that the majority AI pilot initiatives fail — not due to technical shortcomings, however as a consequence of challenges in aligning new know-how with present organizational constructions. Whereas implementing AI fashions could seem simple, the actual obstacles usually lie in integrating these options with the group’s individuals, processes, and merchandise. This idea, generally known as the “3P” pillars of mission administration, supplies a sensible lens for assessing AI readiness.
On this article, I introduce a framework to assist groups consider and prioritize AI initiatives by asking focused, context-specific {custom} questions throughout these three pillars, guaranteeing dangers are recognized and managed earlier than implementation begins.
Whether or not you’re concerned within the technical or enterprise facet of decision-making within the AI course of, the ideas I define on this article are designed to cowl each points.
The problem of implementing AI use circumstances
Think about being offered with a listing of over 100 potential AI use circumstances from throughout a world enterprise. Moreover, contemplate that the checklist of use circumstances breaks down into a wide range of particular departmental requests that the event crew must ship.
The advertising division desires a customer-facing chatbot. Finance desires to automate bill processing. HR is asking for a software to summarize hundreds of resumes. Every request comes with a unique sponsor, a unique degree of technical element, and a unique sense of urgency, usually pushed by strain to ship seen AI wins as quickly as attainable.
On this situation, think about that the supply crew decides to start implementing what seems to be the quickest wins, they usually greenlight the advertising chatbot. However, after preliminary momentum, the issues begin.
First are the individuals issues. For instance, the advertising chatbot stalls as two groups within the division can’t agree on who’s chargeable for it, freezing improvement.
After this subject is solved, course of points come up. For instance, the chatbot wants stay buyer information, however getting approval from the authorized and compliance groups takes months, and nobody is obtainable for added “admin” work.
Even when this will get resolved, the product itself hits a wall. For instance, the crew discovers the “fast win” chatbot can’t simply combine with the corporate’s important backend programs, leaving it unable to ship actual worth to clients till this subject is sorted.
Lastly, after greater than six months, budgets are exhausted, stakeholders are dissatisfied, and the preliminary pleasure round AI has worn off. Happily, this final result is exactly what the AI-3P framework is designed to stop.
Earlier than diving into the framework idea, let’s first have a look at what latest analysis reveals about why AI endeavors go off monitor.
Why do AI initiatives derail?
Enthusiasm round AI — or extra exactly, generative AI — continues to peak day after day, and so we learn quite a few tales about these mission initiatives. However not all finish with a optimistic final result. Reflecting this actuality, a latest MIT study from July 2025 prompted a headline in Fortune journal that “95% of generative AI pilots at companies are failing.”
A part of the report related to our function entails the explanations why these initiatives fail. To cite a Fortune post:
The most important drawback, the report discovered, was not that the AI fashions weren’t succesful sufficient (though execs tended to assume that was the issue.) As a substitute, the researchers found a “studying hole” — individuals and organizations merely didn’t perceive methods to use the AI instruments correctly or methods to design workflows that might seize the advantages of AI whereas minimizing draw back dangers.
…
The report additionally discovered that corporations which bought-in AI fashions and options have been extra profitable than enterprises that attempted to construct their very own programs. Buying AI instruments succeeded 67% of the time, whereas inner builds panned out solely one-third as usually.
…
The general thrust of the MIT report was that the issue was not the tech. It was how corporations have been utilizing the tech.
…
With these causes in thoughts, I need to emphasize the significance of higher understanding dangers earlier than implementing AI use circumstances.
In different phrases, if most AI endeavors don’t fail due to the fashions themselves, however due to points round possession, workflows, or change administration, then we now have pre-work to do in evaluating new initiatives. To realize that, we will adapt the traditional enterprise pillars for know-how adoption — individuals and processes, with a concentrate on the tip product.
This considering has led me to develop a sensible scorecard round these three pillars for AI pre-development choices: AI-3P with BYOQ (Carry Your Personal Questions).
The general thought of the framework is to prioritize AI use circumstances by offering your individual context-specific questions that intention to qualify your AI alternatives and make dangers seen earlier than the hands-on implementation begins.
Let’s begin by explaining the core of the framework.
Scoring BYOQ per 3P
As indicated earlier, the framework idea relies on reviewing every potential AI use case towards the three pillars that decide success: individuals, course of, and product.
For every pillar, we offer examples of BYOQ questions grouped by classes that can be utilized to evaluate a particular AI request for implementation.
Questions are formulated in order that the attainable answer-score combos are “No/Unknown” (= 0), “Partial” (= 1), and “Sure/Not relevant” (= 2).
After assigning scores to every query, we sum the overall rating for every pillar, and this quantity is used later within the weighted AI-3P readiness equation.
With this premise in thoughts, let’s break down how to consider every pillar.
Earlier than we begin to contemplate fashions and code, we must always be certain that the “human component” is prepared for an AI initiative.
This implies confirming there’s enterprise buy-in (sponsorship) and an accountable proprietor who can champion the mission by its inevitable hurdles. Success additionally is dependent upon an trustworthy evaluation of the supply crew’s abilities in areas similar to Machine Studying operations. However past these technical abilities, AI initiatives can simply fail and not using a considerate plan for end-user adoption, making change administration a non-negotiable a part of the equation.
That’s why the target of this pillar’s BYOQ is to show that possession, functionality, and adoption exist earlier than the construct part begins.
We are able to then group and rating questions within the Individuals pillar as follows:
As soon as we’re assured that we now have requested the correct questions and assigned the rating on a scale from 0 to 2 to every, with No/Unknown = 0, Partial = 1, and Sure/Not Relevant = 2, the following step is to test how this concept aligns with the group’s each day operations, which brings us to the second pillar.
The Processes pillar is about guaranteeing the AI use case answer matches into the operational material of our group.
Widespread mission stoppers, similar to rules and the inner qualification course of for brand spanking new applied sciences, are included right here. As well as, questions associated to Day 2 operations that assist product resiliency are additionally evaluated.
On this manner, the checklist of BYOQ on this pillar is conceptualized to grasp dangers in governance, compliance, and provisioning paths.
By finalizing the scores for this pillar and gaining a transparent understanding of the standing of operational guardrails, we will then focus on the product itself.
Right here is the place we problem our technical assumptions, guaranteeing they’re grounded within the realities of our Individuals and Processes pillars.
This begins with the elemental “problem-to-tech” match, the place we have to decide the kind of AI use case and whether or not to construct a {custom} answer or purchase an present one. As well as, right here we consider the soundness, maturity and scalability of the underlying platform, too. Aside from that, we additionally weigh the questions that concern the tip‑person expertise and the general financial match of the Product pillar.
In consequence, the questions for this pillar are designed to check the technical decisions, the end-user expertise, and the answer’s monetary viability.
Now that we’ve examined the what, the how, and the who, it’s time to carry all of it collectively and switch these ideas into an actionable resolution.
Bringing 3P collectively
After consolidating scores from 3P, the “prepared/partially prepared/not-ready” resolution is made, and the ultimate desk appears to be like like this for a particular AI request:
As we will see from Desk 4, the core logic of the framework lies in reworking qualitative solutions right into a quantitative AI readiness rating.
To recap, right here’s how the step-by-step method works:
Step 1: We calculate a uncooked rating, i.e., Precise rating per pillar, by answering a listing of {custom} questions (BYOQs). Every reply will get a price:
- No/Unknown = 0 factors. This can be a purple flag or a major unknown.
- Partial = 1 level. There’s some progress, however it’s not absolutely resolved.
- Sure/Not relevant = 2 factors. The requirement is met, or it isn’t related to this use case.
Step 2: We assign a particular weight to every pillar’s whole rating. Within the instance above, primarily based on the findings from the MIT study, the weighting is intentionally biased towards the Individuals pillar, and the assigned weights are: 40 p.c individuals, 35 p.c processes, and 25 p.c product.
After assigning weights, we calculate Weighted rating per pillar within the following manner:
Step 3: We sum the weighted scores to get the AI-3P Readiness rating, a quantity from 0 to 100. This rating locations every AI initiative into one in all three actionable tiers:
- 80–100: Construct now. That’s a inexperienced gentle. This means the important thing parts are in place, the dangers are understood, and implementation can proceed following customary mission guardrails.
- 60–79: Pilot with guardrails. Proceed with warning. In different phrases, the thought has benefit, however some gaps might derail the mission. The advice right here could be to repair the highest three to 5 dangers after which launch a time-boxed pilot to be taught extra about use case feasibility earlier than committing absolutely.
- 0–59: De-risk first. Cease and repair the recognized gaps, which point out excessive failure threat for the evaluated AI initiative.
In abstract, the choice is the product of the AI-3P Readiness method:
That’s the method for scoring a person AI request, with a concentrate on custom-built questions round individuals, processes, and merchandise.
However what if we now have a portfolio of AI requests? An easy adoption of the framework to prioritize them on the organizational degree proceeds as follows:
- Create a list of AI use circumstances. Begin by gathering all of the proposed AI initiatives from throughout the enterprise. Cluster them by division (advertising, finance, and so forth), person journey, or enterprise affect to identify overlaps and dependencies.
- Rating particular person AI requests with the crew on a set of pre-provided questions. Carry the product house owners, tech leads, information house owners, champions, and threat/compliance house owners (and different accountable people) into the identical room. Rating every AI request collectively as a crew utilizing the BYOQ.
- Type all evaluated use circumstances by AI‑3P rating. As soon as the cumulative rating per pillar and the weighted
AI-3P Readinessmeasure is calculated for each AI use case, rank all of the AI initiatives. This ends in an goal, risk-adjusted precedence checklist. Lastly, take the highest n use circumstances which have cleared the edge for full construct and conduct an extra risk-benefit check-up earlier than investing sources in them.
Now let’s have a look at some necessary particulars about methods to use this framework successfully.
Customizing the framework
On this part, I share some notes on what to contemplate when personalizing the AI-3P framework.
First, though the “Carry Your Personal Questions” logic is constructed for flexibility, it nonetheless requires standardization. It’s necessary to create a hard and fast checklist of questions earlier than beginning to use the framework so that each AI use case has a “honest shot” in analysis over completely different time durations.
Second, inside the framework, a “Not relevant” (NA) reply scores 2 factors (the identical as a “Sure” reply) per query, treating it as a non-issue for that use case. Whereas this simplifies the calculation, it’s necessary to trace the overall variety of NA solutions for a given mission. Though in principle a excessive variety of NAs can point out a mission with decrease complexity, in actuality this will sidestep many implementation hurdles. It could be prudent to report an NA‑ratio per pillar and cap NA contribution at maybe 25 p.c of a pillar’s most to stop “inexperienced” scores constructed on non‑applicables.
The identical is legitimate for “Unknown” solutions with rating 0, which current a full “blind spot,” and presumably needs to be flagged for the “de-risk first” tier if the data is lacking in particular classes as “Possession,” “Compliance,” or “Finances.”
Third, the pillar weights (within the instance above: 40 p.c individuals, 35 p.c processes, 25 p.c product) needs to be seen as an adjustable metric that may be business or group particular. As an illustration, in closely regulated industries like finance, the processes pillar would possibly carry extra weight as a consequence of stringent compliance. On this case, one would possibly contemplate adjusting the weighting to 35 p.c individuals / 45 p.c processes / 20 p.c product.
The identical flexibility applies to the choice tiers (80–100, 60–79, 0–59). A company with a high-risk tolerance would possibly decrease the “construct now” threshold to 75, whereas a extra conservative one would possibly elevate it to 85. Because of this, it’s related to agree on the scoring logic earlier than evaluating the AI use circumstances.
As soon as these parts are in place, you’ve all the things wanted to start assessing your AI use case(s).
Thanks for studying. I hope this text helps you navigate the strain for “fast AI wins” by offering a sensible software to determine the initiatives which might be prepared for achievement.
I’m eager to be taught out of your experiences with the framework, so be at liberty to attach and share your suggestions on my Medium or LinkedIn profiles.
The sources (tables with formulation) included on this article are within the GitHub repo situated right here:
CassandraOfTroy/ai-3p-framework-template: An Excel template to implement the AI-3P Framework for assessing and de-risking AI projects before deployment.
Acknowledgments
This text was initially printed on the Data Science at Microsoft Medium publication.
The BYOQ idea was impressed by my discussions with Microsoft colleagues Evgeny Minkevich and Sasa Juratovic. The AI‑3P scorecard thought is influenced by the MEDDIC methodology launched to me by Microsoft colleague Dmitriy Nekrasov.
Particular due to Casey Doyle and Ben Huberman for offering editorial critiques and serving to to refine the readability and construction of this text.
