Applied AI, On Your Terms

There has never been more applied AI help on offer. Sometime this quarter, in a business review you are already paying for, the pitches will arrive. The CRM provider brings agents and an activation team. The ERP provider brings a copilot. The data platform runs bootcamps. The cloud provider arrives with credits and embedded engineers. And this spring the model labs entered the field in force. Anthropic moved first, launching Ode with Anthropic. OpenAI followed a week later, standing up the OpenAI Deployment Company built to place forward-deployed engineers inside client organizations. Capability that once required years of internal hiring is now packaged, subsidized, and delivered to your door. But, at what cost?

It is worth being honest about enterprise AI adoption. Getting AI to move a P&L is a genuinely hard problem, and the numbers that predate this applied AI services wave tell you how hard. S&P Global found the share of companies abandoning most of their AI initiatives jumped from 17 percent to 42 percent in one year. Gartner expects over 40 percent of agentic AI projects to be canceled by the end of 2027. These are measures of the problem itself. Moving a P&L with AI demands a deep understanding of how a specific business creates value, and the discipline to put that business above everything else, including the platforms it runs on.

Put the business above everything else, including the platforms it runs on.

The default approach often for enterprises to adopt AI is to accept each vendor's offer on the vendor's terms and call the sum an AI Strategy. There is a better way, and it is neither exotic nor expensive. Be thoughtful, intentional, and systematic.

Step back before you dive in. Design the whole before you optimize the parts. And engage an applied AI partner whose interests are your business and your team first, and nothing else. An applied AI partner whose incentives are aligned so completely that your success is their success, irrespective of which data and technology platforms you double down on, which ones you shelve, and how and with whom you optimize your token spend.

In practice, this comes down to four moves, I recommend P&L leaders consider as they review AI adoption within their business:

1. Adapt the data and technology to the business.

Start with the reversal, because it is the biggest opportunity of this era. The ERP decades ran in one direction: enterprises adapted the business to the software, and best-practice templates were the price of standardization. The agentic era can run the other way. Satya Nadella argues that the business logic inside enterprise applications will migrate to an AI tier. If he is even half right, application boundaries are about to get soft, and processes no longer need to be designed around modules and licenses. They can be designed around how work should actually flow, with data and technology pulled toward the business instead of the business bending toward the stack. Enterprises that start from the business will build processes their competitors cannot copy, because those processes will not exist in any vendor's template. That advantage is durable, and it is available to any leadership team willing to do the design work.

2. Optimize the whole.

Designing from the business means seeing all of it. Every vendor's services scope ends where their product ends, and that is fair. The CRM team will make your CRM smarter. The data platform will make your pipelines faster. Take that help. Just remember that enterprise value does not live inside applications. It lives in the processes that run across them. Quote-to-cash. Claims. Supply planning. Service recovery. MIT's data shows AI budgets flowing overwhelmingly to sales and marketing while the better returns sit in operations and finance, which is what happens when the loudest use cases win instead of the most valuable ones. The end-to-end process map is the single most valuable AI artifact your company will produce this decade, and no vendor can produce it for you. Build it, keep it current, and score every applied AI offer against it. The parts will improve either way. The whole improves when you own it.

The biggest returns sit in the seams between your systems, and the seams belong to you.

3. Make the foundational decisions on purpose.

Before the first deployment, answer the questions that determine whether anything compounds. What is the target data architecture? Which processes should exist at all in an agentic era? Where does human judgment sit, and where does it move? What gets built, what gets bought, what gets retired? These are leadership questions, and they are answerable in weeks. BCG's research puts 70 percent of AI success in people and processes and only 10 percent in the algorithms themselves. Sequencing follows the same logic. Settle the seventy first, and the platform choices underneath become straightforward, reversible, and cheap to test. Skip this step and the stack answers the questions for you, which is how enterprises end up with an architecture nobody chose. The deliberate version takes a fraction of the time the accidental version wastes.

4. Align the incentives.

Vendor services are subsidized, and it helps to be clear-eyed about why. Cloud providers attach credits and engineers to consumption growth. Platform vendors price services to accelerate product expansion. None of this is a scandal. It is the business model, disclosed on every earnings call, and it makes vendor services excellent at one thing: executing on their own platform. Use them for exactly that.

For the design of the whole, hold a higher bar. Engage an applied AI partner with no platform to sell and no consumption to grow. Is that partner equally successful when you double down on a platform and when you shelve one? When you grow your token spend and when you renegotiate it, with whomever you choose? A partner who passes that test can tell you which process should disappear, which contract should shrink, and which pilot deserves ten times the investment. I asked recently who your AI agent works for. Ask the same question of the people who arrive to deploy AI within your business, and give the design seat to the ones with a clean answer.

Your success should be the Applied AI partner's success, irrespective of which data and technology platforms you double down on and which ones you shelve.

Parting thought: The same MIT research behind the 95 percent headline found that externally built tools succeed roughly twice as often as internal builds. Partnering works. The full playbook is to take the subsidized capacity from your vendors and point it at execution, and to pair it for the design with an Applied AI Partner whose only stake is your business's outcome. Thoughtful, intentional, systematic. Step back, map the whole, choose deliberately, and the odds move to your side. The capability on offer has never been greater. It is yours to direct.