How Power Ladder Experts Solve Business Problems, with AI as a Tool
Updated: 6 days ago
Start with one real decision
Imagine a business owner coming to me with this situation.
Customers are ready to order more. A supplier offers a discount for buying a larger lot. Sales are growing. But the cash needed for this purchase has other demands on it, too.
How much should we buy?
Today, AI can answer that question in seconds. But the answer becomes useful only when someone checks three things: does the question match the decision we actually need to make, how reliable is the evidence, and who takes responsibility if the outcome is different from what we expected?
Those three things are the experts' work. AI is a tool that helps them do it faster.
Watch the companion video
Jenny and the Deal Before Dawn is a fictional story illustrating how expert judgment supports an inventory decision.
On 16 October, I attended a Snowflake workshop presented by May Phang, Technical Architect, AI/ML. One slide immediately stayed with me because it reflected what I have repeatedly encountered with clients.
"Every AI is a pipeline - and people own most of it" — May Phang, Snowflake (Note 1)
May divided the process into six steps: Frame, Prompt, Retrieve, Reason, Respond, and Act. She described the role of people closest to the business as "Direct what AI does, for whom, and why."
Many people see AI as a box: put a question in and get an answer out. May showed that there are several steps in between, and most still require human decisions. I have used that idea to redesign Power Ladder's service process, making clear who owns each step and where AI helps.
Clients gain three things. First, they know from day one who is responsible for each part of the work, with no stage where AI makes decisions without oversight. Second, AI is used for work it handles well, such as finding evidence and calculating different scenarios, so work moves faster while retaining the checks. Third, executives can trace a recommendation back through every step: the question it addresses, the evidence behind it, and the person who confirmed it.
Here is how I apply this division of work in Business Play.
Step | Work | Responsibility |
|---|---|---|
Frame | Choose the decision to address | Client + business expert |
Prompt | Provide context and constraints | Business expert |
Retrieve | Find relevant evidence | Data expert + AI |
Reason | Analyse the alternatives | Business and data experts + AI |
Respond | Review the recommendation | Business expert |
Act | Approve, implement, and review | Client + systems integration expert |
The sequence comes from May Phang's presentation. The assignment of responsibilities is Power Ladder's application of it.
AI can help considerably in the middle steps. People still own the beginning and the end.
Why three areas of expertise matter
This inventory purchase may look like a single issue, but it touches strategy, finance, sales and stock data, and the existing purchasing system. One person cannot cover every angle. I therefore divide responsibility into three areas and agree with the client, from day one, who owns each.
Strategy and business: help executives define the decision, question assumptions, and understand how each alternative affects cash, customers, and suppliers.
Data and AI: check that each number means what it does in the business, prepare analysis in the client's Snowflake environment, and explain how much confidence to place in the findings.
Systems integration and delivery: connect the approved decision to existing systems, define who does what and who approves it, and agree when to review it.
These are responsibilities the work requires, not a headcount. The actual team depends on the scope, and some work may involve partners.
These three areas exchange work throughout the process. Business questions become checks for the data team. Constraints the data team finds feed back into the recommendation before the integration team puts it into practice.
Business Play: one decision, three alternatives
Business Play is how I approach different business situations. It starts with the executive's most important issue at that moment, then selects a Play that fits the opportunity or risk ahead.
The client receives more than a single answer. They receive a recommendation they can examine: alternatives, evidence, assumptions, constraints, and a proposed action. Whenever I score a Play, I must explain why, and identify the changes in conditions that would change the recommendation.
For the inventory decision, the Play is Calculated Ambition. It considers two things together: Opportunity and Financial Readiness. How attractive is the market, and do we have the financial capacity to act?
Alternative | What it offers | What to watch |
|---|---|---|
Buy the full lot now | Stock ready to sell; a possible discount | A large cash commitment and the risk of unsold inventory |
Buy in stages | Flexibility and less pressure on cash | Potentially higher unit costs or temporary stock shortages |
Wait | Preserve cash while gathering more evidence | Potential lost sales |
An illustrative example. The appropriate choice depends on evidence, financial capacity, and supplier terms.
Real case: Business Play at Belly Thailand
Belly Thailand, the author's glamping-tent business, offers a real business example. The question was which products to buy and how much cash to commit to stock. Business Play brings sales opportunities, inventory and financial readiness into one decision, helping the owner compare options before approving a purchase.
The case-study page also offers a separate learning demo using simulated data. It does not expose customer data or connect to Snowflake live.
See the change before it appears in the financial statements
I have been reading Abundance Economics by John Leslie Millar. Chapter 9 raises a point that fits this decision closely. The author argues that changes in costs and competition often happen before they become visible in financial statements. Section 9.1 encourages businesses to watch signals from competitors and supply chains. Section 9.2.1, The Impact Audit, asks where our assumptions might fail if costs, selling prices, or customer demand change (Reference 7).
Applied to the inventory decision, the questions become clearer. If market selling prices fall faster than expected, is this opportunity still worthwhile? And how much cash pressure can the business carry while waiting to sell the stock? Those answers feed back into both Opportunity and Financial Readiness scores. The client then has a reasoned basis for buying the full lot, buying in stages, or waiting, along with a point at which to review the decision before committing more money.
How each area weighs the trade-offs
Data and AI starts with the numbers this decision needs: how much demand varies, the margin remaining after variable costs, current stock levels, and the lowest projected cash balance. AI can calculate multiple scenarios quickly. Before trusting the figures, however, someone must confirm that each dataset represents what the business actually means.
Strategy and business interprets when each alternative makes sense. A discount may improve the unit price while also putting more pressure on cash. If sales remain uncertain and cash is limited, buying in stages may be worth more than the discount. If demand evidence strengthens or payment terms change, the answer can change too.
Systems integration works with purchasing to establish whether the chosen approach can actually be implemented. Can the supplier deliver in stages? Who approves the purchase? How will the system alert the team if conditions change?
The client knows, before committing money, what needs to hold true, what could go wrong, and when to revisit the decision.
Four questions I use to guide AI's work
To make the division of work between people and AI clear each time, I adapt the AI Fluency Framework by Rick Dakan and Joseph Feller, whose learning materials were developed with Anthropic. It describes four areas of human capability. I turn them into four questions for discussions with clients (Read the framework).
Area | Question |
|---|---|
Delegation | Which work belongs to people, and which can AI do? |
Description | Are the goals, context, and constraints clear? |
Discernment | What evidence makes this recommendation credible? |
Diligence | Who is responsible for checking, using, and following up on it? |
Why think this through so carefully? Research published in 2024 by Michelle Vaccaro, Abdullah Almaatouq, and Thomas Malone offers a useful answer. They combined findings from multiple experiments and found that, on average, people working with AI performed better than people working alone, but did not outperform whichever was better when working separately: the human or the AI. Results also varied by task type, and most experiments predated today's AI models (Nature Human Behaviour).
For me, this means bringing people and AI together is not enough. We also need to test whether that division of work actually improves quality for the particular task.
Clients need to see who is responsible for what
In Chapter 7, section 7.3 of Abundance Economics, Millar argues that some work continues to require people because the person is part of the value the recipient wants. Sensitive decisions are one example: someone must take responsibility and explain the trade-offs. This argument has limits. It does not mean every form of consulting will survive without changing (Reference 7).
I apply this directly. Clients should know who checks the evidence, who explains the reasoning and limitations, and who has authority to approve. The expert's value lies in work the client can examine, from weighing the trade-offs to reviewing a decision when circumstances change.
From advice to work people can actually use
Power Ladder provides business consulting and systems integration (Consulting & Systems Integration). However good a recommendation sounds, it is of little use if it cannot be implemented. A recommendation to buy in stages, for example, works only if stock and cash data are accurate, purchasing has a clear owner, approval points are defined, and there is a way to detect changing conditions.
Stage | What the client receives | Lead responsibility |
|---|---|---|
Frame the decision | Agreement on the decision, its owner, and how success will be measured | Business expert |
Compare alternatives | A recommendation with evidence, assumptions, trade-offs, and review points | Business + data and AI |
Connect the workflow | A process the team can use, with data definitions and approval points | Systems integration expert |
Track outcomes | Actual results compared with the baseline, and a measurable next cycle | All three areas + the client |
I agree on measures with the client at the start: how long it takes to prepare a recommendation, how complete the evidence is, and how many manual handoffs remain. We then track financial and operational outcomes against the baseline to see whether the implementation actually helps decision-making.
AI needs to understand what the data means first
For AI to help with decisions, it first needs to understand what each piece of data means in the business. One example I like is Walmart's Wally, announced in March 2025. Walmart explains that it built a semantic layer to help its AI assistant understand internal sales data and help merchants examine product performance (Walmart's announcement).
When I work on Business Play in the client's Snowflake environment, the first task is therefore to agree on data definitions, then connect the sources and test the results against the decision. Cortex Agents can help by querying structured data and searching documents within configured permissions, with tools for evaluating responses. Clients receive a recommendation and can understand the data and conditions behind it (Cortex Agents · Agent evaluations).
Closer to the Business
For me, Closer to the Business means keeping the client's decision at the centre throughout the work: helping choose what matters, making the trade-offs visible, and putting an approved decision into practice, with AI as a tool that helps us move faster.
Which decision in your business would benefit from someone helping you develop alternatives, check the evidence, and put the result into practice?
Interactive Quiz: what would you decide?
Try five questions about an inventory purchase, then consider which evidence, cash constraints and conditions you would check before approval. This exercise uses simulated data to practise decision-making; it is not a real purchasing recommendation. Select EN for English or TH for Thai.
Or tell us what you are deciding, what constraints you face, and what data you already have. If your business does not yet use Snowflake, start with Discovery Consulting to clarify the decision, evidence, data readiness and implementation options. Business Play delivery described in this article takes place in the client’s Snowflake environment, with scope and readiness agreed before implementation.
References and credits
The application to Business Play, the allocation of responsibilities, and the hypothetical example are Power Ladder's approach.
May Phang · Snowflake workshop Wording from slides at the workshop attended and referenced by the author on 16 October 2026. The link leads to the webinar series page.
Rick Dakan and Joseph Feller · AI Fluency Framework Learning materials developed with Anthropic. The questions in this article are Power Ladder's application.
Vaccaro, Almaatouq, and Malone · Nature Human Behaviour (2024) When combinations of humans and AI are useful: A systematic review and meta-analysis. DOI: 10.1038/s41562-024-02024-1.
Walmart · Wally announcement, 18 March 2025 This example is unrelated to any Power Ladder project or relationship.
Snowflake · Cortex Agents Requires configuration, data preparation, and testing for the intended task.
John Leslie Millar · Abundance Economics: The Economic Mechanisms of the AI Age, Abundance Economics Press. Chapter 7, section 7.3, "Where Human Employment Persists"; and Chapter 9, sections 9.1, "The Recognition Gap for Existing Companies," and 9.2.1, "The Impact Audit." Based on screenshots of book pages supplied by the article's author on 27 October 2026. The book's arguments and their application to Business Play are separate.
Comments