A company can spend months evaluating artificial intelligence platforms and still overlook the question most likely to determine whether the investment succeeds: does it know how its own business works?
Consider a sales department in which customer information sits across several systems, managers disagree on what qualifies as a promising lead, and experienced employees follow procedures that have never been formally documented. An AI tool could help the company respond to enquiries faster or identify potential customers. It could also apply inconsistent rules more efficiently.
The distinction is consequential. Artificial intelligence can improve the execution of a process, but it cannot automatically resolve disagreements about what that process should be.
For many companies, this is becoming the less visible challenge of AI adoption. The technology is advancing rapidly. Organisational readiness is proving harder to buy.
McKinsey's 2025 global survey found that 88% of respondents said their organisations regularly used AI in at least one business function. Yet nearly two-thirds said their companies had not begun scaling AI across the enterprise. Only 39% reported an impact on earnings before interest and taxes at the enterprise level. The findings suggest a gap between deploying AI in individual applications and translating it into broader financial results.
The difficulty may have less to do with the sophistication of a model than with the organisation expected to use it.
The cost of automating ambiguity
Businesses have spent decades accumulating processes that work well enough to survive, even when they are difficult to explain.
A customer order may pass through several departments, with employees correcting errors along the way. A finance team may reconcile figures manually because two systems calculate the same metric differently. A manager may approve exceptions based on experience rather than a documented policy.
These arrangements can persist for years because employees know how to compensate for their shortcomings.
Automation removes some of that informal flexibility. AI introduces another complication: it may be asked to interpret instructions, recommend actions or generate outputs in situations where the underlying rules are unclear.
A company seeking to automate lead qualification, for example, must first decide what makes a lead valuable. Is it the size of the prospective contract, the likelihood of conversion, the customer's long-term value or the speed at which a deal might close? Different answers will produce different priorities.
No model can settle that commercial question on behalf of management.
The sensible response is not to abandon automation. It is to examine the process before automating it. That may involve eliminating unnecessary steps, agreeing on decision criteria and determining which exceptions require human intervention.
The work is less visible than a product demonstration. It is also more likely to determine whether the investment delivers lasting value.
More data does not necessarily mean better decisions
Data quality is a familiar concern in technology projects. Data ownership is a more difficult one.
In many organisations, customer information is maintained by sales, billing details by finance and delivery records by operations. Each department has its own systems, priorities and definitions. A customer's status in one database may not match its status in another.
An AI assistant asked to prepare a customer briefing may combine these records into a polished summary without making the underlying contradictions obvious. The result can look authoritative while remaining incomplete.
The problem is not simply that the company has bad data. It may have no agreed authority over the data it uses.
Who decides which customer record is correct? Who is responsible for updating it? Which information can be shared with an AI system? How quickly must an error be corrected?
These questions become especially important when AI is connected to systems that trigger actions rather than merely generate text.
Companies need not undertake an expensive, organisation-wide data overhaul before every AI project. They do need to establish reliable information and clear ownership for the particular task they intend to automate. A system designed to answer routine internal questions has different requirements from one that influences credit decisions, financial reporting or customer eligibility.
The standard should be proportionate to the consequences of getting the answer wrong.
The knowledge that never made it into the system
A company's most useful information is not always stored in a database.
It may reside with the employee who knows why a particular supplier requires additional checks, the account manager who recognises early signs of customer dissatisfaction, or the finance professional who understands which anomalies indicate a genuine problem.
This accumulated judgement can be difficult to replace. It can also become a liability when the organisation depends on a handful of people to keep essential operations running.
The difficulty becomes apparent when companies introduce internal AI assistants intended to help employees find policies, answer operational questions or navigate procedures. If the relevant information is scattered across outdated documents, private folders and informal conversations, the system lacks a dependable basis for its answers.
The temptation is to treat this as a software problem. Often, it is a management problem that software has made harder to ignore.
Documentation does not need to capture every decision or turn experienced employees into rule-following machines. It should identify the procedures, exceptions and principles that others need to understand in order to perform the work reliably.
That exercise has value beyond AI. It reduces dependence on individual employees, improves onboarding and makes processes easier to review.
AI simply gives businesses another reason to do it.
When management has no single answer
The most persistent obstacle may be disagreement at the top.
One department is rewarded for speed; another is judged on accuracy. Sales promises delivery dates that operations considers unrealistic. Managers apply different standards to customer complaints because the company has never established a consistent policy.
Introducing AI into these environments can turn local inconsistencies into organisation-wide ones. It may also force decisions that management has postponed.
This is particularly important when systems are expected to act with limited supervision. A tool that drafts a response can be reviewed before it reaches a customer. A system that issues refunds, changes records or initiates transactions requires much clearer boundaries.
The National Institute of Standards and Technology's AI Risk Management Framework places governance, accountability and ongoing risk management at the centre of responsible AI use. These are not merely compliance concerns. They determine who is responsible for a system's behaviour, how its performance is evaluated and what happens when it fails.
Technology can support a decision. The organisation must still own it.
Readiness is a business decision, not a finishing line
None of this means companies should wait until every process is documented and every database is clean. Such a standard would make AI adoption impractical.
Readiness depends on the task. A company experimenting with AI-generated marketing drafts can tolerate a different level of uncertainty from one automating financial approvals or handling sensitive customer information.
A more disciplined approach begins with a defined business problem. Establish how the work is performed today, identify the information and decisions it depends on, and set a measurable standard for improvement. Test the system on a limited scale. Examine errors, exceptions and the amount of human intervention required, not just the time saved.
If the results are reliable, expand the application. If they are not, determine whether the limitation lies in the model, the data, the process or the assumptions behind the project.
That distinction matters. Replacing a tool will not resolve a disagreement over company policy. Buying a more capable model will not make fragmented records consistent. And automating a process that creates unnecessary work may simply make the waste more efficient.
The real investment may therefore begin well before the technology is deployed.
Businesses that understand their processes, assign responsibility for their information and make their decision-making explicit will have a stronger foundation for using AI effectively. Those that do not may find that the most expensive part of adoption is not the software, but the organisational work they assumed the software would eliminate.
The question for executives is no longer simply which AI system to buy. It is whether the company has made the decisions, established the controls and clarified the processes that allow such a system to be useful.
At Quantazone, we examine the business decisions behind successful technology adoption, from AI and business systems to cloud and marketing. Stay tuned for more perspectives on turning technology investment into measurable business value.
Top comments (1)
Very true! Just buying AI tools isn’t enough. Businesses also need the right processes in place to actually make the most of them.