The Real Barriers to Agentic AI in FP&A: Why Technology Is Not the Biggest Challenge_News_北京智达方通科技有限公司

According to industry reports from last year, although many enterprise finance teams have experimented with leveraging artificial intelligence tools and innovative technologies to optimize financial planning strategies, most teams remain constrained during actual execution by issues such as data silos, manual workflows, and insufficient operational maturity. The key insight behind this phenomenon is that technology is readily available; what truly hinders enterprises has never been the technology itself.

Over the past two years, large language models have quietly integrated into the daily work of finance professionals, with applications ranging from analyzing market trends and explaining forecast variances to drafting financial reports and troubleshooting data errors. However, when the focus shifts to agents capable of autonomously executing tasks, triggering workflows, retrieving data from financial systems, running forecasting models, detecting anomalies, and escalating issues for resolution, the number of enterprises that have truly achieved implementation remains extremely low. Most organizations are still stuck in pilot phases, failing to generate substantial business impact. What is the root cause of this phenomenon?

The High Bar for Data Quality

For years, data quality and standardization have been frequently discussed topics in the field of finance transformation. Yet the reality is that many enterprises still struggle with this fundamental aspect. Although current AI technologies possess the ability to process data in multiple formats, their requirement for high data quality remains an unavoidable prerequisite. As artificial intelligence continues to evolve, traditional Generative AI (GenAI) is gradually transitioning towards Agentic AI, which can autonomously plan, execute, and optimize workflows to complete complex tasks. For Agentic AI to truly deliver value, enterprises need to establish unified definitions and a single source of truth across departments, which undoubtedly imposes even higher demands on data quality.

However, the reality is that no enterprise has 100% accurate data, especially large organizations formed through mergers and acquisitions. The key is to identify data that is truly value-driving and ensure it is consistently defined and captured. For enterprises at the current stage, rather than spending vast amounts of time and effort cleaning all data, it is more effective to focus on core data sources that support key decisions. For example, when designing budget consolidation mechanisms, the Intcube EPM system places greater emphasis on implementing rigorous data validation during the consolidation process, automatically collecting and consolidating data from various levels through automated aggregation rules while ensuring data accuracy. The value of this approach lies in not waiting for data to be perfect before acting, but in continuously guaranteeing quality throughout the data flow process via systematic data governance mechanisms.

The Systemic Thinking Required for Agentic AI

As mentioned earlier, the core difference between Agentic AI and traditional AI is that it can not only analyze data and generate insights but also trigger decisions and task execution. The realization of this actionability requires that end-to-end processes be clearly defined and documented—exceptions need to be clarified in advance, escalation paths need to be clear, and a data governance framework needs to be established. If an enterprise cannot explain how a specific piece of data was generated or why a particular workflow was triggered, user trust in the system will quickly erode.

To address this, finance teams can choose a single domain (including areas such as revenue and demand forecasting, operational expense planning, cash flow forecasting, scenario planning, etc.—these areas need to encompass dimensions like profitability impact, data accessibility, process repeatability, governance readiness, and executive support) to delve deeply into and build a vertically and horizontally integrated architecture. This means that implementing Agentic AI should first address the design of the operating model, and only secondarily focus on technical deployment. Enterprises need to treat Agentic AI as part of the operating model, operating collaboratively across FP&A, IT, and business departments, rather than as an isolated technology add-on.

Overcoming Personnel and Cultural Barriers

If data and technical architecture constitute hard constraints, then personnel and culture are equally challenging barriers. When new tools are introduced, teams naturally feel resistance and skepticism. Therefore, enterprises need forward-thinking, proactively developing clear change management plans. Through continuous communication and training, they can help finance professionals understand the value of Agentic AI, alleviate fears about technology replacing them, and foster a cultural atmosphere of "human-machine collaboration." On the other hand, AI transformation should be planned with people at the center, positioning AI as a tool to unlock value rather than a replacement for personnel. Simultaneously, enterprises should invest in upskilling employees, maintain a sustainable pace of progress, and include employee experience as a metric.

Driving AI from Pilot to Production

Another emerging issue is: why do so many AI projects stall at the pilot stage? Often, the problem is not in the technology validation phase but in the transition from validation to production. In practice, the technology is real and reliable, the intent is sincere, but progress stagnates at the transition point between potential and production. Common obstacles include: pilots being embedded in financial models rather than real business processes; success criteria being vague and lacking quantitative metrics aligned with business value; and investment intensity mismatched with transformation ambition.

To bridge the gap from pilot to production, enterprises need to build a complete implementation mechanism. Pilot projects should be designed to closely fit actual business scenarios and have clear, well-defined success criteria to objectively measure pilot effectiveness and provide a basis for subsequent scaling. Simultaneously, enterprises should allocate resources reasonably according to transformation goals, ensuring investment intensity matches the scale and depth of the transformation. Cross-departmental collaboration mechanisms should be established to jointly address technical issues and process bottlenecks encountered during the pilot.

Taking the Intcube EPM system as an example, it has built a relatively complete application matrix for AI function integration. Leveraging the semantic parsing capabilities of the DeepSeek large language model, the system can intelligently identify and extract data items, dimensions, and hierarchies highly relevant to user questions from complete metadata. It supports complex semantic understanding, accurately captures key elements of user intent, and automatically generates optimized query structures. Intcube deeply embeds AI capabilities into the daily operational workflows of FP&A. Users simply describe their needs in natural language, and the system automatically generates the corresponding analytical views. This lowering of the interaction threshold is a crucial prerequisite for Agentic AI to be effectively utilized by business personnel.

The application of Agentic AI in the FP&A field is gradually unfolding. Its essence is not merely a technology upgrade but a reconstruction of the operating model. Data governance, process definition, governance frameworks, and personnel mindset—these non-technical elements determine the extent to which technology value is released. For enterprises, the most pragmatic action at present is not to chase the latest large language models, but to honestly assess their organizational data readiness, process clarity, and cultural receptiveness, choose one truly critical area to cultivate deeply, and integrate artificial intelligence into the team's working rhythm.