AI — how to start thinking about it

Business executives encounter frequent mentions of buzzwords like AI, machine learning, and analytics. The challenge lies in understanding what these mean for their organizations and vision. While the impact of growing system intelligence on business performance is evident, executives implementing AI face significant obstacles.

When seeking to adopt AI, leaders confront difficulties in understanding feasibility, technology choices, and execution approaches. This challenge stems from rapid technological advancement, futuristic messaging, and limited prior knowledge. Both buyers and sellers of AI services struggle with scope definition and expectation-setting, with substantial misalignment between business and technology executives driven by unclear boundaries and varying timeframes for understanding AI capabilities.

Two extreme viewpoints exist:

1. AI can solve every problem
2. AI cannot solve any problem

Most executives operate between these extremes. Successful AI implementation requires accurate understanding of organizational problems combined with knowledge of AI methods.

A simple framework:

Data + Algorithm = Intelligence

Key determining factors include:

1. What data gets generated and how?
2. How much task-specific knowledge is required to train algorithms? Is this widely available or highly specialized?
3. What is the optimization goal — human task replacement or human-machine output maximization?

Three categories of AI applications:

The first involves unconstrained data generation with widely available knowledge and limited organizational specificity. Examples include computer vision, speech generation, and voice translation. Leaders should focus on adopting technologies from large vendors.

The second category requires combining human cognitive abilities with process-specific knowledge. AI-driven intelligent automation addresses document processing, customer review analysis, and email routing. Executives should subdivide automation goals and replace individual task chunks.

The third utilizes production system data that is structured, unstructured, fragmented, and siloed. These applications face constraints from data management complexity and organizational context requirements. A discovery-led approach with use-case-driven capability building works best, emphasizing human-machine collaboration rather than substitution.

---

This framework still guides how Softsensor scopes AI case studies and projects — discovery-led, problem-first, and built for human-machine collaboration rather than substitution. Originally published on LinkedIn →

← All insights