摆脱数据困境 构建AI信心(第二卷)

Table of contents Executive summary03 Avoiding the perfect data trap: five keyreminders for AI success Confidence in data: a maturity marker07 Data confidence is a capability: a call to action 14 Executive summary Data governance — including data integration, preparationof data for AI models and training people to use dataresponsibly — is a key lever for advancing AI maturity andmaximising ROI.As organisations advance along the AI maturity continuum data is both the fuel AND the friction.In this latest edition of our AI Pulse Survey series, we explore the state of data quality and readiness to supportAI efforts, uncovering both encouraging advancements andpersistent challenges.For instance, the survey reveals the value of having AI doesn’t exceed expectationsby accident — it does so on thefoundation of trusted data.Peter Mottram (see the following page), the nature of their data challengesshifts — though some issues remain of concern. For thoseat more advanced stages, issues related to data reliability,completeness and accessibility persist — often stemmingfrom unclear definitions, inconsistent taxonomies andfragmented aggregation (i.e., data that isn’t explicitlydefined or classified, or is inferred in ways that can lead toconfusion or errors). While respondents at the higher end ofAI maturity did not rank these issues as their top concerns,challenges like siloed data and integration difficulties — firstidentified in our inaugural AI Pulse Survey — continue toimpact organisations across all levels of maturity.2 confidence in data. It shows that:•Respondents who are “very confident” or “confident” are more likely to report that AI investments have significantlyexceeded their return on investment (ROI) expectations.•Conversely, those respondents who are “not confident”or only “somewhat confident” in the quality of their Enterprise Data and AnalyticsGlobal LeaderAI Leader data are more likely to report that their return on AIinvestments falls below their expectations.This highlights the need to strengthen the foundations ofdata confidence. Issues like bias and misinterpretation can Data governance is the framework of policies, processes and roles that helps ensure data is accurate, consistent, secure and used responsibly — supporting high data quality and compliance across an organisation. transparency-enhancing solutions. These include implementing data standards, data lineage tools, high-quality dashboards and metadata management services. Such measures are foundationalto building a robust data governance framework and mitigating risks associated with poor data quality. When left unchecked, low-quality data can introduce or amplify bias in several ways —including inaccurate labelling, outdated information, imbalanced outcomes and ambiguous interpretation of inputs. understanding and no strategicinitiatives. Key performanceindicators (KPIs) have not yetbeen defined. pilot programmes to assessfeasibility and benefits. Role-based perspectives reveal striking differences in how much people trust the data they work with. IT professionals reportNot all data is trusted equally: why confidence varies by role the highest levels of confidence — unsurprising, given their proximity to the raw data and their role in curating, cleansing andvalidating it. C-suite leaders also express strong confidence, likely because they interact with polished, decision-ready insightsthat have already passed through multiple layers of refinement.Respondents from business functions like sales and customer service report more varied or lower confidence levels. They often Investments in data qualitymust reach front-endsystems — where trust isoften weakest. Strengtheningthese entry points may bethe best opportunity to buildlasting confidence from theground up.Matt McGivern encounter data in its raw or less processed form. These teams are closer to the “data trenches,” where inconsistencies, gapsand quality issues are more visible.This pattern suggests a clear relationship: the more curated and validated the data, the higher the confidence in its reliability.Conversely, roles that interact closely with raw or unrefined data — but aren’t directly involved in its validation — tend to express greater scepticism, likely due to their exposure to inconsistencies before the data is fully processed.Mature AI organisations tend to progress along parallel tracks, making steady — and sometimes simultaneous — gains in bothAI adoption and confidence in their data. The journey from data confusion to AI confidence is not linear. However, with the right foundations in place, including robust governance, cross-functional alignment and a culture of data literacy, organisationscan unlock the full potential of AI to drive meaningful, measurable outcomes. Notable findings 01Progress with AI closely correlates with the quality and management of data.Early-stage AI adoption often begins with not-fit-for-purpose, incomplete or imperfect d