This strategic framework was created by George Westerman, a Research Scientist in the MIT Initiative on the Digital Economy. It is used to determine if automating a specific task with generative AI is economically and practically viable compared to an existing manual process.

The equation evaluates viability by comparing the total costs of the AI implementation against the Cost of the Manual Process (the status quo). For an initiative to be sensible, the following condition should generally be met:

(Cost of Using AI + Cost of Adaptation + Cost of Detecting and Fixing Errors) < Cost of Manual Process

The three main cost components are defined as:

1. Cost of Using AI: The direct expenses involved in accessing the technology, such as API usage fees, subscriptions to providers (e.g., OpenAI, Azure), or the computing power required to run open-source models.
2. Cost of Adaptation: The investment of time, resources, and specialized expertise required to fine-tune or adapt the model to achieve the specific level of accuracy needed for the task.
3. Cost of Detecting and Fixing Errors: The cost of the necessary supervision process—often involving a "human-in-the-loop"—to review AI outputs and correct mistakes, acknowledging that generative AI is not perfect and can produce hallucinations

You should perform this practice to validate the economic and practical viability of a project before investing resources. Specifically, it helps you:

• Determine Economic Sense: It ensures that the total investment in AI (usage fees + fine-tuning + human supervision) is actually lower than the "Cost of the Manual Process" (the status quo). It prevents you from spending more on automation than you save on labor,.

• Avoid the "Hype" Trap: It provides a disciplined framework to look past the excitement of new technology. Instead of implementing AI just because it is trendy, this equation forces you to analyze if it truly solves a business problem efficiently.

• Uncover Hidden Costs: It highlights costs that are often overlooked, specifically the "Cost of Adaptation" (hiring experts to fine-tune models) and the "Cost of Detecting and Fixing Errors" (the human time required to review AI outputs), which can be significant,.

• Assess Risk ("Cost of an Error"): It compels you to evaluate the consequences of an AI hallucination. If the "Cost of an Error" is too high (e.g., a wrong medical diagnosis or a legal liability), the initiative may not be viable regardless of the implementation costs,.

Phase 1: Establish the Baseline

Step 1: Calculate the "Cost of the Manual Process" Before looking at the technology, you must quantify the status quo.

• Action: Determine exactly how much it costs to perform the task the way it is done today "by habit". This includes employee hours, salaries, and overhead associated with the manual workflow.

• Key Roles:

    ◦ Process Owner/Business Analyst: To map the current workflow and measure the time currently spent.

    ◦ Finance/HR: To provide accurate data on labor costs associated with those hours.

Phase 2: Calculate the "Cost of Executing" with AI

Step 2: Determine the "Cost of Using AI" Calculate the direct expenses of the technology itself.

• Action: Estimate the fees for accessing the technology. This could be API usage fees (e.g., tokens), monthly subscriptions to commercial providers (like OpenAI or Microsoft Azure), or the infrastructure costs to run open-source models on your own servers,.

• Key Roles:

    ◦ IT Architect/Tech Lead: To select the right model and estimate compute/API consumption.

    ◦ Procurement: To handle vendor subscriptions or cloud budget allocation.

Step 3: Estimate the "Cost of Adaptation" Generative AI is rarely perfect "out of the box" for specialized tasks.

• Action: Calculate the investment required to fine-tune the model or engineer prompts to achieve the necessary level of accuracy. This is often the "hidden" cost involving specialized expertise and time.

• Key Roles:

    ◦ Data Scientists/AI Engineers: To perform the fine-tuning or Retrieval-Augmented Generation (RAG) setup.

    ◦ Subject Matter Experts (SMEs): To provide the "ground truth" data and examples needed to train/adapt the system.

Step 4: Calculate the "Cost of Detecting and Fixing Errors" (Human-in-the-loop) Because GenAI hallucinates, you cannot simply automate and forget.

• Action: Estimate the cost of the human supervision required to review the AI's output and correct mistakes,. If reviewing the AI output takes as long as doing the work manually, the project fails this test.

    ◦ Example: In software, existing debugging processes might absorb this cost easily. In legal contracts, this might require expensive lawyer hours.

• Key Roles:

    ◦ SMEs/Reviewers: To act as the "curators" validating the output.

    ◦ Project Manager: To design the workflow that integrates this review process efficiently.

Phase 3: The Risk Assessment

Step 5: Assess the "Cost of an Error" Even with supervision, errors may slip through. This is the "Go/No-Go" gate.

• Action: Evaluate the consequences if the AI is wrong and the humans fail to detect it.

    ◦ If the cost is low (e.g., a slightly imperfect marketing email), the project can proceed.

    ◦ If the cost is high (e.g., a wrong medical diagnosis or a plane crash), the project may not be viable regardless of the financial savings,.

• Key Roles:

    ◦ Legal/Compliance Officer: To assess liability and regulatory risk.

    ◦ Senior Leadership: To decide if the risk appetite aligns with the organization's strategy.

Phase 4: The Final Decision

Step 6: Execute the Comparison

• Action: Sum the costs from Steps 2, 3, and 4. Compare them to Step 1.

    ◦ If Total AI Cost < Manual Cost AND the Cost of an Error is manageable, the project is a candidate for implementation.

• Key Roles:

    ◦ Decision Maker (Sponsor): To give the final approval based on the economic logic and risk profile.