A manufacturing company had already set aside funds in the millions to help modernize their business using AI and nobody had yet to explain why this was. The CFO was seeking efficiencies in operation. The CIO’s desire was to get infrastructure modernized. The sales people were calling for generative AI copilots. Legal felt that governance needed to be further tightened, as employees had already been copying and pasting information from the private tools which included confidential information into public AI tools.
Nobody was aligned.
This is an incredibly common scenario these days and companies are frantically trying to look AI-ready in the eyes of their boards, investors and customers. The issue is that there are companies that still think that they’re adopting AI, but they don’t have an Enterprise AI Strategy.
It’s hardly comparable.
One provides measurable business leverage over time. The other makes for disjointed tooling, duplicated costs, security issues and an ever-increasing pile of disconnected pilots that quietly die six months after.
Some companies are buying AI tools faster than they can govern them
I believe the market was surprised to see how quickly the adoption of AI purchasing would take place within enterprises.
Five years ago, a lot of technology purchases were typically made by the IT leader. That control layer is pretty much broken with AI products, since teams can individually sign up to different cloud-based solutions, rather than having to go through lengthy enterprise sign-off processes.
AI Copy Tools are used by marketing teams.
HR test AI recruitment systems.HR test AI recruitment systems.
Coding copilots are experimented by engineering teams.
Financial institutions analyze AI predictive tools. AI predictive platforms get assessed by finance organisations.
Then suddenly, leadership discovers that the sensitive company data is going through 15 different AI vendors that have a variety of governance policies.
All things begin to cost money from there.
Generative AI investment accelerates even further, reports McKinsey & Company, albeit with a large number of organizations still facing challenges in realizing a measurable value for the enterprise from experimentation. This isn’t that surprising, as many companies continue to work on AI on a tactical basis, rather than a business basis.
They want to pursue tools and don’t know what operating models they need.
The first step in an actual strategy is a much less glamorous one.
The first step is to determine what business processes are truly impactful economically from the use of AI, and what are mostly fads that they can focus on now.
That’s pretty self-explanatory, unless you’re in an actual executive planning meeting.
Why your data quality matters more than your AI model selection
This is likely the least sexy aspect of enterprise AI planning, but that’s why companies put it on the back burner until their projects go awry.
One of the logistics companies I worked for was so fascinated by implementing AI forecasting systems in inventory management. Leadership took months to research vendors and to compare the performance of the models.
Meanwhile, there were a lot of issues with their internal operations.
There was not uniformity of naming conventions across the region for regions warehouses. History of shipments had duplicate fields. Customer fulfillment time was different for each system as there had never been a complete operational integration of the acquisitions.
The AI system didn’t cause the issue; that wasn’t the main issue. There was a business platform supporting data for the underlying business environment.
That organization ultimately postponed deployment almost eight months to normalize operation data pipelines to get reliable outputs from the models for forecasting.
It was not anticipated that the most challenging time of implementation will be the “boring infrastructure cleanup” phase.
However, this takes place frequently.
A good Enterprise AI Strategy typically relies on the adoption of simpler models and more on whether:
- internal systems are standardized
- governance exists
- APIs communicate reliably
- historical data is usable
- business processes are documented clearly
If you don’t have those building blocks, AI adds to organisational confusion, rather than efficiency.
I’m growing more and more skeptical about enterprise AI roadmaps that place a disproportionate focus on the capabilities of the model, and not much else on maturity of the data.
The companies getting the strongest AI outcomes are surprisingly conservative
This is the opposite of the hype surrounding AI transformation, but it’s consistent with the experience of doing this day-to-day.
The best enterprise AI use cases are typically limited, managed and very mundane.
Not revolutionary.
Not organization-wide disruption.
[It] is just well-targeted operational improvement.
While others were publicizing their large-scale AI programs, one health care provider I coached was hesitant to get into any fancy-schmanzy customer experience with generative AI. Rather, they turned their attention to the process of internal claims processing, and how to minimize administrative overhead by employing tightly-controlled automation processes that are linked to existing processes.
That project cut down the delays in processing times by a considerable amount in a matter of months.
There was no press coverage relating to a virus. There were no executives that came up on the conference stages to speak about “AI reinvention.”
However, the company had started to add value in terms of operations before others were still testing out the proof-of-concept deployments.
The experience taught me that one of the things that I now strongly believe is that in enterprise AI maturity, it is typically not apparent early in the journey.
Best organisations make the most of quietly, prior to promoting loudly.
What enterprises misunderstand about AI infrastructure costs
Even though the topic of Artificial Intelligence (AI) has moved beyond the confines of software licensing, most budgeting discussions are still centered on software costs.
The consequences of the infrastructure damage are far larger later on. Particularly after organizations go beyond the point of being in isolated experiments.
A large retail business I had had worked for, initially thought that their AI spend would be mainly on SaaS subscriptions and API access. After 18 months they were having to deal with:
- GPU resource allocation
- vector database scaling
- inference latency optimization
- cloud egress costs
- identity access governance
- internal AI monitoring systems
- data retention policies
They put themselves into an operational structure that was very complicated, much more so than what leaders had imagined.
That’s one of the many driving forces for cloud providers to be investing so heavily in enterprise AI infrastructure services.
Platforms like:
Are moving toward becoming a complete enterprise Artificial Intelligence (AI) ecosystem and away from being a model provider.
They know that enterprises are looking for more governance, deployment, monitoring and compliance controls to consolidate their operations.
The use of the actual AI model is now one of many layers to the broader enterprise architecture discussion.
A surprising number of AI copilots create negative productivity initially
The surprise that all too many executives find difficult to accept:
Even if the AI is functioning correctly, its implementation can negatively impact productivity for a while.
This is what I witnessed when implementing enterprise software where development teams took a big bite out of AI coding assistants. Leadership demanded engineering velocity gains were to be realized virtually right away.
On the other hand, there was a delay in the development of the city, first because:
- workflows changed
- code review patterns shifted
- output validation increased
- security teams introduced additional governance controls
- junior engineers over-relied on generated code
This reduction in productivity required several months to get things back under control.
It’s not that it was a failure, it’s just that it wasn’t successful.
It involves the price of operational adaptation of the organization.
Too many AI conversations have missed this adjustment period and it makes sense that vendors would sell on the high side of productivity during the sales cycle.
In the real enterprise world, things are certainly different.
It takes time to get used to the changes that are in place with new systems, particularly those that impact the way decisions are made.
What a practical Enterprise AI Strategy actually looks like
Most enterprises do not need fifty AI initiatives simultaneously.
They need three things:
- operational prioritization
- governance discipline
- measurable business outcomes
That is it.
Had the opportunity to start building an enterprise AI roadmap in 2026, it would differ from my own and what most organizations are doing right now.
Identify one process, where employees do a measurable amount of work in a manual fashion that is repetitive, in which they waste time. For most use cases, it’s more likely to be customer support routing, claims processing, contract review, knowledge retrieval and classification for procurement, or other internal use cases where customers are not directly interacting with the AI.
Second, set up governance prior to scaling up deployment. Define:
- approved vendors
- acceptable data usage
- retention policies
- human review requirements
- escalation procedures
Prior to employees building their own out-of-control AI processes.
Thirdly, take bold steps to measure operational improvement.
Not a measure of the adoption of AI.
Not prompt volume.
Actual business outcomes:
- reduced processing time
- lower support backlog
- fewer manual reviews
- faster onboarding
- lower operational cost
Today’s organizations are making the most of AI, not merely as buzz.
Why leadership alignment matters more than technical sophistication
A major threat to enterprise AI momentum is having distinct measures of success in each department.
I saw this go down during one of the financial services implementations that I worked on, where:
- IT prioritized infrastructure scalability
- legal focused on regulatory exposure
- operations wanted automation
- executives wanted public AI positioning
- customer support wanted staffing relief
All legitimate goals.
However, they were never deployed together.
The outcome of this was the lack of prioritisation, duplication of tooling assessment and intra-company disputes regarding governance ownership. They put in a tremendous amount of effort in planning yet not doing.
To create a successful AI roadmap, it’s important to have executive buy-in with regard to:
- primary business objectives
- acceptable operational risk
- implementation sequencing
- success measurement
- governance ownership
If they were not aligned, it would be second guessing chaos management.
Organizational issues are typically more complicated than technology issues.
Enterprises adopting AI too late may face a real competitive problem
I also believe it’s a risky thing to go to the other end.
However, some are so conservative that they are falling behind the times in terms of their business operations.
IBM Global AI Adoption Index research over the past few years shows that AI is being used more in enterprises further in their core workflows than as an experimentation environment. The shift is significant because operational AI maturity builds up over time.
Companies that are adopting AI into:
- internal search
- software development
- customer operations
- forecasting
- analytics
- workflow automation
Are slowly creating competitive advantages within the organisation that are hard to be easily copied afterwards.
Especially once they begin to compound internally with knowledge of the product, and process optimization.
It’s not just about whether to use AI anymore; it’s about how to use it safely and effectively.It’s no longer just about the question of whether to use AI, but how to do it safely and effectively.
Now the question is: “How long can we wait until an acceptable efficiency gain is achieved with operational integration before the competitors start making real progress?”
There are a lot more challenging questions about that.
The smartest first step is usually smaller than leadership expects
A successful Enterprise AI Strategy rarely begins with a bold transformation announcement.
Instead, organizations usually start with a single operational improvement that delivers measurable value, helps teams understand responsible AI governance, and builds the confidence needed to scale AI initiatives over time.
I’d rather see one of the painful internal processes successfully automated instead of announce 20 disjointed pilots of AI not to end up in production systems.
Volume of experimentations is not as important as operational discipline as the market comes into a stage where it is expected.
But to be honest, that’s something that’s long overdue.
Author Bio
Talha Qureshi is an enterprise technology analyst and blogger with over a decade of hands-on experience across cybersecurity, cloud infrastructure, B2B SaaS, and enterprise AI. He writes about the gap between how enterprise technology is marketed and how it actually performs in real organizational environments.
