Artificial intelligence has become embedded in everyday business operations. From writing emails and analysing data to developing business plans and reviewing processes, AI can help people work through problems, explore ideas and make better use of their time.
But there is a subtle risk that people using AI should understand, it does not have to produce incorrect information to give advice that leads to a poor decision. The recommendation can be perfectly reasonable, but it doesn’t fit the business, its customers or decisions that have already been made.
This becomes more noticeable when AI is used over several weeks or months on the same project. As conversations develop, people can become accustomed to relying on the assistant’s recommendations, assuming the AI has kept up to date with the conversation and this is where the tendency to over rely and stop checking can creep in.
What is AI drift in business?
Imagine working with an AI assistant to develop a new business website. Over several conversations, you establish the target customers, decide how the services should be presented, agree on the tone of voice and choose a straightforward page structure.
A few weeks later, you ask the assistant to help update some content. Instead of working within the existing structure, it recommends redesigning sections, changing the language or introducing features you previously decided against. The suggestions might be sensible for another business, but they no longer reflect the direction you’ve established.
This is one way people may experience what is known as AI drift: the responses move away from the original context, decisions or objectives. It is a useful way to describe the problem, although it doesn’t establish the technical reason for the change.
For a business, the practical concern is straightforward. If nobody notices that the recommendations have changed, time and money could be spent revisiting decisions or replacing work that was already appropriate.
Can AI forget previous conversations and business decisions?
AI assistants don’t necessarily have access to every detail of your previous conversations, this depends on the system being used, the conversation, its settings and how relevant context is provided.
Even when earlier information is available, an assistant may fail to use it correctly. It might overlook an important decision, misunderstand the reason behind it or respond primarily to the latest question without considering the wider project.
We experienced something similar while working on a project. After moving work from a long-running conversation into a new one, some of the advice became more generic. Decisions about the website structure were revisited, and assumptions about the technical setup needed correcting.
We had retained working notes and records of previous decisions, so we could identify the inconsistencies and restore the relevant context. Without those records, it would have been much harder to distinguish a genuinely useful new recommendation from one that simply overlooked earlier work.
This experience doesn’t prove why the assistant’s behaviour changed as there may have been several contributing factors. What it demonstrates is the importance of keeping essential business decisions somewhere other than an AI conversation.
Why does AI sometimes give generic business advice?
AI assistants can draw on familiar approaches to business problems. This is often useful, particularly when someone is exploring a subject for the first time or wants to understand common practice. However, a widely accepted approach isn’t necessarily the right one for every business.
Consider a company that values informal, friendly communication with its clients. Following a meeting, the owner sends a short email summarising the discussion and inviting the client to correct anything that has been misunderstood. The owner asks AI to improve the wording, and the assistant suggests:
“If this accurately reflects your understanding of our discussion, please confirm and we can agree the next steps.”
There is nothing inherently incorrect about that sentence. It is clear and professional, and it asks for confirmation and in some circumstances, it might be entirely appropriate. But the business owner might prefer:
“If this is all okay with you, please let me know and we can proceed.”
That version is less formal and more consistent with the relationship the business wants to build. It still gives the client an opportunity to raise questions or correct a misunderstanding. The difference isn’t about grammar or accuracy. It’s about understanding how the business works and what it wants to achieve through its communication. This is why AI-generated advice needs context. A recommendation can follow accepted professional practice while still being unsuitable for the people who will use it.
What are the risks of relying on AI advice without checking it?
When businesses use AI to support decisions, the consequences of an unsuitable recommendation depend on what is being changed. An overly formal email might simply make a conversation feel less natural. A recommendation to replace a reporting system, reorganise a business process or change a technology platform could have much greater implications.
For example, a business might ask AI how to improve its customer reporting. The assistant recommends introducing a new dashboard and reporting tool, explaining the benefits of automation and centralised information. Those benefits may be genuine, but they don’t establish that the business needs a new system. The existing reports may contain everything managers require, but the underlying data is inconsistent. Employees may not understand how to interpret the information or, alternatively, the reports might be accurate and useful, but arrive too late to support decisions.
Each situation requires a different response and introducing new technology without understanding the actual problem could add costs and complexity without improving the outcome. The same principle applies when AI recommends changing an approach that has already been agreed. Before accepting the recommendation, businesses need to understand whether something has genuinely changed or whether the assistant is approaching the question without the full context.
How can businesses keep AI advice consistent?
One useful practice is to keep a brief record of important decisions as a project develops. For example, recording that a particular website layout was selected is helpful. Recording that it was selected because customers needed to find service information quickly, and that a more complicated layout had already been rejected, is much more useful.
The reasoning gives future decisions something to be assessed against. If AI later recommends the rejected layout, the business can ask what has changed to make it a better option. It also helps to provide relevant background information when starting a new AI conversation. Business objectives, technical arrangements, agreed working practices and previous decisions can all influence the quality of a recommendation. An assistant should not be expected to reconstruct a project’s history reliably without the information it needs.
Businesses can also ask AI to identify uncertainty rather than make assumptions. If the assistant doesn’t know which systems are being used or why a decision was made, acknowledging that gap is more useful than producing a confident explanation based on incomplete information. Most importantly, a new recommendation should be considered on its merits. AI should be able to challenge an earlier decision, but the challenge should be supported by a reason that matters to the business.
Should businesses challenge recommendations made by AI?
Yes, especially when those recommendations affect costs, operations, customers or established plans. Challenging AI doesn’t mean rejecting its advice, it just means understanding the reasoning before deciding whether to act.
A useful question is: “What new information makes this approach better than the one we already agreed?” A business requirement may have changed, new information may have become available or an earlier assumption may no longer be valid. Or, the recommendation may simply reflect a more common approach, without explaining why it would work better in that particular business.
Human judgement remains important because business decisions involve more than identifying technically possible solutions. They also depend on resources, priorities, customer expectations, staff experience and the practical consequences of making a change. An AI assistant can help investigate these considerations, but the business still needs to make the final decision.
How should businesses use AI to make better decisions?
The starting point should be the business problem rather than the technology available to address it. If a company wants to improve reporting, it should first understand what decisions the reports need to support. If it wants to automate a process, it should understand how the process currently works and why improvement is needed. If it wants to use AI to develop a strategy, it should establish its objectives and constraints before asking for recommendations.
AI can then help explore the options, identify questions worth investigating and challenge assumptions. Used in this way, it becomes a useful part of the decision-making process rather than an unquestioned source of instructions.
Before making a significant change, it is worth asking whether the recommendation reflects new evidence, a genuine business need or simply a different way of approaching the same problem. After all, when an AI assistant suddenly suggests changing direction, the question may not be whether it has found a better answer. It may be that it has forgotten why you chose the original one.
Is your technology helping your business move forward?
AI and other technology can create real opportunities, but introducing new tools isn’t always the answer. Sometimes the best place to start is by understanding what’s already working, what isn’t and where improvements would make a genuine difference.
WiseWhale’s Tech Roadmap service provides an independent assessment of your existing technology and practical recommendations based on your business needs.Explore our Tech Roadmap service