Finance Superpower

This week, we take a break from Strategy. This post was idea inspired by Secret CFO's recent newsletter & playbook related to AI for CFOs & Roger L Martin's write ups on Strategy & AI. These are :

1. Industrial Tech CFO - Secret CFO
2. AI for CFOs (Parts 1 to 5) from Secret CFO Playbook Series
3. Becoming an AI Augmented Enterprise - Roger L Martin
4. A Leader’s Role in Fostering AI Superpowers - Roger L Martin

Being involved in multiple Finance System Implementations & Transformation Projects, I always loved jumping into rabbit holes to understand how software & systems work and how can we as finance team, enhance or augment our value add. I believe finance can use systems as tools to deliver huge value adds :

1. Finance, as a department, is the only department which can partner with business to identify whether the organization strategy is working & on the right track.
2. Knowledge of industry, competition & external factors gives context & can help provided guided recommendations.
3. Understanding how to dive deep into the sea of data, convert the data into insights & framing them to stakeholders in a way they can take timely decisions is a competitive advantage.

Understanding & picking the right tools & software for this is critical & more so with the tools becoming better & better.

Excel & Software

Till now, we have all seen or used (still use!!) tech mostly as tools for speeding up our work or in other words, productivity tools. And for finance professionals, the first thing that comes to mind is.... Microsoft Excel .

I remember the first time I laid my eyes on excel & immediately fell in love with it. Numbers used to (& still does!!) come alive. We could put in rows & columns of data, run scenarios & use formulas. It was love at first model.

As the years passed, tools like Excel &. other software improved. Finance professionals could generate better models & scenario outcomes.

Depending on the industry, finance professionals deal with industry specific software like ERP, core banking systems or treasury solutions. The companies that serve these tools to customers are mostly software or or SaaS (Software as a Service) companies with their

1. Business & operating models,
2. Revenue & cost drivers.

The banks or industrial enterprises using these software have their own

1. Business & operating models as well as
2. Cost & revenue drivers or unit economics.

The software used by these companies (ERP, core banking solutions etc) form part of their capex (capital expenses) or opex (operating expenses) to

1. Improve / introduce new products &/or sell better services to customers &
2. Improve or enhance internal operations & processes.

The context behind the processes or operations were held & updated by the humans running the show while the tools provided the means & output.

The business models of the industries were not impacted by these tools per se. But the way business was done changed beyond recognition. For e.g., storing data in cloud instead of on-prem hardware, storing customer information in databases instead of manual registers etc.

The Balance Sheet & Income Statement had additional lines incorporating these capex spends (for new software implementations) or cost lines (for software subscriptions).

Artificial Intelligence (AI)

From 2022 onwards, a new paradigm was introduced - Generative Artificial Intelligence or Generative AI. It started with chats, image generation, video generation etc & now evolved into agents.

Lots of applications or enterprise software now incorporate AI into it's core design to enhance service offerings or user experience.

Now, in terms of tool usage, what has changed?

Instead of humans using tools like Word, Excel etc to do some work to get a desired output, we just "prompt" an LLM to "do" the work with proper context & clear steps with objectives. The LLM uses a set of tools (e.g., Python Libraries, Excel skills etc) to carry out the task & provide the desired output.

Earlier, the humans held the context & knowledge. The tools were used to test scenarios & assumptions using that context / knowledge & we get the desired output. Now, that context is provided to the agent & it generates the output using the tools.

The earlier tools process data while we held the context & objective to evaluate the output. If we provide the LLM the context with clear steps & objectives, the AI generates the output.

Secret CFO has put it best - AI is a model reasoning through a problem using context, working out what to do next without being explicitly told (probabilistic)

This context & the process can also be put as a skill which the agent can use repeatedly to generate the desired output. This is automation of a repetitive process using AI.

We will not get into the hallucination or second level review by humans at this stage for now as models are not advanced yet to generate reliable outputs on their own without human review. But the direction is unmistakable - LLMs are getting better at what they do & any work which may require verification now may be independently done by an LLM in the near future (at the rate at which LLMs are developing currently).

So, what changed from the legacy ERP or Excel tools that we used for ages from the current AI paradigm? I see this change from two lenses :

1. AI as a personal augmentation tool
2. AI changing unit economics

AI as a personal augmentation tool

Thera are lots of write ups & articles where AI has enhanced or improved workflows especially where agentic AI is involved, especially research. Roger L Martin has extensively written about how AI can be a superpower to augment human capacity.

He wrote that to realize the potential value & make AI an augmentation superpower, there are five levels to consider - Research, Task Automation, skill leveling up, Blindspot detection & counterbalancing Groupthink.

1. First Level - How AI Augments knowledge & research - "AI is now a common first-pass research assistant. It can map a space, surface sources, and draft a structured brief in minutes. Sophisticated deep research tools incorporated into AI products help analysts quickly collect information that would have earlier taken days."
2. Second Level - How AI Augments through Task Automation - "Beyond research, AI can quietly take work off the plate of busy leaders, leveraging their time for other higher-value activities."
3. Third Level - How AI Augments through Skill Leveling-Up - "AI is a force multiplier for uneven skill profiles. Very few of us have completely consistent skill levels across the spectrum required in our jobs. Each of us has our stronger and weaker skill areas. AI can be used to level-up those weaker skill areas to give us a more consistent skill set through AI augmentation."
4. Fourth Level - How AI augments through blind spot detection - "AI is relentless at check-listing the things we forget. Even highly skilled professionals need checklists to avoid blindspots. But AI can go far beyond a simple standard checklist. Point AI at a plan and ask: What’s missing? What could fail? It will probe dependencies, non-obvious stakeholders, compliance constraints, and capacity cliffs. It will test for coverage gaps."
5. Fifth Level - How AI augments through counter balancing Groupthink - "Groupthink is a well-documented problem by which teams converge too quickly on suboptimal decisions. To counterbalance groupthink, leaders can assign AI the role of devil’s advocate. Prompt it to argue the strongest opposing case, quantify downside scenarios, and stress-test assumptions."

AI Changing Unit Economics

For the purpose of this discussion, we will divide enterprises into two (From Roger L Martin) :

1. AI Enterprises - Companies whose product is AI like Open AI, Anthropic etc.
2. AI Augmented Enterprises - Companies which use AI to augment their strategy.

The unit economics impact is different for both the above types of enterprises. For AI enterprises, the only way to enhance model capabilities is to increase compute power & that is a major bottleneck (E.g., Anthropic buying compute from SpaceX-AI - Source).

Figuring out how to allocate compute to activities in line with strategy (E.g., for Anthropic - use the compute internally for model testing or externally for better customer usage) requires a different mindset from heavy industries than tech as it is a long term commitment. Krishna Rao, CFO of Anthropic lays it as below in his podcast with Patrick O'Shaughnessy :

1. "The compute that we procure, it's the lifeblood of our business. It is the most important thing in the company. It's like the canvas on which everything else gets built.
2. And then as we go out and do these deals to procure compute, flexibility is really important to us. And so we build that flexibility into the deals themselves. We build that flexibility into how we use the compute as well.
3. Because the way in which we bridge from a position we are today to where we want to go when the business is growing exponentially is to use that compute as efficiently as possible. I would say I spend 30 or 40% of my time on compute, even today.
4. Importantly, we basically want to utilize each chip to its best purpose within the company."

The amount of capex budget for AI enterprises is mind-boggling with long pay off periods.

Combined 2026 infrastructure spend by Microsoft, Amazon, Alphabet & Meta is now guided toward ~$700–725 Billion — roughly double 2025's ~$383 Billion — underscoring the scale of the capex commitment. (Source : here)

Some questions to ponder :

1. By when would these companies recoup their huge capex costs?
2. How should they price their models in a way that is sustainable?
3. How does compute costs vary as the business scales? Is there a linear relationship or does it follow another logic?

For Finance professionals, capital allocation decisions are based on growth & revenue forecasts. Implementing AI does not guarantee either but involves huge outlays. AI is an evolving tool which is highly capable but still not there yet in terms of delivering reliable outputs. There is a learning curve to understanding & implementing AI in organizations.

Secret CFO has put it clearly - "You need to allocate capital (money and time) to AI implementation and treat it as a learning budget. Then scale your spend appetite as you learn what works"

For AI augmented enterprises, AI usage shows up in the cost line as token usage costs which can be high (For example, Microsoft cancelled most internal Claude Code licenses in its Experiences & Devices division (effective June 30, 2026) after token-based billing burned through the division's annual AI budget far ahead of schedule).

The cost is variable but not in line with sales or revenue but in line with usage which is independent of revenue generating activities. Hence, forecasting that cost line needs more context or data. What benefits are being generating from token usage?

Token cost is the line item finance doesn't yet understand and can't yet forecast — and it scales invisibly with usage. A poorly designed workflow can "quietly rack up material spend." (From Secret CFO)

Roger L Martin gave an example of JP Morgan using AI to augment their strategy where "the firm has deployed internal generative-AI tools that allow bankers, analysts, and other professionals to query large volumes of internal documents and data using natural language. In investment banking, for example, AI tools help analysts review financial disclosures, extract key information from documents, and prepare draft materials for client work. Importantly, the AI does not replace the professional judgment of the banker or analyst; instead, it reduces the time spent on manual information gathering and document review."

Embracing the New

Now most of us use AI one way or the other. Bolting AI into an existing process without revisiting it to make it more efficient with AI will not produce any tangible benefits.

The same thing happened with the introduction of ERPs or core banking systems in banks. Before these systems, manual ledgers were maintained for customer data & GL accounts. Processing & accessing these records required a set of processes & controls entirely different from post introduction of ERPs where customer data & GLs are maintained in databases requiring entirely new processes for access & processing of information.

Similarly with AI, it is not about only replacing an existing workflow. It is about looking at it from First Principles. The most important requirement is an open mind to embrace the new.