every region in which the organisation does business? Some of the most prominent models are not globally available – which means regionality and data sovereignty requirements can become blockers if not addressed early.
From a purely engineering standpoint, implementation planning must also address issues of scalability. AI workloads, for example, are famously “bursty”: a user submits a task to an agent, requiring a heavy computational lift, and expects near-instant results. Engineering teams need to account for capacity and availability and test for it rigorously before go-live, not after. In production, simple pings won’t suffice to monitor infra health, rather the probes should account for examples of complex natural language queries and requests.
Define what “good” looks like and maintain focus
The third pillar of practical AI is knowing what success looks like. This means tying AI outcomes directly to the organization’s existing North Star metrics. If the priority is revenue growth, select and configure AI that demonstrably contributes to revenue. If the goal is operational efficiency, quantify the time savings the system will deliver and hold it accountable to those numbers.
AI adoption does not mean abandoning the performance metrics that already guide the organisation. The most durable AI initiatives are those that map AI outcomes back to established organizational goals – whether that’s reducing risk, accelerating throughput, or freeing skilled professionals from low-value tasks.
Once those metrics are in place, the imperative is focus. AI adoption tends to go sideways when the scope creep enters the picture and the use case becomes fragmented – and teams lose sight of the core value proposition. The answer is not to be rigid, but to be disciplined about maintaining momentum on the initiatives most likely to deliver measurable impact.
Stay flexible
One final note. The AI landscape is changing with extraordinary speed. Models are improving. Unit economics are shifting. Use cases that were frontier eighteen months ago are now commonplace.
In this environment, there should be plenty of room for experimentation. This means being willing to revisit your ROI metrics as you learn. It means being open to nudging a use case slightly – and discovering that the adjusted version unlocks far greater value than the original. It means giving teams permission to iterate, to get things wrong, and to course-correct without losing momentum. This is frontier work for most organizations.
Aim for as much clarity upfront as you can around use cases, implementation, and what “good” looks like but remain flexible as circumstances evolve. That combination of rigor and agility is what separates the AI systems that enterprises tolerate from the ones they can’t imagine working without.
We feature the best Robotic Process Automation (RPA) software.
This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.
The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit
