Director, Learning Design & Development| PMIAsheville, NC, United States
Validating and checking outputs is critical when working with AI systems like Generative AI. Such validation approaches may include establishing clear criteria, implementing strong testing protocols, and continuous refinement.
In your experience with AI, what are some best practices for ensuring the results you receive are accurate, relevant, and aligned with your original goals?
For ensuring that the outputs are accurate and relevant to what was intended by the original goals, it is crucial to have proper intentions right from the start of the task. This can be achieved by understanding the objective of the task, the target audience, and the desired output format.
Always check critical information from reliable sources, particularly for data-driven or high-stakes decisions. AI outcomes must be viewed only as drafts or suggestions, not as decisive. Iteration through prompts, asking probes, and seeking explanations helps improve the quality and relevance. The AI is presently in a evolution phase as the data for specific domain, region demography, culture is not available.
Finally,I believe use of human judgement to check outcomes for bias, practicality, as well as relevance to objectives must be done as of now. While artificial intelligence works best as a tool to support decision making, the implementation of the same must be correlated with the available data for accuracy. Saving Changes...
Sanjeev JhaQA Manager| NLB Services for client NTTData
Few best practices which I often follow:
(1) Validate the response by going through the output myself.
(2) Ask the same LLM to cross check its response.
(3) Ask the same LLM or AI model to explain the process by which it draws the response. (Prompt chaining)
(4) Ask further questions on the given response to know if given result is correct.
(5) Sometimes, I validate the response by giving the same prompt to some other similar LLM or AI model and compare the result.
Using the creative approach as opposed to the RTF approach is a good start, but then be prepared for iterative tweaking based on the responses received.
That's a very good question. In my response, I am assuming that the question refers to an LLM-based chatbot.
From my experience, the best results are achieved the more context I provide to the LLM. This means providing as much information as possible that describes both the project itself and the project context.
A second very important step is the quality of the request, also known as the prompt for the LLM. This is similar to human communication, where the quality of the question determines the quality of the answer. Therefore, a good prompt strategy is required, for example:
1. Data and context about the project
2. The goal of the request
3. The task that the LLM should fulfill
4. The format in which the output should be delivered.
In subsequent requests, it is possible to build on the context and results of the previous request. It is important that this process takes place within a chat, as otherwise the context is lost.
Use a proven framework like CREATE and continue to iterate till you get the results you are looking for.
A very good question and also difficult to answer as well. However you have to go to the basics and say as far as you are concerned, how well are you versed with the subject at hand ?. There are facts which the AI will generate and if you can verify these facts the more reliable the generated response will be. The fewer the facts then it means that the Generative AI response is far from meeting your original goals. Then it becomes very critical that you review the accuracy , relevancy and the alignment of the response to your original need. Unfortunately there are no clearly defined metrics that one can use a model to evaluate an AI generated response. So from my personal experience I basically restrict AI to an area where i have sound knowledge of , else it becomes almost impossible to verify details generated by an AI if you venture into unchartered territory. However with long usage and exposure your confidence also tend to increase as well. The best practice and protocol to follow would be to consult subject matter expects to validate the AI generated response before making critical decisions based on it to avoid any inherent associated risks which you might be not aware of.
AI is not an oracle of the ancient times that intermediates responses from the gods. It is much more similar to a dedicated interim that will deliver literally deliver what it was asked to.
So Mr. Chitsamatanga's advice of submitting AI suggestions to SMEs is wise.
Recently I heard a professor saying AI should be preferably used in two scenarios:
1. You know a lot about a subject and need to refine or evolve on it
or
2. You know nothing about a subject and need some directions.
To get accurate, relevant, and goal aligned results from AI, you need to be precise about your objective and give the system clear constraints, roles, and priorities. Require the AI to show its reasoning, surface uncertainties, and challenge its own conclusions to reduce hallucinations. Break complex tasks into structured steps and iterate rather than accepting the first output. And above all, maintain your own critical judgment instead of outsourcing decisions to the model. Saving Changes...
Hard to answer! Theoretically looks like an easy response, however the tricky part is bringing to words the abundant tacit behaviour and knolewdege in every process. Very much liked the proposal of act like teaching a Mentee.
From a project management standpoint, ensuring AI outputs are accurate and aligned starts before the prompt is written. Some best practices I’ve found effective are:
Anchor AI use to a clear objective: Define the decision, outcome, or deliverable the AI is supporting. If the goal isn’t clear, even a “good” output can be misleading.
Apply human validation, not blind acceptance: Treat AI as an assistant, not an authority. Outputs should always be reviewed against project context, constraints, and stakeholder expectations.
Use triangulation: Cross-check AI outputs with known data sources, SME input, or historical project artifacts (lessons learned, risk registers, KPIs).
Iterate prompts intentionally: Refinement is less about wording tricks and more about progressively narrowing scope, assumptions, and success criteria.
Embed governance and ethics: Be mindful of data sensitivity, bias, and traceability—especially when AI informs decisions that impact people, cost, or compliance.
In practice, AI delivers the most value when it enhances judgment, sense-making, and decision quality, not when it replaces them. This reinforces the PM’s role as a critical thinker and steward of outcomes, rather than just a consumer of outputs. Interested to hear how others are operationalizing validation in real project environments. Saving Changes...
PMI Team Member
