Where to Start With AI: What Does Your Company Need Most?

Short version: Start with where AI can create the most immediate value for your company. That might mean helping your team use AI more consistently, building AI into a specific workflow, or doing both together. Here is how the options compare, and five questions to help decide whether a build is worth pursuing.

Two Kinds of AI Spending That Look Identical on a Proposal

There are two different things a company can buy when it decides to do something about AI, and they arrive in the same slide deck.

The first is AI implementation: building AI into your software. Computer vision, knowledge assistants grounded in your own documents, intelligent document processing, embedded features inside products you already run, predictive models, agentic workflows. This is real work and it produces something you can point at.

The second is AI enablement: making your team more capable and consistent with AI. Standardizing the tools your developers use, bringing AI into requirements, testing and code review, prompt and context engineering, governance and policy, and the change management that makes any of it stick.

Both are legitimate, and we do both. The question is where your organization will get the most value from its next AI investment.

Which Brings the Most Immediate Value

Look at where AI could save time, reduce costs, or improve the quality of your work. If your team already uses AI but struggles to get consistent results, enablement can improve the work happening today. If a defined workflow is costly or slow, and the team is ready to support it, implementation may offer more immediate value.

The market is not short of AI access. In McKinsey's State of AI survey, fielded in May and June 2026 with 1,719 participants across 97 nations, nearly nine in ten organizations report regular use of AI in at least one business function. Only 37 percent attribute at least some EBIT impact to that use, which is about the same share as the year before. Adoption widened. Enterprise financial impact did not keep pace.

The survey does not say why, and a survey result is not an explanation. Our own read is that the gap is not about access. Nearly everyone has access. It is about consistency. Six developers running six different setups can produce six different quality levels, and nobody can tell which practices caused which outcome.

In that situation, shared practices may be more useful than another AI feature. A team with those practices in place may get more value from a focused build. Enablement and implementation can also happen together; the priority is the work that addresses your most pressing need.

Training is one part of enablement, not the whole of it. Enablement is a set of practical tools, shared expectations, and repeatable ways to use AI that stay behind after the engagement ends. In practice that means an approved toolset, shared context standards, an agreement about what gets AI assistance and which outputs require human review before they are used, and a written answer to what data is allowed to go where. For teams using coding agents, it also means preparing repositories with clear project instructions and reliable tests, and setting limits on what agents can access or change. None of it requires a custom model or a new AI product, and all of it changes what happens the next time somebody proposes an AI project.

Five Questions to Ask Before You Fund a Build

When a build looks like the best use of your budget, these five questions help decide whether it is worth scoping.

1. What business problem will it solve? Describe the cost in time, money, or risk, and who bears it.
2. Is the data available and permitted for this use? Confirm that it exists, is accessible, and can be approved for the project.
3. How will the output fit the workflow? Identify who reviews it, what happens next, and what it replaces.
4. Who is accountable when it is wrong? Name an owner and a path for reviewing uncertain cases.
5. How will you measure results after launch? Set success and failure thresholds, and keep checking quality as prompts and models change.

If all five have credible answers, you have a candidate worth scoping, and the next step is to validate security, privacy, technical feasibility, integration effort, and expected value before committing to a build. If two or three do not, you do not have an AI project yet. You have a data problem, a workflow problem, or an ownership problem wearing an AI project's clothes, and fixing that is the honest next step.

What Usually Goes Wrong Is Not the Model

RAND interviewed 50 practitioners across more than 50 organizations and identified five recurring root causes of AI project failure: misunderstanding or miscommunicating what problem needs solving, lacking the data to train an effective model, focusing on the latest technology rather than on real problems, lacking adequate infrastructure to manage data and deploy models, and applying AI to problems too difficult for it. Our read is that most of those begin with organizational choices rather than technical ones. Many AI projects start to fail before model development, or outside it entirely, in the weeks before anyone opens an editor.

How We Approach It

Sometimes the honest answer is a simpler automation. If the inputs are predictable and the business needs the same answer every time, a deterministic automation or a plain integration will often create more value with less complexity, cost, and risk. When that is the case, we will tell you. We will cover that trade-off in a follow-up.

We start with where AI can actually help, not with the tool. We work with you to decide whether that calls for team enablement, a focused build, or both. If you are weighing an AI initiative and want a straight answer about where to start, that is a useful first conversation to have.

Frequently Asked Questions

What is AI enablement, and how is it different from AI implementation?

AI enablement makes an existing team more capable and consistent with AI through approved tooling, upskilling, AI-augmented delivery practices, and governance. AI implementation applies AI to a defined product or workflow. Enablement creates the conditions for repeatable, responsible use; implementation turns a validated use case into a working solution.

Should we do AI enablement before AI implementation?

Not necessarily. Enablement is a useful starting point when your team needs more consistent practices for using AI. If you have a validated use case with reachable and permitted data, an understood workflow, clear ownership of errors, and a way to measure results, a build may be the better investment. Scope it and put the necessary guardrails around it. When both are needed, team enablement can happen alongside implementation.

Not sure whether your next step is enablement, a simpler automation, or an AI build? Talk with us about your starting point.