The Best AI-First Organizations Are Built Bottom-Up

Too many AI-deployment guides commit the same “AI-first” mistake: promoting trendy solutions without diagnosing the real problem. I recently came across Microsoft's “Best practices for effective AI implementation” resource (found here) and was concerned with how they go right to chatbots and KPIs without identifying the friction points best positioned to be solved with AI. Rather, effectively introducing AI begins with bottom-up problem analysis that engages every stakeholder to solve core pain points.

What does this look like?

Let’s take a look at Microsoft's deployment pipeline (under “Project planning for AI deployment”). After assessing AI readiness they identify that an organization should set clear objectives. This means first selecting some key performance metric most aligned with your business goals and then pushing for a measurable improvement through AI integrations. Take the following example Microsoft provides as a “specific [area] where AI can deliver the most impact”:

Decrease average time response for customer inquiries by 40% using AI chatbots. Monitor the average time it takes to resolve customer issues before and after chatbot implementation.

In this case, Microsoft guides deployments as a top-down process where the best solution is one without any recognition of the problems or pain points in the customer service process that cause poor average response times. Their suggested pipeline provides an arbitrary, flashy AI solution—a marketable plan of action that executives can get behind—without any problem identification. Suppose that the true friction point slowing down response times is a legacy CRM or entitlement lookup system call (e.g. to an LDAP server) that happens before the response. If that CRM is slow then the time to acknowledge the user won’t budge even with an instant greeting afterwards.

Consider as well:

Increase conversion rates for personalized marketing campaigns by 25%. Track the conversion rates of campaigns using AI-generated recommendations versus traditional methods.

Again, their identified KPI, the marketing campaign conversion rates, will supposedly improve by implementing an AI-generated recommendation system with little research into the causes of the low conversion rate. If the poor rate was caused by a slow or glitchy mobile checkout experience, no matter the improvement to the recommendation system, customers would still drop at payment. Thus upgrading the user-experience or payment options would advance your KPI far more than any AI-generated recommendation.

Why is this an issue?

Without proper analysis that identifies the bottlenecks across your organization that are most suited for AI assistance or automation—those that are data-intensive, repetitive, and require little decision-making—you’re ignoring solutions to problems that may have a magnitude greater rate of return than a sweeping AI-first initiative. These AI low-hanging fruit are often data-driven, high-volume tasks that require minimal creativity or collaboration, though they cause recurring friction for workers in the long run. By targeting these areas first, you capture a significantly higher rate of return. For example, consider implementing a queryable, AI-powered knowledge base that quickly surfaces answers, speeding up customer inquiry resolution, rather than building an expensive AI chatbot that doesn’t address the core issues in your current process. Building the AI chatbot requires investment in an entirely new business process and displaces your current workers—instead of boosting their productivity. Identifying the problem before prescribing the solution ensures that you invest where the highest ROI lies, instead of chasing flashy initiatives that are likely to underperform.

But this isn’t just an issue of uncaptured potential. Instead, it reflects a divisive approach that is likely to slow down your transition to becoming an AI-first organization by alienating your employees. Why? When leaders expend resources on shiny new equipment instead of making repairs or value-added improvement to the existing machines, it demonstrates that they don’t value the old machines and intend on replacing them shortly. In other words, in order to get buy-in from your workers, you must listen to their needs—asking what the friction points are, where they are most bogged down, and if—not how—AI can help alleviate certain pressures in their workflows. Rather than presenting AI as the ultimate solution, involve all the stakeholders of a business process or unit to create solutions. This bottom-up approach will not only uncover the highest-impact opportunities for AI but also create trust between management and employees, making for smoother adoption and real productivity gains.

In short, stop jumping straight into high-level KPI goals or flashy solutions. “AI-first” doesn’t mean AI solutions should be considered before the problems at hand, but that AI solutions are the first in line to be tested. By leading with the real problems your workforce faces, you maximize return on investment with employee support, unlocking the true potential of an AI-first organization.