Budgets are approved. Tools are licensed. Pilots are running.
And yet most leadership teams, when asked directly, cannot answer a simple question: where is our organization actually getting value from AI?
The honest answer, in most cases, is the same place: the easy stuff. One-off tasks. Individual productivity wins. A handful of people using ChatGPT or Claude to draft emails faster.
That's a start. It is not a strategy.
The broader data points in the same direction. McKinsey's latest State of AI survey found 88% of organizations now use AI in at least one business function, yet only about a third have scaled anything beyond pilots. Gartner found that only 20% of low-maturity organizations keep an AI project running for three years or more. Adoption is no longer the hard part. Durability is.
The organizations pulling ahead are not just using AI more. They're using it more deliberately. They know which investments compound and which ones consume budget without returning anything durable. They can tell the difference between a prompt that should stay a prompt and a process that's ready to become infrastructure.
This framework is built for that conversation.
Not another maturity model
Most AI frameworks are ladders. Gartner and McKinsey each publish a version: five stages, climb from experimentation to transformation, measure progress by which stage you've reached.
Ladders are useful for benchmarking. They're less useful for the decision actually in front of you: should this initiative be a prompt, a configured workflow, or a system build?
This grid is a map, not a ladder. There is no bottom rung to be embarrassed about and no top rung where everything belongs. A well-run organization operates in many cells at once, deliberately. The CEO reframing a board memo in a chat window and the engineering team running an AI platform in production are both doing it right, as long as each initiative sits in the cell that matches what it's trying to accomplish.
The goal is not to climb out of the "low" cells. The goal is to put every initiative in the right cell, and to know why it's there.
Two questions that cut through the noise
Before approving any AI initiative, two questions determine whether the investment makes sense.
What do you want AI to do? A one-time answer, recurring analysis, and repeatable action are three different jobs. Treating them the same is where most AI spend gets wasted.
How deeply does this need to be built? A well-crafted prompt, a configured workflow, and a custom-built platform are not interchangeable. Each has a different cost, a different owner, and a different maintenance requirement. Choosing the wrong level is expensive in both directions. Over-build a prompt into a platform and you've wasted both human and token capital. Under-invest in something that deserved real infrastructure and you've capped the return. In production, under-investment looks like a fragile, undocumented integration quietly creating duplicate billing records.
Map those two questions on a grid and nine distinct investment categories emerge. Most organizations operate in three or four of them. The ones seeing compounding returns have learned to use all nine, at the right time, for the right reason.
The AI Utilization Grid
Plotted together, those two questions become a grid. The columns are what you want AI to do. The rows are how deeply it's built. Hover or tap any cell to see how different teams put it to work.
The rows · Deployment model
How the capability is built and owned. This is about permanence and ownership, not sophistication.
- Prompt: a person runs it directly. Minutes to hours. Nothing persists when the conversation closes.
- Process: a configured workflow runs it. Days to stand up, a business owner maintains it. Assembled from existing tools, not custom-built.
- Platform: a purpose-built system the business runs on. Weeks to months to build, with sustained ownership and investment to keep it alive. Durable infrastructure, however it's built.
The columns · What you want AI to do
The job you need done, independent of how it's delivered.
- Answer: an ad hoc output: an answer, a draft, a summary.
- Analyze: recurring insight and decision support.
- Act: repeatable work, executed reliably.
The four cells that matter most right now
Quick Wins: necessary, not sufficient
Every team should be operating here. A CEO reframes a board memo in minutes before a 9am call. A CFO gets a plain-language read of a 40-page analyst report before an investor meeting. A sales leader drafts a competitive response to an RFP that landed this morning.
If your team isn't here yet, this is the gap to close before anything else. It builds the baseline familiarity that makes every subsequent investment easier to trust and justify. Plenty of one-off work lives here for good; the mistake isn't using Quick Wins, it's letting it become the only cell you operate in.
Consistency: the most underused cell
This is where trust in AI gets built, and where most organizations leave the most value on the table.
The difference from Quick Wins isn't the task, it's the repetition. A sales team saves its best follow-up prompt as a reusable skill every rep runs after a customer call. An ops director turns the weekly status update into one standard prompt applied across six teams. A marketing lead locks in a brand-voice skill so all outbound content starts from the same reliable baseline. Same quality, every time.
No workflow tools. No engineering. Just a prompt good enough to save and reuse on a repeating task.
Most organizations skip Consistency not because they don't value reliability, but because it doesn't feel like progress. There's no launch, no announcement, no demo. A prompt template doesn't generate a press release. But six months of consistent AI output across a sales team, an ops function, or a content team builds something harder to replicate than any tool purchase: organizational muscle memory.
The ROI compounds with every repetition. So does the trust. Organizations that skip this cell and jump to automation or platform investment often find their teams resistant to the output, because consistency was never established. Trust in AI is a function of repeated, reliable output over time. You can't mandate it, and you can't rush it. But you can start today with no budget approval required.
Trust in AI is a function of repeated, reliable output over time.
Experiment: where smart budget decisions get made
Before any significant Process or Platform investment, something should be validated here.
A CFO greenlights a two-week test connecting an AI tool to the data warehouse before approving a four-month analytics build. A CTO validates whether an AI agent can reliably navigate an internal system before any engineering time is allocated.
The cost of skipping Experiment isn't just wasted engineering budget. It's the credibility loss when a Platform investment fails publicly inside the organization. A failed pilot in Experiment costs days or weeks. A failed platform build costs months, political capital, and often sets back AI adoption across the team by a year.
Many expensive AI misses start by skipping this step: moving from Quick Wins straight to a Platform build without proving the pattern first. Experiment is cheap insurance on every investment above it.
Engine: where competitive advantage compounds
This is the conversation happening at board level in the organizations moving fastest, and it's where the numbers stop being hypothetical.
From our published customer work: an internal knowledge agent that answers the questions employees used to file tickets for, recapturing roughly 20,000 hours a year. That capacity doesn't just come off a cost line, it widens the margin on everything the business does. A competitor still paying people to handle those questions carries a cost you've already removed.
That's the quieter half of the Engine: margin you expand by lowering the cost to operate. The louder half is revenue, AI built into the product or the go-to-market motion itself, so the system doesn't just cost less to run, it sells. Either way, the economics of the business change.
Each Engine makes the next one cheaper and faster to build, because the organization now has the infrastructure, the data patterns, and the institutional knowledge to deploy AI faster than it could a year ago.
This is the top of a progression the Act column makes visible: a prompt that proves consistent becomes a workflow that scales, and the rare capability that throws off real, repeatable return gets built into the core. Not everything should become an Engine. The ones that earn it travel this path.
The earlier this foundation gets built, the harder it becomes for competitors to catch up.
What the middle of the grid looks like
The four cells above get the attention. Three quieter cells do a lot of the work in practice. All three examples below come from our published customer work.
- Pulse (a workflow that watches): an order-to-cash monitoring system that surfaces integration issues in minutes instead of weeks, cutting sync failures by 70%.
- Scale (a workflow that does): purchase-order processing that cut errors by 90%. Same-day orders went from exception to standard.
- Control Tower (a system view of the business): a customer data platform unifying CRM, support, and product usage data, answering questions 14x faster. Leadership stopped cleaning data and started using it.
The full set is at onesolve.io/customer-stories.
A note on the most expensive mistake in AI right now
Before any system build gets approved, one question: can this be handled by an ad hoc prompt?
In most cases, it can.
A custom pipeline built to answer questions about a single contract. An API integration coded from scratch to pull data for one board presentation. A bespoke tool scoped by IT for a request a well-crafted prompt would have resolved in an afternoon. These are not hypothetical. They happen in organizations with good engineers, reasonable budgets, and no framework for matching effort to use case.
The Overkill cell exists to name the pattern so it can be caught before the invoice arrives. The most expensive AI mistake isn't a bad vendor choice or a failed model. It's a six-week build for a problem that belonged in a prompt.
Adoption is no longer the hard part. Durability is.
Where does your organization fit?
Most organizations running multiple AI initiatives will find themselves in several cells at once. That's normal. The question is whether each initiative is in the right cell for what it's trying to accomplish.
Bring these to your next leadership meeting:
- Which cells have active initiatives right now? Map them and see where the weight falls.
- Are any Platform-level builds solving problems that belong in Prompt? If so, that's a budget conversation worth having before the build is complete.
- Is your team operating in Consistency, or skipping it? If the answer is skipping it, the fastest ROI available requires no budget approval.
- What's been validated through Experiment before moving to Scale? If the answer is nothing, the next Platform investment carries more risk than it needs to.
- Which teams are closest to an Engine-level investment, and what do they need to get there?
Treat the grid as a working tool. Run it in a leadership or strategy session. Map your active initiatives. Ask which cell each one belongs in today, not where it started. Most teams find at least one initiative in the wrong row, and at least one opportunity in Consistency or Experiment that nobody owns yet. That conversation is usually more useful than another vendor demo.
What comes next
This framework is a starting point. Each cell has more depth beneath it.
The build vs. buy question deserves its own treatment, specifically why AI-assisted coding changes the capability barrier without changing the total cost of ownership. The Consistency cell deserves a longer look at what trust in AI actually requires from leadership, and why it can't be compressed on a timeline. The Engine deserves a serious examination of how AI-forward organizations are widening their competitive position, and what it takes to build one that holds.
Those are coming.
If you're trying to figure out where your organization stands, or want to think through the right next move, we're here. No pitch required, just a practical conversation.
OneSolve works with organizations at every part of this grid, from establishing Consistency in day-to-day operations to building the agentic infrastructure that powers competitive advantage.