{"id":16291,"date":"2026-04-23T12:37:30","date_gmt":"2026-04-23T11:37:30","guid":{"rendered":"https:\/\/bdg.io\/uk\/?p=16291"},"modified":"2026-04-23T15:09:21","modified_gmt":"2026-04-23T14:09:21","slug":"will-ai-replace-the-cfo","status":"publish","type":"post","link":"https:\/\/bdg.io\/uk\/will-ai-replace-the-cfo\/","title":{"rendered":"Will AI replace the CFO? Not quite."},"content":{"rendered":"\t\t
\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

Agentic AI is fundamentally reshaping the finance function \u2013 faster than most people expect. CFO Kyanush Kay and Jochen He\u00dfler (bdg) discuss the reality of Excel, the AI hype, and what it all means for the CFO role.<\/h4>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

TL;DR<\/h4>
  • Agentic AI is not a revolution, but the next stage of a transformation that has long been underway.<\/li>
  • The technology isn\u2019t the problem. The data is.<\/li>
  • The CFO won\u2019t be replaced \u2013 but someone using AI will replace those who don\u2019t.<\/li>
  • Starting beats waiting. Always.<\/li><\/ul>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
    \n\t\t\t\t
    \n\t\t\t\t\t\t\t\t\t

    You walk into the office in the morning, and the monthly report is already on your desk. Variance analysis complete, outliers commented on, an initial assessment prepared for the board. No copy-paste, no hours spent wrestling with data.<\/p>

    This isn\u2019t utopia. It is the current state of play \u2013 at least for those making use of it.<\/strong><\/p>

    In the bdg webinar From Excel to Agentic AI<\/em><\/strong>, CFO Kyanush Kay<\/strong> and Jochen He\u00dfler<\/strong>, VP Growth at bdg, discussed exactly this: where finance teams truly stand today, what is currently changing, and what role the CFO plays in it \u2013 or may no longer play.<\/p>

    “AI won\u2019t take my job”<\/h3>

    Let\u2019s start with the question many people ask themselves, but rarely out loud: Will Agentic AI make the CFO obsolete?<\/strong><\/p>

    Kyanush Kay has a clear answer:<\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t

    \n\t\t\t\t
    \n\t\t\t\t\t\t\t
    \n\t\t\t

    \n\t\t\t\t\"AI won\u2019t take my job. A person using AI will take my job.\"\t\t\t<\/p>\n\t\t\t\t\t\t\t

    \n\t\t\t\t\t\t\t\t\t\t\tKyanush Kay, CFO<\/cite>\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t<\/blockquote>\n\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
    \n\t\t\t\t
    \n\t\t\t\t\t\t\t\t\t

    Kay comes from a controlling background \u2013 Air Berlin, start-ups, EPM software \u2013 and understands both sides: the operational reality of finance and what technology is already capable of delivering today.<\/p>

    His assessment is pragmatic: any CFO primarily focused on preparing the past \u2013 reporting, closing, historical analysis \u2013 is in an increasingly vulnerable position. This work can be automated.<\/p>

    By contrast, the CFO who runs scenarios, prepares decisions and steers the business has little to fear. Quite the opposite.<\/p>

    A shift 15 years in the making<\/h3>

    The evolution of finance \u2013 from number provider to business partner \u2013 is nothing new. It has been a core theme in controlling for at least 15 years. And yet, in many organisations, it has not truly materialised.<\/p>

    Why? Not due to a lack of intent, but because of day-to-day reality. Anyone spending their time merging data from multiple systems, consolidating files and manually building reports simply lacks the capacity to think strategically.<\/p>

    Agentic AI can break this bottleneck \u2013 but only if it is actually used.<\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t

    \n\t\t\t\t
    \n\t\t\t\t\t\t\t\t\t

    \"\ud83d\udca1\" The reality in finance departments<\/h3>
    • 99 % of CFOs<\/strong> report significant challenges in their finance function
      (BARC CFO Agenda 2025, n=194)<\/li>
    • 50 % of companies<\/strong> aim to reduce their reliance on spreadsheets
      (BARC CFO Agenda 2025)<\/li>
    • 35 %<\/strong> still rely on manual data transfers between planning and consolidation tools
      (BARC FPM Score 2025)<\/li>
    • 71 % of CFOs<\/strong> face increasing pressure to provide up-to-date steering information
      (Horv\u00e1th CFO Study 2025)<\/li><\/ul>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
      \n\t\t\t\t
      \n\t\t\t\t\t\t\t\t\t

      Where do you really stand? The five stages<\/h3>

      To understand what distinguishes Agentic AI from previous developments, it helps to look at the evolution of analytics \u2013 not as theory, but as an honest self-assessment.<\/p>

      1. Descriptive Analytics \u2013 What happened?<\/strong>
      Classic reporting. Representing the past. Excel remains very much at home here. Most organisations are familiar with this stage \u2013 some with this stage alone.<\/p>

      2. Diagnostic Analytics \u2013 Why did it happen?
      <\/strong>Analysing variances, breaking down drivers, going deeper. Modern EPM systems perform strongly here. Those who have reached this stage already hold a tangible advantage.<\/p>

      3. Predictive Analytics \u2013 What will happen?<\/strong>
      Algorithms, machine learning, automated forecasts. Technologically feasible ffor some time \u2013 yet many organisations struggle to fully adopt it. We will come back to why.<\/p>

      4. Prescriptive Analytics \u2013 What should we do?<\/strong>
      The system provides not only a forecast, but a recommendation. \u201cIncrease price in segment X by 3%.\u201d This requires both a solid predictive foundation and trust in the outcome.<\/p>

      5. Agentic AI \u2013 The agent acts
      <\/strong>No longer just output awaiting interpretation. The agent plans, decides and executes \u2013 within defined boundaries and under human oversight.<\/p>

      Kyanush Kay describes it as a digital junior controller: independently identifying patterns, comparing drivers, flagging outliers and suggesting initial solutions.<\/p>

      Why so many get stuck at predictive<\/h3>

      The honest answer: because you are expected to stand behind numbers you do not fully understand.<\/p>

      An algorithm that produces a forecast without explaining how drivers were weighted, what role randomness played, or where uncertainty lies presents a real challenge in finance.<\/p>

      As a CFO, you stand in front of investors, the board and the executive team. A black box is not an option.<\/p>

      This leads many back to what feels safe: Excel. At least there, they understand what is happening. That is understandable \u2013 but ultimately misguided.<\/p>

      The solution is not to go backwards, but to adopt systems that explain their outputs.<\/p>

      What Agentic AI can already do today<\/h3>

      Data preparation<\/strong>
      Filling gaps, identifying inconsistencies and integrating data from multiple sources \u2013 faster and more reliably than any manual process. This is not future thinking. It is already reality.<\/p>

      Proactive monitoring<\/strong>
      Instead of spending hours searching for anomalies, the agent does it for you. It detects patterns, flags outliers and prepares initial commentary. Teams can focus on thinking, not searching.<\/p>

      Forecasting with statistical confidence<\/strong>
      Forecasts provide not just a number, but a range. If a planner enters a value far outside historical norms, the system flags it: \u201cAre you sure?\u201d
      This improves quality without overriding human judgement.<\/p>

      Scenario analysis with external factors<\/strong>
      Inflation, commodity prices, exchange rates \u2013 external influences can be directly integrated into planning. Scenarios can be created in minutes rather than overnight.<\/p>

      Self-service for business units<\/strong>
      With a solid data foundation, business units can conduct simple analyses themselves. This relieves finance \u2013 and strengthens its role as a partner rather than a service provider.<\/p>

      The prerequisite no one likes to talk about<\/h3>

      None of this works without reliable data.<\/strong><\/p>

      Kyanush Kay puts it bluntly: \u201cPrescriptive or predictive analytics only works when the data is ready.\u201d\u00a0If data is still being copied from multiple systems, versions shared via email and Excel files manually consolidated, Agentic AI will have limited impact \u2013 regardless of the tool.<\/p>

      That is not a reason to wait. It is a reason to start right there.<\/p>

      The most expensive mistake: waiting for the right moment<\/h3>

      A familiar pattern: management discusses Agentic AI, while the reality in departments remains Excel, with a seemingly vast gap in between. The result? Inertia.<\/p>

      \u201cWhere do I even start?\u201d is a valid question. It should not become an excuse.<\/p>

      Jochen He\u00dfler sees this regularly in practice. His advice is clear: start, do not wait. A well-defined initial use case can demonstrate tangible results within days. This creates clarity and builds trust \u2013 neither of which emerges in strategy meetings.<\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t

      \n\t\t\t\t
      \n\t\t\t\t\t\t\t\t\t

      \ud83d\udca1 PRACTICAL INSIGHT<\/h4>

      A mid-sized company started with a single module for monthly P&L planning. Three months later, they added cash flow forecasting.\u00a0Reporting that previously took three days was reduced to two hours.<\/strong><\/p>

      What did the team do with the time gained? They analysed<\/strong> for the first time \u2013 rather than simply preparing data.<\/p>

      Source: bdg project experience<\/em><\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t

      \n\t\t\t\t
      \n\t\t\t\t\t\t\t\t\t

      The hamster wheel will not stop on its own<\/h3>

      Another reason for hesitation: lack of time. Day-to-day operations leave little room for change. That is true \u2013 but it is also a trap.<\/p>

      If time is consumed by constant data preparation, it will remain scarce until that work is automated.\u00a0The solution is not more time. It is a clearly defined first step.<\/p>

      What really changes for the CFO<\/h3>

      Less preparation, more judgement<\/strong>
      When reporting and monitoring are automated, time is freed for what CFOs should be doing: evaluating outcomes, considering options and advising the business.<\/p>

      Faster responses when it matters<\/strong>
      Geopolitical events, commodity prices, interest rate shifts \u2013 those who can model scenarios in minutes rather than weeks make better decisions.<\/p>

      Forecasts you can trust<\/strong>
      Not because AI is infallible, but because it identifies patterns humans miss and makes uncertainty transparent rather than concealing it.<\/p>

      A different role in the room<\/strong>
      A CFO who can tell sales in a budget discussion, \u201cPushing this segment will not pay off \u2013 the margins will not support it,\u201d and back this up with current data rather than a three-week-old report is a very different counterpart.\u00a0That is the difference.<\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t

      \n\t\t\t\t
      \n\t\t\t\t\t\t\t\t\t

      CONCLUSION<\/h3>

      Kyanush Kay’s message is clear: Agentic AI is not a passing trend<\/strong>. It is a stage in an ongoing evolution \u2013 much like Big Data and predictive analytics before it. And further stages will follow.\u00a0Those who do not start today will be at a disadvantage tomorrow.<\/p>

      So, back to the original question \u2013 will the CFO still be needed?\u00a0<\/strong>Yes. But not as the person producing reports.\u00a0As the one making decisions based on them.<\/p>

      Everything else comes down to getting started.<\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t

      \n\t\t\t\t
      \n\t\t\t\t\t\t\t\t\t

      Kyanush Kay is a CFO with a background in controlling, including roles at Air Berlin and in the start-up environment. Jochen He\u00dfler is VP Growth at better decisions group (bdg), an international EPM and BI consultancy.<\/em><\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t

      \n\t\t\t\t
      \n\t\t\t\t\t\t\t\t\t

      Frequently asked questions (FAQ)<\/h3>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
      \n\t\t\t\t
      \n\t\t\t\t\t\t\t
      \n\t\t\t\t\t\t
      \n\t\t\t\t\n\t\t\t\t\t
      What differentiates Agentic AI from the AI features my EPM tool already has? <\/div><\/span>\n\t\t\t\t\t\t\t\n\t\t\t<\/path><\/svg><\/span>\n\t\t\t<\/path><\/svg><\/span>\n\t\t<\/span>\n\n\t\t\t\t\t\t<\/summary>\n\t\t\t\t
      \n\t\t\t\t
      \n\t\t\t\t
      \n\t\t\t\t\t\t\t\t\t

      Traditional AI features deliver output: forecasts, insights, dashboards. An agent goes further \u2013 it acts. It executes steps without the requirement for manual triggers. It’s the difference between an assistant that suggests and one that implements.<\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/details>\n\t\t\t\t\t\t

      \n\t\t\t\t\n\t\t\t\t\t
      Do I need to clean up my data before starting? <\/div><\/span>\n\t\t\t\t\t\t\t\n\t\t\t<\/path><\/svg><\/span>\n\t\t\t<\/path><\/svg><\/span>\n\t\t<\/span>\n\n\t\t\t\t\t\t<\/summary>\n\t\t\t\t
      \n\t\t\t\t
      \n\t\t\t\t
      \n\t\t\t\t\t\t\t\t\t

      Not completely \u2013 but partly, yes. A pragmatic approach: start with the use case where data is already relatively clean. That delivers quick results and shows where to go next.<\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/details>\n\t\t\t\t\t\t

      \n\t\t\t\t\n\t\t\t\t\t
      How do I explain AI-generated forecasts to the board? <\/div><\/span>\n\t\t\t\t\t\t\t\n\t\t\t<\/path><\/svg><\/span>\n\t\t\t<\/path><\/svg><\/span>\n\t\t<\/span>\n\n\t\t\t\t\t\t<\/summary>\n\t\t\t\t
      \n\t\t\t\t
      \n\t\t\t\t
      \n\t\t\t\t\t\t\t\t\t

      By using systems that explain their results: which drivers were weighted how, where uncertainty lies. This isn\u2019t optional \u2013 it\u2019s a requirement. Any system that can\u2019t do this doesn\u2019t belong in finance.<\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/details>\n\t\t\t\t\t\t

      \n\t\t\t\t\n\t\t\t\t\t
      How long does it take to get started? <\/div><\/span>\n\t\t\t\t\t\t\t\n\t\t\t<\/path><\/svg><\/span>\n\t\t\t<\/path><\/svg><\/span>\n\t\t<\/span>\n\n\t\t\t\t\t\t<\/summary>\n\t\t\t\t
      \n\t\t\t\t\t
      \n\t\t\t\t
      \n\t\t\t\t
      \n\t\t\t\t\t\t\t\t\t

      A prototype for a clearly defined use case can be built in a few days. Not a finished system \u2013 but a realistic understanding of what\u2019s possible. And a far better basis for decisions than theory.<\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/details>\n\t\t\t\t\t\t

      \n\t\t\t\t\n\t\t\t\t\t
      What about data security? <\/div><\/span>\n\t\t\t\t\t\t\t\n\t\t\t<\/path><\/svg><\/span>\n\t\t\t<\/path><\/svg><\/span>\n\t\t<\/span>\n\n\t\t\t\t\t\t<\/summary>\n\t\t\t\t
      \n\t\t\t\t\t
      \n\t\t\t\t
      \n\t\t\t\t
      \n\t\t\t\t\t\t\t\t\t

      A question that must be answered before implementation. Where is data processed? Is it used to train models? What about GDPR and industry-specific compliance? Serious providers have clear answers.<\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/details>\n\t\t\t\t\t<\/div>\n\t\t\t\t\t