Livia Serbanescu

The through-line

Everything Livia Serbanescu says circles one claim: forecasting is the core act of finance, and its quality depends less on the sophistication of the model than on the quality of the conversation around it. "Forecast accuracy is key, because that's what's driving good decision making" [2], and accuracy at Microsoft has improved from 3% to 1.5% at company level over roughly a decade, moving from Excel-based processes to machine learning models with embedded AI [2]. But she is consistent that the machinery is the easier half. "You can't have accuracy without a good quality conversation. Our models are amazing and they can get better and better, but it's that level of qualitative insights and quality conversation you have with the business that's always going to be the differentiator" [2].

The second recurring idea is that the finance role itself is changing shape rather than acquiring new tools. "The CFO is no longer in a sense the financial officer, is the chief future officer" [2], and AI is the forcing function: "AI is not the productivity tool that you just layer on top of existing processes. It's changing the way you need to operate" [2].

On forecast accuracy as the point of finance

The improvement from 3% to 1.5% [2] is presented as the outcome of a decade-long shift in method, not a one-off project. Machine learning models are run centrally, trained on years of data, while local finance teams add market flavor, business insights, and judgment on improvement versus baseline, in a monthly cadence with deeper quarterly work [2]. The granularity matters: forecast data is analysed at contract level, by month and by business segment, with cloud data lakes feeding the models and Power BI handling visualisation [2]. She identifies the availability of massive data as the key enabler of the whole approach [2]. The consequence she draws is that forecasting cannot be a periodic finance exercise bolted to the calendar; it has to be embedded into the business rhythm as continuous dialogue [2].

On finance and the business as one conversation

Her sharpest formulation is a warning against the standard mental model. "The CFO, you should not think about this as finance and then the business. It's never going to work out. It's finance and the business together in a dialogue" [1][2]. Siloing finance from the business is, in her framing, the first crack that appears in any forecasting process [2]. This is why she treats the local layer of the forecast as substantive rather than cosmetic: the central model supplies the baseline, and the judgment about whether a number represents genuine improvement comes from people talking to the business [2]. The models can keep getting better and it will not change where the differentiation sits [2].

On AI requiring redesign, not layering

She rejects the productivity-tool framing of AI outright. It should not be layered on top of existing forecast processes; the process has to be completely redesigned AI-first [2]. The claim is about operating model rather than tooling: "It's changing the way you need to operate" [2]. This connects directly to her view of the CFO's remit, where questions such as company-wide AI and Copilot adoption now land on the finance leader's desk as decisions to be made [2].

On the CFO as chief future officer

The custodian-of-financials definition is one she treats as finished. "The CFO is no longer in a sense the financial officer, is the chief future officer" [2]. The temporal discipline she attaches to that comes from Satya: "finance should think in decades and operate in quarters" [2]. The role expansion is not aspirational in her telling; it is already visible in the kind of decisions finance owns, including enterprise-wide AI and Copilot adoption [2].

On "learn it all" culture

She returns repeatedly to the cultural shift: "We kind of shifted as a company from we knew it all to we learn it all. And it is so visible in the way it adds value, the way we operate" [1][2]. She credits it as driven from the top, and identifies its practical effect on finance specifically: it empowered finance teams and earned finance a seat at the table beyond reporting [2]. The link back to forecasting is direct, since a learn-it-all posture is what makes a finance team willing to hold an open conversation about a number rather than defend it.

On agility under pressure

The pandemic serves as her reference case for what finance can do when the cycle breaks. Microsoft redrew a full budgeting cycle for Western Europe within a few weeks, using fast trial-and-error decisions with post-mortems attached [2]. She describes it as formative for how she thinks about finance agility [2], which fits the wider argument: forecasting is a running dialogue that can be re-run quickly, not an annual artefact.

Takeaways

• Treat forecast accuracy as the driver of decision quality, not as a reporting metric: Microsoft moved from 3% to 1.5% accuracy over roughly a decade [2].
• Run models centrally on years of data, then have local finance teams add market flavor and judgment on improvement versus baseline, monthly and more deeply quarterly [2].
• Forecast at contract level, by month and business segment, with cloud data lakes feeding the models and Power BI for visualisation; data availability is the enabler [2].
• Never structure the process as "finance and then the business": "It's never going to work out. It's finance and the business together in a dialogue" [1][2].
• Watch for siloing as the first crack in any forecasting process, and embed forecasting into the business rhythm as continuous dialogue [2].
• Redesign finance processes AI-first rather than layering AI on: "It's changing the way you need to operate" [2].
• Adopt the "learn it all" posture over "know it all"; at Microsoft it was driven from the top and won finance a seat at the table beyond reporting [1][2].
• Use "think in decades and operate in quarters" as the CFO's time horizon, and expect to own decisions such as company-wide AI and Copilot adoption [2].