Follow zdnet: Please add it as a priority source On Google.
Important points of ZDNET
- Only 5% of AI projects will be delivered. It comes down to the ability to customize.
- With a partnership in place, AI will double the success.
- Ask the right questions before making a decision between the building and the purchase.
There is an incredible gap between AI aspirations and actual successful projects – this was shown in recent MIT study It turns out that only 5% of the generated AI projects provide measurable value to the company. What do the top 5% do differently? Their general denominator is that their technology teams are mastering the art and science of high customizing AI in their business, while promoting partnerships and a Go-it-Alone approach.
They're deeper – very deep.
Successful AI projects do differently
According to a research team of Aditya Challapally, Chris Pease, Ramesh Raskar and Pradyumna Chari, AI efforts “focus on narrow but valuable use cases, deep integration into the workflow, expanding through continuous learning rather than a broad set of features.” “Domain flow and workflow integration are more important than flashy UX.”
Also: I retested my coding skills for GPT-5 using Openai's guidance – and now I trust it even more
Ultimately, it's not about building or purchasing AI just to use it. This is about how businesses benefit from AI. Instead of “the fight against outdated SaaS playbooks,” they added, experts “need to attract enterprise attention through aggressive customization and collaboration with real business issues.” “Exceptional performers don't build generic tools, they embed themselves in their workflows, adapted to context, and scaled from narrow but valuable scaffolding.”
It is worth noting that “plug and play AI is a myth.” Paul McDonna Smitha senior lecturer at MIT Sloan Executive Education told ZDNET. (McDonagh-Smith was not directly involved in this study.) “External tools save time, but the actual work uses “plug and personalized AI” to customize the AI tools to suit your workflow. ”
The reason why Gen AI tools such as ChatGpt are successful in pilots is “because of flexibility,” he added. “However, mission-critical work often fails due to factors including lack of memory, which reduces the ability to learn, adapt and customize to the extent necessary to effectively integrate with your daily workflow.”
Also: Stop using AI for these nine work tasks – this is why
Establishing strategic partnerships to advance AI makes a huge difference, according to MIT research. The co-authors observed much more builds than they purchased the initiative, and the partnership was successful twice as often as internal development efforts. Such partnerships “have high value, reduce total costs and provide better alignment with operational workflows. Companies avoid building overhead from scratch while achieving tailored solutions.”
When to buy
Still, AI advocates and developers need to work with partners such as vendors and network partners and weigh them when it's best to develop in-house. When making such a decision, “it creates a turning point when speed, scale or specialized expertise is required, and the company's team is not ready to reach the timeline needed for the business,” said McDona Smith. “It makes sense to build internally when a project is at the heart of its competitiveness, but it needs to be careful.
There is concern that using external solutions will reduce the opportunities for customization needed, but McDona Smith believes this fear is unfounded. “This is not a 'plug and play AI' case for existing and emergency workflows. They argue that AI success does not depend on the choice of sourcing external AI tools, but does not depend on internal ability to ensure that companies are suited to the way they think, work and act. ”
Also: These CFOs dedicate 25% of their AI budget to Agent AI
Other industry leaders agree that successful AI depends on individual circumstances. “Deciding to use your internal team or outsource to other vendors depends on what your organization wants from AI,” he said. David's FriendCEO and co-founder of Wasabi Technologies. “If technology such as AI is part of the company's central differentiation, or if the business competes for price and needs to be developed and managed internally, outsourced if it is not the core part of the company's offering.
This process should also begin by asking the right questions. “The question is not whether they can build technology, but whether they should.” Adrian Murraythe founder and CEO of Fisent Technologies pointed out. “You may have a very capable and well-funded technology team, but your capabilities are inherently limited and you need to focus on the most valuable effort. Teams need to focus on the efforts that can create differentiated value. These technologies need to be applied to specific business issues rather than building a core technology infrastructure that can be easily licensed from solutions.”
The nature of partnership relationships is also a determinant of AI success. This should be more than a transaction arrangement. “Top buyers treat AI startups like software vendors, like business service providers, and keep them on benchmarks close to those used by consulting companies and business process optimization providers,” according to the MIT report. This includes deep customizations tailored to the internal processes and data associated with the results.
Also: Four Ways to Turn AI into a Business Advantage
“Make sure that existing emergent workflows are broken down and analyzed whether they work with genai or not, and evolve and recombine when genai is ready,” advised McDona Smith.
Employing grassroots AI within the company
Often, successful AI efforts begin at the grassroots level, and this study also shows. “Many of the most powerful enterprise deployments started with employees who had already experimented with tools like ChatGPT and Claude for their personal productivity,” the study co-author reported. They “intuitively understood the capabilities and limitations of genai and became the early champions of internally approved solutions. Rather than relying on central AI capabilities to identify use cases, successful organizations allowed budget holders and domain managers to represent issues, veterinary tools and lead rollouts.”
Also: How AI-enabled autonomous business will change the way you work forever
Agent AI architectures are also emerging and are supported by frameworks such as Model Context Protocol (MCP), Agent-to-Agent (A2A), and NANDA, which allows agent interoperability and coordination. “These frameworks form the basis for the Emerging Agent Web, a mesh of interoperable agents and protocols that replace monolithic applications with dynamic tuning layers.”
Culture shifts required for AI
Working with AI vendors “can start a critical early momentum, but heavy lifting will translate AI solutions into processes, policies, practices and, of course, people and culture,” says McDonagh-Smith.
