Perspectives

Executive summary

As businesses race to leverage the potential of AI, technology is no longer the main thing holding them back. The tools teams have at their disposal are more powerful, and more versatile than ever, capable of meeting big customer and performance ambitions. 

But success depends on deploying these tools effectively, and ensuring they’re embraced by end-users. Too often what prevents that is trust - in the data that ultimately powers AI projects, and in using that data as a basis for business decisions. Employees and customers are quietly abandoning AI agents and chatbots because of substandard output, and concerns about credibility. The rapid growth of agentic AI is only raising the stakes and the complexity, given the potential risks posed by more autonomous, non-deterministic systems trained on low-quality data. 

In this edition of Perspectives Thoughtworks experts explain why the best response to the data trust crisis is a renewed focus on the basics. By strengthening the often under-invested foundations of a sustainable data strategy - data quality, data governance and data utility - businesses can create a virtuous cycle of data that ensures AI-enabled insights and capabilities feed directly into positive outcomes. 

Key takeaways

Too many AI projects are still destined for the ‘POC cemetery.’ Enterprises have no shortage of ideas about how to deploy AI agents, and new solutions have made it relatively easy to churn out prototypes. The limitations emerge when these systems are brought into the production environment, and must contend with real-world use cases and expectations. 

Lapses rooted in language can prevent the flow of high-quality data through systems, and the business. A lack of universal and up-to-date organizational semantics can lead to gaps and inconsistencies that threaten the quality of the data resources vital to successful AI. The potential ‘blast radius’ of agentic systems means the negative consequences of low-quality data can go well beyond degraded system performance. 

Data strategy needs to be oriented around a common understanding of value. Building this understanding is a delicate process that often requires balancing the interests of multiple stakeholders, but can be achieved by focusing on needs and designing a process accordingly. Any benefits from planned AI projects must also be measured against the risks, which can present too high a cost. 

A good strategy means being selective. The speed at which technology is evolving makes it easy to get swept up in the excitement about the latest and greatest solutions. But with new AI startups and services emerging constantly, not all of them will be destined to last, or fit for organizational purpose. A ‘bite sized’ approach that emphasizes small investments and experimentation over all-in bets both tightens focus and makes sure data and AI products can be steadily improved over time. 

Effective use of data ultimately depends on people. Businesses often invest heavily in the early stages of the data value chain, where data is ingested and modeled. But realizing value has far more to do with how effectively that data is absorbed and put into action by teams on the ground. Closing the loop can require change management and workforce training, especially as sentiment towards AI sours. 

Rather than an end-goal, enterprises should aim for data & AI readiness. While companies are generally confident about their levels of AI preparedness, Thoughtworks research indicates some of this confidence may be misplaced. Embarking on structured evaluation of maturity across the multiple dimensions needed to succeed with AI at scale - including data, governance and human experience - will build the organization’s capacity to harness innovations and manage risks as technology evolves.