Senior Director of LinkedIn’s CFO Systems Says AI is Changing Finance from a Reporting Function to an Operating System

My journey didn’t start with AI. It started with frustration. Frustration of sitting in rooms where two systems showed two different numbers. Frustration of finance teams spending more time reconciling the past than shaping the future. Frustration of seeing incredibly smart finance leaders constrained by fragmented systems and delayed insights.

AI is a forcing function. It’s exposing everything that was already broken:

• Rigid processes
• Delayed insights
• Overdependence on manual judgment
• Systems designed for control, not intelligence

What excites me about this moment is not automation — it’s amplification. For the first time, finance leaders can move from hindsight to foresight, from reporting to influencing, and from efficiency to effectiveness.

Today, my focus is on building what I call a Finance Technology Strategy & Platforms model — where architecture, product thinking, and AI come together to create a system that doesn’t just track the business, but actively shapes it. Because the future of finance isn’t about closing books faster. It’s about helping the business make better decisions before the numbers are even written.

Pankaj Shares

AI is a forcing function. It’s exposing everything that was already broken.

Leading LinkedIn's CFO organization

I lead Finance Technology & Platforms within LinkedIn’s CFO organization — a global, multi-billion-dollar business operating at scale across geographies, products, and regulatory environments.

The complexity is what you’d expect at this stage:

• Multiple revenue streams across Talent Solutions, Marketing, and Premium
• A global footprint with region-specific compliance, tax, and operating models
• Deep integration across finance, sales, HR, and product systems

But the real challenge isn’t scale. It’s interconnectedness.

Finance today sits at the center of every critical decision:

• Planning and forecasting across dynamic business cycles
• Managing capital allocation with increasing precision
• Supporting growth while maintaining discipline and controls

We operate less like a traditional finance tech team and more like a product organization embedded inside finance — accountable not just for systems, but for outcomes. Because at this scale, finance isn’t just about managing numbers. It’s about shaping how the company makes decisions at speed, with clarity, and with confidence.

Where CFOs Must Start with AI

Most CFOs start their AI journey thinking about use cases, tools, vendors, and ROI. But AI doesn't operate in isolation. It sits on top of your data, your processes, your operating model, your decision discipline.

If those are fragmented, AI doesn’t simplify them; it amplifies the fragmentation. You don’t get better answers. You get faster wrong answers.

So, before asking: “Where can we apply AI?” Ask: “Is our finance system designed to produce consistent, decision-ready data?” Because AI is not a strategy. It’s an accelerator of whatever system you already have.

Focus on:

• A clear source of truth (definitions, metrics, governance)
• Standardized processes across finance domains
• A strong data and architecture backbone
• A culture that values questioning outputs, not accepting them

How AI integration Enhances Finance Workflows

Here are the improvements I've seen with AI:

• Cycle time reduction (30–60%): Variance analysis that used to take 2–3 days now happens in a few hours, and monthly close support activities (analysis, reconciliations support) have been reduced significantly. This is where AI delivers immediately — speed.
• Forecast accuracy (5–15% improvement in near-term horizons): Short-term forecasts (4–8 weeks) improved meaningfully; AI picked up leading indicators humans were ignoring (usage trends, behavioral signals). But important nuance: Accuracy improved more in stable environments, less in volatile ones.
• Productivity lift (20–40% analyst capacity unlocked): Analysts spend less time gathering data and more time interpreting it. One analyst can now cover what previously needed 1.5-2 FTEs. This doesn’t always translate to cost reduction — it translates to better allocation of talent.
• Earlier risk detection (days → near real-time): Cash flow risks, anomalies, and outliers identified faster; reduced “surprise factor” in business reviews. This is subtle but powerful — fewer last-minute escalations.

Here's what didn't improve (or got worse):

• Overconfidence in outputs: AI-generated narratives sound right even when they’re wrong. Early on, teams started trusting outputs without challenge, which is more dangerous than manual errors.
• False precision: AI creates an illusion of certainty: “Forecast says 97.2% confidence”. But finance decisions are not probabilistic exercises alone — they involve strategy, timing, and context.
• Data dependency exposed weak foundations: AI didn’t fix data problems. It magnified them: inconsistent hierarchies, poor master data, fragmented systems. Garbage in, amplified garbage out.
• Limited impact on long-term planning: AI struggled in annual planning, strategic scenarios, and capital allocation decisions. Because these are not pattern problems — they are judgment problems.

The uncomfortable truth is that AI didn’t reduce finance costs as much as people expected. It shifted the cost from labor to compute, from manual effort to model governance, and from analysts to engineers and data infrastructure.

AI-generated narratives sound right even when they’re wrong. Early on, teams started trusting outputs without challenge, which is more dangerous than manual errors.

Share This Quote on:

• Share on Twitter
• Share on LinkedIn
• Share on Facebook

Pankaj PrasoonOpens new window

Senior Director, CFO Systems & Technology, LinkedIn

Why AI Informs Finance Decisions, but Humans Lead Judgment

Here's the simplest way to look at the split between AI tasks and human tasks: AI should inform patterns; humans should make commitments.

I rely on AI in areas where scale, speed, and pattern recognition matter more than judgment:

• Forecasting (short-term and rolling): AI is excellent at ingesting multiple signals — historicals, seasonality, pipeline data, external indicators — and continuously updating forecasts. It removes the bias of static, calendar-driven planning.
• Variance analysis: AI can break down variances across thousands of dimensions instantly: what changed, where it changed, and what likely caused it. This used to take days. Now it’s near real-time.
• Cash flow patterning: Not decision-making, but signal detection: payment behavior trends, collection risks, and liquidity patterns. AI helps surface risks earlier than traditional reports.
• Risk detection (operational and financial): AI is very effective in identifying anomalies: unusual transactions, control deviations, outlier behaviors. It doesn’t “understand risk,” but it flags where to look.

But I keep it explicitly human in areas where context, trade-offs, and accountability matter:

• Capital allocation: No model understands strategy, timing, or competitive positioning the way leadership does. AI can inform scenarios — but cannot decide where to place bets.
• Long-range planning and budgeting: Budgets are not forecasts — they are commitments. They reflect intent, priorities, and trade-offs across the organization.
• Final risk decisions: AI can highlight anomalies, but what is acceptable risk? What trade-off is worth taking? Those are leadership calls, not model outputs.
• Business partnering and influence: Finance is not just about numbers — it’s about influencing decisions: Challenging assumptions, reading the room, and understanding the intent behind asks. AI has no context for human dynamics.

If AI is making your decisions, you’ve gone too far. If AI is not informing your decisions, you’re not using it enough.

Why AI Hasn't Fully Automated Finance

The narrative was: “AI will automate finance.”

That didn't happen. It automated parts of workflows, but introduced new layers:

• Model validation
• Exception handling
• Governance and audit requirements

The reality is that you don’t remove humans — you shift them to oversight roles.

How AI Adoption Is a Behavior Problem

Speaking of humans, AI adoption is not a technology problem — it’s a behavior problem.

Initially, I assumed that if we built the right models, people would use them → insights would be faster → decisions would improve → insights would happen faster → accuracy would improve → and trust would follow.

All logical. All incomplete. We built something technically strong: better variance detection, faster insights, and cleaner narratives. And yet, adoption was uneven.

Because finance teams didn’t struggle with lack of insight. They struggled with letting go of how they’ve always done.

Analysts still rebuilt models in Excel "just to validate." Leaders trusted what they understood, not what the model suggested. Reviews were still anchored on old formats, not new insights.

The system changed. The behavior didn’t — at least not immediately.

I wish I had done this:

• Start with operating model, not model accuracy: Define how decisions will change, what meetings will look like, what gets removed, not just added. Instead of asking: “Is the model right?” Ask: “Will this replace something people do today?”
• Design for trust, not just intelligence: Trust doesn’t come from accuracy metrics or technical sophistication. It comes from transparency, explainability, and consistency over time. I would have invested earlier in making outputs challengeable, not just impressive.
• Over-index on change management: We treated it like a system rollout. It wasn’t. It was a mindset shift from building analysis to reviewing insights, from control to collaboration with AI, and from certainty to probabilistic thinking.
• Fix data and definitions first (more than we did): We knew data mattered. We underestimated how much. AI doesn’t tolerate ambiguity.

I could have avoided:

• Rework cycles where teams rebuilt what AI already produced
• Over-engineering models that didn’t change decisions
• Early skepticism that was actually lack of trust, not lack of accuracy
• The false belief that better output = automatic adoption

Trust doesn’t come from accuracy metrics or technical sophistication. It comes from transparency, explainability, and consistency over time.

Share This Quote on:

• Share on Twitter
• Share on LinkedIn
• Share on Facebook

Pankaj PrasoonOpens new window

Senior Director, CFO Systems & Technology, LinkedIn

Why AI Reshapes Financial Control Assumptions

In finance, we’re trained to think: more controls = more accuracy; more approvals = less risk; more structure = better outcomes. So, we built systems that were: rigid, sequential, and designed to prevent mistakes.

We were optimizing for predictability, not responsiveness; for auditability, not always for decision quality.

It worked — but at a cost: speed, adaptability, and sometimes even clarity.

AI doesn’t operate well in rigid environments. It thrives on: continuous data flows, dynamic inputs, and real-time signals. It showed me that control doesn’t come from restriction — it comes from visibility.

I had to shift from:

• Preventing errors upfront → Detecting and correcting them quickly
• Periodic reviews (monthly/quarterly) → Continuous monitoring
• Human checkpoints everywhere → Exception-based intervention

Here's where this showed up:

• Forecasting: moved from static, calendar-driven forecasts → continuously updated, signal-driven forecasts
• Financial controls: reduced heavy pre-approvals → increased real-time anomaly detection and post-event validation
• Capital allocation: moved from certainty → probabilistic, scenario-based thinking

The shift is simple, but hard. You have to go from control through restriction to control through intelligence and visibility. And once you see that, you can’t go back. Because the goal of finance isn’t to prevent every mistake. It’s to enable better decisions, faster — with enough control, not maximum control.

Why CFOs Should Redesign Forecasting

Most finance organizations still run forecasting like it’s 2005: monthly cycles, static assumptions, heavy manual consolidation, and backward-looking reviews.

AI doesn’t fit into that model.

We shifted from calendar-driven forecasting, monthly updates, post-close variance analysis, backward-looking reviews, and heavy prep work to signal-driven forecasting:

• Continuous forecasting layer: real-time signals and continuously updated forecasts
• AI-first variance detection: system flags changes and finance reviews exceptions
• Decision-oriented reviews: Shift from “What happened?” to “What will happen?”

Here are the results:

• Time to insight ↓ ~50%
• Forecast relevance ↑ significantly
• Finance role shifted to decision partner.
• Analyst productivity ↑ 30–40%

Redesign forecasting into a continuous, signal-driven system. Because once that changes, finance becomes an active operator of the business.

How AI Improves Forecasting and Variance Analysis

It’s still early days for me at LinkedIn, so here's an example from a prior transformation where we embedded AI directly into the forecasting and variance analysis process.

Finance teams were spending a disproportionate amount of time explaining what happened:

• Manual variance analysis across hundreds of cost centers
• Static forecasts updated monthly or quarterly.
• Heavy dependence on Excel models and human pattern recognition
• By the time insights were ready, the business had already moved.

We had data — but no scalable way to interpret it. So, we introduced an AI-driven layer on top of our planning and actuals data that could:

• Automatically detect anomalies and variance drivers across dimensions (region, product, cost category)
• Generate first-level narratives explaining “why” — not just “what.”
• Continuously update forecasts based on leading indicators, not just historical trends.

But the real shift wasn’t the model — it was the operating model:

• Finance moved from building analysis to reviewing and challenging insights.
• Forecasting became more continuous, less calendar-driven
• Business reviews shifted from “what happened last month” to “what will happen next.”

The improvements were clear. Cycle time for variance analysis dropped from days to hours. Forecast accuracy improved because we incorporated real-time signals instead of lagging inputs. And most importantly, finance earned a seat earlier in decision-making, not after the fact.

The biggest learning? AI didn’t replace finance judgment — it forced us to use it where it actually matters. Because the value of finance is not in explaining the past better. It’s in helping the business act on the future sooner.

Why AI Struggles with Variance Analysis Narratives

One area where we saw clear limitations was AI-generated variance analysis narratives tied to financial reporting.

The numbers were right. But the narrative wasn’t always right. For example, the model attributed increased operating expense to “increased hiring in key regions” — plausible, but wrong. The actual driver was one-time accounting reclassification and timing shift in accruals. The model matched patterns, not reality.

So, it was inaccurate, not traceable, and it eroded trust in the team. Here's how we fixed it:

• AI generates signals and hypotheses, not final narratives.
• Every output must be traceable.
• Introduced “confidence + evidence” layers

Human validation became explicit, not assumed.

How CFOs Can Lead AI-driven Finance Transformation

Here's my advice: Don’t chase AI. Fix finance — AI will follow.

• Start with the system, not the tool.
• Redesign work, don’t just automate it.
• Separate signal from judgment.
• Invest in trust before scale.
• Accept that costs will shift, not disappear.
• Lead the behavior change yourself.
• Build a “Finance OS,” not isolated pilots.

AI will not make you a better CFO. But it will expose whether your finance function was designed to scale intelligence in the first place.

Pankaj Shares

Don’t chase AI. Fix finance — AI will follow.

Follow Along

Follow along as Pankaj Prasoon continues to shape the future of finance at LinkedIn. And check out his personal website.

More expert interviews to come on The CFO Club!