Every few months, a variation of the same conversation plays out in leadership meetings across the country. Someone (usually a board member, sometimes a competitor’s press release) raises the question: “What’s our AI strategy?”

The instinct is to start looking for AI solutions. To evaluate vendors. To launch a pilot. To do something with AI so there’s an answer to the question next time it comes up.

This instinct is understandable. It’s also backwards.

The Solution in Search of a Problem

“We need AI” is not a strategy. It’s a technology in search of a use case. And when you start with the technology rather than the problem, you end up with implementations that are technically impressive and operationally useless.

This isn’t hypothetical. Research from MIT Sloan and Boston Consulting Group found that only 10% of companies report significant financial benefits from their AI investments.1 Gartner estimates that through 2025, at least 30% of AI projects will be abandoned after the proof-of-concept stage.2 The graveyard of AI initiatives is filled with projects that answered the question “Can we use AI for this?” without ever seriously asking “Should we?”

The pattern is predictable: a team identifies a process that seems like a good fit for AI, builds or buys a solution, launches a pilot, and then watches adoption stall. The problem isn’t the technology. The problem is that nobody verified whether the process was actually a priority worth solving, or whether the people doing the work would trust a new system to do it.

Start With the Business Problem

The organizations that extract real value from AI don’t start with AI. They start with a business problem that’s costing them money, slowing them down, or limiting their ability to scale.

What’s taking too long? What’s too expensive? What breaks when volume spikes? What depends on people who won’t be there forever? What decisions are being made with incomplete information because the data takes too long to compile?

These questions lead to a prioritized list of operational pain points. Some of those pain points will have AI-enabled solutions. Many won’t. The point is that you’re evaluating AI as a potential tool for solving a real problem, not adopting AI for its own sake.

When you start with the problem, you also define what success looks like before you build anything. If the problem is that manual invoice processing takes 40 hours per week, success is measurable: reduce that time by 70%. If the problem is that customer inquiries take 24 hours to get a first response, success is a metric: first response under 2 hours. These targets create accountability. They also create a kill switch: if the AI solution isn’t hitting the target, you know to stop investing.

The Data Question Nobody Wants to Ask

There’s another reason to start with the problem rather than the technology: it forces you to confront your data environment honestly.

AI systems are only as good as the data they’re trained on and the data they have access to. If your customer data is scattered across three systems that don’t sync, an AI-powered customer service tool will give inconsistent answers. If your inventory data is manually updated once a day, an AI demand forecasting model will be working with stale information. If your institutional knowledge lives in email threads and the heads of senior employees, there’s nothing for an AI system to learn from.

A 2023 survey by NewVantage Partners found that 82% of organizations cite data quality and accessibility as primary barriers to AI adoption.3 The data problem often isn’t a technology problem. It’s a process problem and a governance problem that predates any AI initiative.

Starting with “We need AI” lets you skip over these uncomfortable realities. Starting with “We need to reduce invoice processing time by 70%” forces you to look at where the invoices come from, what format they’re in, how they’re routed, and why exceptions happen. That investigation might reveal that AI is the right solution. It might also reveal that you need better integrations, cleaner data, or documented processes before AI is even viable.

Source: Composite from NewVantage Partners, Gartner, McKinsey surveys

Governance Before Implementation

AI governance is often treated as an afterthought: something to figure out once the AI is working. This is a mistake.

Without governance, AI adoption becomes chaotic. Different departments buy different tools. Sensitive data ends up in systems that weren’t evaluated for security. Employees use consumer AI products to process proprietary information. Decisions get made by algorithms that nobody fully understands and nobody is accountable for.

The organizations that succeed with AI establish governance frameworks early. They define what types of problems are appropriate for AI solutions. They create processes for evaluating vendors and vetting data security. They establish accountability: someone who owns outcomes, not just implementation. They build in review mechanisms so that AI-assisted decisions can be audited.

This governance work isn’t exciting. It doesn’t demo well. But it’s the difference between an AI investment that compounds in value over time and an AI investment that creates liability.

The Right Question

“What’s our AI strategy?” is the wrong question, not because AI doesn’t matter, but because it puts the technology at the center of the conversation instead of the business.

The right question is: “What problems are costing us the most, and what’s the best way to solve them?”

Sometimes the answer will involve AI. Sometimes it will involve better integrations, cleaner data, or more documented processes. Sometimes it will involve hiring. The goal isn’t to use AI. The goal is to solve problems. AI is one tool among many.

Organizations that approach AI this way don’t have flashy pilot programs to announce. What they have is a disciplined process for identifying high-value problems, evaluating potential solutions, and implementing changes that actually stick. That’s less exciting than “We’re using AI.” It’s also more likely to work.

Citations

1 S. Ransbotham et al., “Winning With AI,” MIT Sloan Management Review and Boston Consulting Group, 2019.

2 Gartner, “Predicts 2021: AI and Advanced Analytics,” 2020.

3 NewVantage Partners, “Data and AI Leadership Executive Survey,” 2023.