The Gap Between Buying AI and Actually Using It
Deployment & Adoption
The AI adoption challenge is usually about people, not the tech.
Every enterprise has already bought the tools. The models are capable. The demos work. And yet, months later, usage adoption is still low, pilots never move beyond the pilot stage, and the expected business value is nowhere close to showing up in the numbers.
The gap is not simply between having AI and using AI. It is the gap between what the technology can do and what people and the business are actually ready to use.
My PhD research into enterprise AI adoption highlights this gap repeatedly: trust, communication, governance, unclear ownership, organisational culture, and different perspectives between technical teams and business stakeholders can all affect whether an AI moves from an idea into real-world use.
Deployment and adoption are not the same problem
Deployment is the technical question does the system work, is it integrated, is it secure, can it scale.
Adoption is the human question: do people actually use it, does it change how work gets done, does the value show up where the business can see it.
The architecture can be sound. The integration can be clean. The security controls can be in place. And six months later, almost nobody outside the original pilot team has touched it.
Common barriers to AI adoption
My PhD research applied a qualitative approach, based on 30 interviews with enterprise AI practitioners globally. It points to a recurring pattern: the barriers to AI adoption are often less about whether the technology works and more about whether the organisation is ready to use it
Trust, governance, unclear ownership, organisational culture, communication, and the lack of a shared understanding between technical teams and business stakeholders can all create friction.
Why this is a business problem now, not a technical one
Getting access to AI is no longer the hard part. The harder part is making it work across the business: knowing which problems to solve, deciding who owns the rollout, measuring whether it is delivering real value, and connecting what AI can do with what the business actually needs.
The graph highlights where this gap can emerge. An organisation can move quickly from AI experimentation to technical deployment, while the business side is still trying to understand the use case, ownership, process impact, governance, and what success actually looks like
That creates a familiar pattern:
This is where the translation gap becomes important.
Someone needs to connect the technical capability with the business problem. That means translating business needs into something that can actually be deployed, while also translating the technology’s capabilities back into something stakeholders can understand, trust, and use.
That translation work is the real bottleneck. It’s not a coding problem. It’s a stakeholder problem, and it’s the reason so many AI initiatives stallgraph between “we bought it” and “we use it.”
That translation is often the real bottleneck.
It’s not just a coding problem. It’s a stakeholder problem, and it’s one reason AI initiatives stall somewhere between “we bought it” and “people actually use it.”
What actually closes the gap
From my research and experience with enterprise AI adoption, several practical principles stand out:
• Start with a small number of clearly defined use cases, rather than running a hundred parallel experiments.
• Give deployment clear ownership, including someone who can translate between technical teams and business stakeholders.
• Measure usage and business outcomes from day one, not simply whether the technology was purchased or deployed.
• Build training and change management alongside the technical rollout, rather than treating them as an afterthought.
• Use successful deployments as templates, so that what works can be repeated and scaled across the organisation.
The goal isn’t simply to deploy more AI.
Making the translation gap visible
One reason AI initiatives stall is that the decision to deploy AI is often made differently from the way the solution eventually needs to work in the business.
A useful way to think about this is to separate fast, instinctive reasoning from slower, evidence-based deliberation. The point isn’t to eliminate instinct. It’s to recognise when instinct is driving a decision that needs more evidence.
Take a scenario many enterprises are facing right now: a CIO that his enterprise run on Google cloud platform (GCP) deciding between two major AI LLMs models, such as Google Gemini and OpenAI Enterprise.
The early reasoning might sound like:
“We already run on Google Cloud, so Gemini should integrate smoothly.”
“But OpenAI is clearly leading on capability.”
Both statements may contain useful signals, but neither is enough to make a deployment decision.
One reflects existing technology investments and familiarity. The other reflects a perception of capability. Neither answers the questions that ultimately determine whether the AI will work in the organisation.
The real questions are much more practical:
• Does it perform well on our actual business workloads?
• Can we integrate it into the processes and systems people already use?
• What will it really cost to deploy and operate at scale?
• Can we govern it appropriately?
• Who will own the rollout?
• Will employees actually use it ? & How will we measure whether it is delivering value?
The important question isn’t simply:
“Which AI platform should we choose?”
It’s:
Ramy Salem
SAP ANZ
PhD Candidate – AI Agents for Enterprise Architecture & AI Adoption
Follow me on LinkedIn for more on AI deployment, adoption, and enterprise AI research.
#SAPCHANNEL