Most articles on this topic are written by companies that sell AI products. The conclusion is always the same: yes, you’re ready, here’s how to get started, contact us.
We build AI implementations for businesses, so we have a commercial interest in this question too. The difference is that we also tell people when the answer is no — which happens often enough to be worth writing about.
Here is an honest framework for the question.
The question is usually too broad
“Is our business ready for AI?” is the wrong question because it treats AI as a single thing. It isn’t. AI-assisted document review is a different application from an AI scheduling agent, which is different again from a private model trained on your operational data.
A better question is: is there a specific, repetitive task in our business where AI would create measurable value — and can we implement it in a way that is reliable and doesn’t create more problems than it solves?
That question has a real answer. The generic one rarely does.
What AI is actually good at
These are the conditions under which AI reliably creates value for businesses.
You have a volume problem, not a complexity problem. AI works best when the challenge is how many rather than how hard. Reviewing two hundred contracts for a specific clause is a volume problem. Advising on the strategic implications of a non-standard clause in a particular relationship is a complexity problem. AI helps substantially with the first. It can assist with the second, but it cannot replace the expertise.
The work follows a consistent pattern. AI learns and applies patterns. If the task varies significantly in structure from instance to instance, the AI’s output will vary in quality in ways that are hard to predict and harder to catch. Extracting and categorising data from invoices follows a pattern. Diagnosing why a client relationship has deteriorated does not.
A human is reviewing the output before anything consequential happens. The highest-value AI applications in most businesses are ones where the AI drafts or processes and a person reviews before the output matters. The AI reduces the time cost; the human maintains accountability for the result. Applications where output goes directly into a consequential process — with no review — require a much higher confidence threshold than most implementations meet.
Time is the cost you’re already paying. If you can describe a task and estimate how many hours per week it consumes across your team, you can assess whether an AI implementation is worth building. Tasks that cost two hours a week are rarely worth it. Tasks that cost twenty hours a week almost always are.
What AI is not good at
Being honest about this matters, because overconfident AI implementations create problems that are harder to fix than the ones they were meant to solve.
Processes that exist only in someone’s head. If the way a task gets done cannot be described step by step — if it lives in institutional knowledge or individual judgement that has never been written down — AI cannot reliably replicate it. The prerequisite for augmenting a process with AI is being able to describe it clearly enough that you could hand it to a new employee. If you cannot do that, start there.
Problems that are actually people problems. AI makes workflows faster. It does not fix unclear ownership, poor communication, or misaligned incentives. If those are the real issues, an AI implementation is an expensive way to discover that.
Replacing professional judgement entirely. This is where the hype is most dangerous. AI can draft a contract, but it cannot determine whether the terms are appropriate given the client relationship, the negotiating context, and what is actually at stake. AI can flag anomalies in financial data, but it cannot tell you whether a given anomaly is significant. The right frame is augmentation of expertise, not replacement of it.
Your first hire. If you are understaffed and hoping AI will substitute for a person you need, it usually won’t. The implementation cost, the oversight required, and the reliability limitations mean AI typically saves time for people you already have — it rarely eliminates the need for people you haven’t hired yet.
The questions that tell you whether a specific workflow is worth it
If you are trying to assess a particular task in your business, work through these.
Can you describe the process step by step? Not at a high level — step by step, including the decisions made and the inputs required at each stage. If you can describe it clearly, you can evaluate whether AI can replicate it. If you cannot, the process needs to be documented before anything else happens.
How many times does this happen? Per day, per week, per month. Volume is the primary driver of return on investment. A process that happens twice a month will rarely justify a significant build. A process that happens fifty times a week almost always will.
What does it cost you now? In time, in errors, and in the downstream consequences of those errors. Be honest about the number — including the parts that do not show up on a timesheet. The context-switching, the double-checking, the work that falls through the cracks.
What happens if the AI gets it wrong? Every AI output has an error rate. The question is what the consequence of an error is in this specific workflow. If a person reviews every output before it matters, the risk is managed. If the output goes directly into a client-facing document or a financial record, the accuracy threshold is much higher — and the implementation needs to reflect that from the start.
Is the data sensitive? If the workflow involves client information, financial records, or anything subject to professional confidentiality obligations, the architecture matters as much as the capability. A cloud-based AI tool may handle the task well but create data handling problems you discover later. An on-premise deployment — where nothing leaves your network — may be the appropriate approach. This question needs to be answered before the implementation is designed, not after.
What “ready” actually looks like in practice
It is not a checklist completed before anything is built. It is not an eighteen-month data infrastructure project. It is not hiring a strategy consultant to produce a report before any AI is deployed.
In practice, the businesses that get the most value from AI identify one specific, high-volume, time-consuming workflow — one where the current approach is clearly costing them — build something focused on that, measure the result, and expand from there.
A small accounting firm that automates extraction and categorisation of data from client documents saves real hours in a practice where time is the inventory. A plumbing company that generates job completion reports from technician voice notes instead of having field crews fill out forms at the end of the day saves the time and gets better data. A legal practice that summarises long documents before a senior lawyer reviews them changes the economics of certain types of work.
None of these required the business to be “AI-ready” in the abstract. They required a specific problem, an honest assessment of the approach, and an implementation built for the actual workflow — not a generic use case the vendor happened to have on the shelf.
The honest conclusion
Most businesses we talk to are not unready for AI. They have not yet identified the right starting point.
The question worth answering is not “are we ready” but “what is the first thing worth building.” The first has no real answer. The second almost always does — if you are willing to look honestly at the business rather than at the hype around the technology.