AI inside a Spherical Cow
I should explain that title. In an earlier post I explained how ‘Assume a spherical cow’ was joke shorthand for a very useful approach to complexity, which is to use drastically simplified models to start with. Then we looked at six core business functions:
- Discover a Thing people need or want
- Create that Thing
- Let the world know what Thing you have created
- Sell the Thing
- Deliver the Thing
- Make sure you can afford to keep going
These are based heavily on the ‘5 Parts of Every Business’ in Josh Kaufman’s book The Personal MBA.
What I would like to do in this post is continue the conversation we started earlier about where in the enterprise AI is making an impact.
First an elephant in the room at the time of writing. The July 2025 report from MIT NANDA, ‘The GenAI Divide. State of AI in Business 2025’ caused some ruckus on LinkedIn, probably because of this line in the executive summary:
Despite $30–40 billion in enterprise investment into GenAI, this report uncovers a surprising result in that 95% of organizations are getting zero return.
Headlines ran on this line alone, and I think it’s a shame because the conversation is more interesting than that 95%. Take for example another quote from the same executive summary:
But these tools primarily enhance individual productivity, not P&L performance.
So it really depends what you are measuring and where - how can the time saved on task A (tactical), so that time can be re-invested in task B (strategic), be captured in a line on the P&L within the measurement window of a study?
Bottom line is that it’s an interesting article that raises some interesting questions about how one should ‘do’ AI, and also an article written by a particular team with a particular point of view on the technology.
Anyway, hopefully the elephant is dealt with and we can go back to the six spherical cows. Let’s have a closer look at these functions by pretending for the time being that they are actual departments. Whether or not they map to a single real department in your company will vary, but we will make some observations as we go.
Cow Departments
The Department of Discover a Thing people need or want
Typical Real Department(s) are:
This function tends to be spread out across R&D, Product Management, Market Research and Corporate Strategy.
Typical Activities you might see are:
- Market research & trend scanning
- Customer discovery interviews
- Competitor analysis
- Problem framing & opportunity identification
- Early hypothesis testing (MVP concepts, experiments)
Corporate Words you might find to describe this ‘department’:
Market Research & Opportunity Identification (Customer Insights, Product Discovery, Front-End Innovation)
The Department of Create that Thing
Typical Real Department(s)
This function is often centralized around Engineering, other departments might be Product Development, Design/UX
Activities
- Product design & engineering
- Prototyping and iteration
- Usability testing
- Scaling from prototype to production
- Preparing product documentation and specs
Corporate Descriptors
Product Development & Engineering (R&D, Design, Operations)## Let the world know about the Thing you created
The Department of Let the world know what Thing you have created
Typical Real Department(s)
Definitely Marketing (Brand, Demand Gen, PR, Content), Communications, Product Marketing
Activities
- Crafting the brand message and narrative
- Running awareness and demand campaigns
- Lead generation and nurturing
- Customer education (content, webinars, collateral)
- Public relations and external communications
Corporate Descriptors
Marketing & Communications (Brand, Demand Generation, PR, Go-to-Market)
The Department of Sell the Thing
Typical Real Department(s)
Definitely Sales, Business Development, E-commerce (for digital firms)
Activities
- Qualifying leads and managing pipeline
- Customer engagement (calls, demos, proposals)
- Negotiation and pricing
- Closing deals/contracts
- Capturing customer feedback
Corporate Descriptors
Sales & Business Development (Commercial, Revenue Generation, Account Management)
The Department of Deliver the Thing
Typical Department(s)
Spread across Operations, Supply Chain, Professional Services, Customer Service, Support
Activities
- Order acceptance
- Order fulfillment & logistics
- Customer support/helpdesk
- Maintenance or updates
- Collecting customer satisfaction and usage data
Corporate Descriptors
Operations & Customer Success (Fulfillment, Supply Chain, Service Delivery, Support)
The Department of Make sure you can afford to keep going
Typical Department(s)
Finance & Accounting, Legal & Compliance, Executive Leadership
Activities
- Managing cash flow and profitability
- Resource allocation and budgeting
- Risk and compliance management
- Strategic planning & investment decisions
- Reporting financial performance to stakeholders
Corporate Descriptors
Finance & Corporate Governance (Financial Management, Strategic Planning, Risk & Compliance)
An observation we can make is that some of these functions are very centralized, with the real departments being named for the function (Marketing, Sales, Finance). Other functions are spread out and don’t have an appropriately named home (No such department of Discover a Thing people need or want). Sometimes a business can struggle with a function that does not belong in a single real department.
AI Impact on Functions
Next we want to see how much of an impact Generative AI can make on each of these functions. We can turn to some other recent research to see what we can learn.
An Anthropic article called ‘Which Economic Tasks are Performed with AI?’ looked at 4 million anonymized conversations with Claude to see how the tool was being used across different work tasks and job types. The research suggests that usage is concentrated around software development and writing.
From OpenAI there is an article titled ‘How People Use ChatGPT’. It takes a different approach to the Anthropic paper in terms of task description and whether or not the tasks are for work. This research also shows writing tasks dominating work tasks, but for second place surfaced ‘Asking’ (information-seeking/decision support).
Between the two then we have Writing, Software Development and Information-Seeking/Decision Support. Just for interest lets compare with the Robot Head Scores we got in our last post
| Tool/Work Level | Deep Work & Creative | Collaborative | Administrative & Routine |
|---|---|---|---|
| Word | 🤖 🤖 🤖 | 🤖 🤖 | 🤖 🤖 🤖 |
| Excel | 🤖 🤖 | 🤖 | 🤖 🤖 🤖 🤖 |
| Email (Outlook) | 🤖 🤖 | 🤖 🤖 🤖 🤖 | 🤖 🤖 🤖 🤖 |
| PowerPoint | 🤖 🤖 | 🤖 🤖 | 🤖 🤖 |
| Teams Calls | 🤖 | 🤖 🤖 🤖 🤖 | 🤖 🤖 🤖 |
| Teams Messages | 🤖 | 🤖 🤖 🤖 | 🤖 🤖 🤖 |
| Just Thinking | 🤖 | 🤖 | 🤖 |
Word scored eight robot heads, Excel seven, Email got ten, PowerPoint six, Teams Calls had eight, Teams Messages seven and thinking got three. The writing oriented tools (Word, Email) scored highly, which makes sense in the context of both the Anthropic and OpenAI research. Thinking probably needs a boost in robot heads, perhaps, and our previous analysis did not include software development tools.
I am also going to leave software development out of this next step, and see if we can estimate amount of Writing and ‘Asking’ (Information-Seeking/Decision Support) in each of our Cow Departments:
| Tool/AI Use Type | Writing | Asking |
|---|---|---|
| Discover a Thing people need or want | 🤖 🤖 🤖 | 🤖 🤖 🤖 🤖 🤖 |
| Create that Thing | 🤖 🤖 | 🤖 🤖 |
| Let the world know what you have created | 🤖 🤖 🤖 🤖 🤖 | 🤖 🤖 |
| Sell the Thing | 🤖 🤖 🤖 🤖 | 🤖 🤖 |
| Deliver the Thing | 🤖 🤖 🤖 🤖 | 🤖 |
| Make sure you can afford to keep going | 🤖 🤖 | 🤖 🤖 |
These estimates suggest that GenAI would help The Department of Discover a Thing People Need Or Want the most, which as we noted above is a function spread over multiple real departments. A close second is the real department of Marketing.
A key observation here is that GenAI utility may not necessarily be focused in a single department, but rather more in a task our outcome.
What are we supposed to do with this? Well, one suggestion would be to perform a similar analysis on your own work and in your own business. Your mileage will definitely vary. The other thing we can do is put this next to some earlier discussions on the same topic:
- Create a simplified model of your business (Spherical Cows)
- Think about the kinds of Knowledge Work being done by individuals (Will AI Take Away My Work?)
- For individual tasks, consider Ethan Mollick et-als Jagged Frontier, and how it is impacted by tacit knowledge (The Jagged Frontier Is Personal)
- Consider the tools used for work and where GenAI fits (Working in the Matrix)
- Look at the tasks or outcomes, rather than departments (This Post)
One of the failure points discussed by MIT NANDA is that GenAI tools alone (aps on the desktop or online) lack memory - users complain of having to re-teach the tool each time. This is the domain of Workflows and Agents, which will will visit in future posts.
https://mlq.ai/media/quarterly_decks/v0.1_State_of_AI_in_Business_2025_Report.pdf
https://assets.anthropic.com/m/2e23255f1e84ca97/original/Economic_Tasks_AI_Paper.pdf
https://www.nber.org/system/files/working_papers/w34255/w34255.pdf