Why Adding AI to an Old Structure May Leave the Same Problems
A customer enquiry arrives. An AI tool drafts a reply in seconds, summarises the account history and prepares a quotation.
This article explores leadership and business strategy, a theme developed in depth in SMA Code by Dr Sam M Attia.
Then the quotation waits for approval. Nobody is sure who can authorise the discount. The sales team has different availability information from operations. The manager who usually resolves the disagreement is in meetings.
The preparation is faster. The customer is still waiting.
This illustrative situation captures a practical challenge for businesses adopting artificial intelligence. A tool can improve a task while leaving the process around it largely unchanged. If responsibility is unclear, information is unreliable or decisions depend on one person, those weaknesses still shape the outcome.
Recent research puts organisational design at the centre of this conversation. The question for leaders is how the business needs to work differently for the technology to produce a useful result.
What recent research is telling leaders
In its report published on 12 October 2026, The Conference Board identifies work redesign as a leadership responsibility as AI develops. Its findings highlight the need for clearer accountability, defined decision rights and deliberate human review at important stages.
IBM’s September 2026 research adds another perspective. Its surveys covered 1,500 senior HR leaders and 8,800 employees globally. Among the HR leaders, 80% believed AI adoption creates additional work such as validating recommendations, correcting mistakes and managing exceptions. Meanwhile, 42% of employees said AI increased their work or that their work went unrecognised.
These are reported experiences and perceptions from the surveyed groups. They do not establish that every AI deployment produces the same outcome. They do, however, give leaders a reason to examine what happens around the tool, including the work needed to make its outputs usable.
A faster task can expose the next bottleneck
Consider a team preparing proposals. Drafting used to take several hours, followed by a short review from a senior manager.
AI reduces drafting time considerably. The team now produces more proposals each day, but the same manager must still review every one. The queue moves from preparation to approval.
There may still be a useful saving. Yet the overall improvement depends on whether the rest of the process can handle the change.
The team could examine which proposals genuinely require senior judgement, which can be approved within agreed limits and what information reviewers need. That work addresses the bottleneck directly.
Without it, a successful demonstration of faster drafting can coexist with a disappointing customer experience.
Unclear responsibility becomes harder to manage
When an AI tool produces a recommendation, someone must decide what authority that recommendation carries.
Can an employee send it to a customer? Does a manager need to review it? Who checks that the information is current? Who takes responsibility if the recommendation is unsuitable?
If these questions remain unanswered, employees may behave differently. One person checks every detail. Another assumes the output is ready to use. A third sends everything upwards because they are unsure what they can approve.
The resulting inconsistency comes from how the work is organised.
A practical starting point is to assign responsibility for each stage, including checking the source information, reviewing the output and authorising the final action. People also need a clear route for raising concerns or stopping a process when something appears wrong.
Conflicting information still needs an owner
Imagine that a sales spreadsheet lists one service promise, an operations document lists another and an old presentation contains a third.
Introducing an AI assistant does not, by itself, settle which promise the business should make. Someone must establish the agreed position and maintain the information that supports it.
Before connecting a tool to business records, examine how those records are managed. Identify the authoritative source for important facts, who updates it and how outdated material is removed or marked.
This is particularly useful where several teams contribute to one customer outcome. Sales, finance and operations need a shared understanding of what has been agreed.
Good information management gives both people and technology a more dependable starting point.
Measure the work that follows the output
A quick first draft is easy to demonstrate. Its usefulness becomes clearer after someone has checked and applied it.
Suppose a report takes five minutes to generate but requires substantial correction before it can support a decision. The generation time tells only part of the story.
A sensible assessment follows the work through to completion. Include the time spent reviewing, correcting and handling exceptions, alongside the time saved preparing the output.
Also examine the result. Did the customer receive an accurate answer sooner? Were fewer orders returned for correction? Could the team complete more work to the required standard?
These questions help distinguish visible activity from a meaningful improvement in performance.
Start with a process people can explain
Choose one recurring process with a clear purpose. It might be preparing quotations, responding to customer enquiries or producing an internal report.
Ask the people doing the work to describe what happens from start to finish. Where does information come from? Where does work wait? Which decisions require experience? What usually causes rework?
Agree what a successful outcome looks like before introducing the change. For quotations, that could include the time from enquiry to an accurate, approved offer, together with the number needing correction.
Then decide where AI could contribute and where human judgement remains necessary. Set out the checks, approval limits and escalation arrangements in language the team can use.
Run a limited trial and review the complete process. If one step improves while another becomes overloaded, adjust the workflow before extending it.
The structure may need refinement rather than replacement
An established organisation can have valuable experience, strong relationships and effective controls. Its age alone says little about whether its structure is suitable.
The issue is whether that structure supports the work the business now needs to do.
Some improvements may be modest. A clearer approval limit, a shared source of information or a named process owner could remove a recurring delay. Other changes may require responsibilities to be redistributed across teams.
Leaders need evidence about where the difficulty lies. Changing an organisation chart without understanding the work can leave the same confusion under different job titles.
Make the operating model part of the AI decision
AI adoption gives a business an opportunity to examine habits that have become familiar. It can reveal where decisions wait, where information conflicts and where responsibility has never been made explicit.
The opportunity is to improve those arrangements while introducing the technology. Give people clear authority, reliable information and a practical way to review what is happening.
A useful test is whether work reaches a better conclusion, with fewer avoidable delays and a standard the business can sustain.
For a deeper exploration of business structure, leadership and turning strategy into consistent execution, discover The SMA Code by Dr Sam Attia.