A useful approach is to ask questions such as: What problem are we trying to solve? What process or outcome needs improvement? What are the current challenges? How is the process handled today? What measurable benefit is expected? Saving Changes...
You will need to understand the requirements of the project before zeroing down to which AI to use, as many people talk about AI, but they don't have enough information to back it up Saving Changes...
When someone says “we should use AI,” it’s important to unpack the intent. First, clarify the problem they want solved and the objectives behind using AI. Then, assess how AI will integrate into workflows and whether the organization has the right data and infrastructure. Ethical and compliance considerations must also be addressed to avoid risks. Finally, ensure AI adoption is aligned with business goals so it delivers measurable value rather than becoming a buzzword. This way, the vague idea of “using AI” becomes a clear, actionable strategy. Saving Changes...
Nguyen Van SonProject Management Board| Van Xuan GroupHo Chi Minh, Viet Nam
I agree with you, many times people are just pressured to use AI, but it is necessary to get the requirements clear first. Saving Changes...
Mohamed Abbas AliPM Consultant| Future Consult CompanyRiyadh, Saudi Arabia
Mar 19, 2026 11:15 AM
Replying to Omar Jabbar
...
I’ve been asked this many times, and my first response is always: what do you want to achieve with AI? Once the outcome is clear, we can define the right approach, tools, and path forward.
Signals to Differentiate AI Work Different types of AI work can be identified by their underlying architecture, data requirements, and primary objectives.
Generative AI: Recognizable by natural language or visual outputs. It uses large language models (LLMs) to create new content based on prompts.
Predictive AI: Recognizable by numerical risk scores, trends, or classification outputs. It uses statistical machine learning to analyze historical data.
Automation AI: Recognizable by systemic, repetitive task execution. It uses robotic process automation (RPA) or deterministic algorithms to streamline workflows.
Risks of Lumping AI Work Together When universities treat all AI as a single category, several operational and ethical issues occur:
Mismatched Guardrails: Applying strict plagiarism rules meant for Generative AI to predictive grading analytics tools stalls legitimate institutional research.
Resource Waste: Investing in expensive generative compute infrastructure when a project only requires a simple, lightweight regression model.
Compliance Failures: Blurring data privacy lines by feeding sensitive student records into public generative tools instead of secure internal databases.
Signs of Misaligned AI Conversations In academic settings, the term "AI" is frequently used as a catch-all, leading to conceptual friction between different stakeholders.
What Tipped Me Off
Clashing Expectations: Faculty members discuss AI as a threat to academic integrity (Generative AI text generation), while IT administrators discuss AI as a tool to optimize enrollment forecasting (Predictive AI).
The "Magic Wand" Fallacy: Users assume an AI tool can automatically clean unstructured data, build a pipeline, and make decisions without human oversight or custom training.
Incompatible Metrics: Evaluating a generative brainstorming tool using predictive accuracy metrics, or judging an automation script by its creativity.
Navigating the Request: "We Should Use AI" When university leadership or department heads suggest using AI, the actual underlying need usually falls into one of three academic categories. 1. "We have too much administrative paperwork."
Real Request: Basic workflow automation.
Academic Example: Processing transfer credit applications. Instead of an LLM, the university needs an automated parser to match course codes against an existing articulation database.
2. "We need to improve student retention."
Real Request: Predictive analytics.
Academic Example: Identifying at-risk students. The institution needs a predictive model to flag low Learning Management System (LMS) activity engagement, allowing advisors to stage early interventions.
3. "We want to modernize our teaching methods."
Real Request: Generative assistance and interactive learning.
Academic Example: Creating personalized study aids. Professors need a sandboxed LLM trained on specific textbook chapters to act as a 24/7 virtual teaching assistant for student Q&A.
Saving Changes...
Naji El HakimDesign and Business Development Director| United Technology for Construction Co.
The biggest signal is specificity. When someone says "we should use AI" without naming a task, that's a red flag — they're selling a feeling, not a solution. In my work managing architecture projects across data centers, hotels, offices, and residential developments in Saudi Arabia, I've learned to ask immediately: "AI to do what, exactly?"
It's like telling someone "go use Microsoft Office to make me a balance sheet" — without knowing that Excel is the right tool, not Word, not PowerPoint. The suite is irrelevant if you can't match the tool to the task. AI is the same. Saying "use AI" without specifying the pattern is meaningless — and sometimes dangerous.
That one question — "AI to do what, exactly?" — separates the people who've thought it through from those chasing a trend. When everything gets lumped together, teams end up buying tools that don't match the problem, and then blame AI when it underdelivers.
PMI Team Member
