The pressure to "do something with AI" is coming from every direction: software vendors, the board, competitors who claim to be adopting it. And amid that pressure, many companies make rushed decisions: they buy tools they don't use, hire consultancies that produce documents without implementation, or try to automate processes that first need to be put in order.
The uncomfortable truth: not every company is ready to implement AI today, and forcing it too early is one of the most effective ways to waste budget.
This guide is meant to help you know where you really stand.
Question 1: Do you have historical data for the process you want to automate?
AI — machine learning in particular — learns from data. If you want a model that predicts which customers are at risk of not renewing, you need historical renewal and cancellation data. If you want to automate support ticket classification, you need thousands of tickets that are already classified.
If the process you want to improve with AI doesn't have enough historical data, the implementation won't work. The dataset has to be built first.
Important exception: Language models (LLMs) like GPT or Claude are different. They don't need your data to work — they come pre-trained. If you want a customer service assistant that answers questions about your products, you can start with the documentation you already have.
Question 2: Is the process you want to automate documented and standardized?
There's a phrase we use internally: "automating a messy process only produces chaos faster".
If today your credit approval process depends on each analyst's personal judgment, with no documented criteria and no clear rules — you can't automate it with AI. First you have to design the right process, document it, validate it. Then automate it.
AI can handle variability within known rules. It can't invent the rules.
Alternetica
Want to evaluate an AI use case in your company?
We tell you honestly whether it makes sense for your current situation, what is needed and how long it would take. No cost.
Request an evaluation →Question 3: Do you have at least one internal person who will own the implementation?
This is the factor that most often determines whether an AI implementation works or dies forgotten.
An AI implementation is not a project that gets "delivered" and is done. It is a system that needs continuous feedback, adjustments, improvements and someone who actively adopts it within the team. Without an internal owner — they don't have to be technical, but they do have to be committed — the system ends up unused within 6 months.
The role can be the operations manager, the head of customer support, or whoever owns the process being automated. What matters is that this person understands what problem is being solved and cares that it works.
Question 4: How long does the process you want to automate take today?
A useful exercise: calculate the monthly cost, in human hours, of the process you want to automate.
If a process takes 2 hours a week of an analyst earning USD 800/month, you're talking about roughly USD 100/month in time cost. An AI implementation costing USD 15,000 would take 12 years to pay for itself.
The cases where AI has a clear ROI are processes that consume a high volume of time or that happen at scale: classifying 5,000 documents a month, answering 2,000 customer inquiries, reviewing 500 contracts.
If the process is small, first consider whether an automation without AI (simple rules, basic RPA) solves the problem for a fraction of the investment.
Question 5: Is leadership willing to tolerate an adjustment period?
AI implementations rarely work perfectly on day one. A classification model starts at 75% accuracy and improves to 92% after 3 months of feedback. A chatbot makes mistakes at first and improves with real cases.
If your organization expects everything to work perfectly at launch, the project will look like a failure even when it's working exactly as it should.
The honest question to ask leadership: are they willing to give 3-4 months of margin for the system to learn and adjust? If the answer is no, the timing isn't right.
Question 6: Are you clear on which business problem you want to solve?
"I want to use AI" is not a business case. "I want to reduce customer response time from 4 hours to 30 minutes" is. "I want to automate document classification so my legal team spends less time on administrative tasks" is too.
AI projects that fail almost always started from the technology ("we want to implement AI") instead of the problem ("we have this process that consumes time/money/errors and we want to solve it").
If you can clearly articulate which metric you want to improve and how much improving it is worth to your business, you're ready to have a real conversation about AI.
Summary: the maturity checklist
| Condition | Do you have it? | |---|---| | Historical data for the process (or LLMs that don't need it) | — | | Documented and standardized process | — | | Committed internal owner | — | | Enough volume to justify the investment | — | | Tolerance for the adjustment period | — | | Clear and measurable business problem | — |
If you checked four or more: you're in a good position to explore an implementation.
If you checked two or three: there is preparatory work to do, but you don't have to postpone it indefinitely. Sometimes that preparatory work (organizing the process, starting to record data) is part of what we build together.
If you checked one or none: the priority now isn't AI. It's building the foundations.
Get in touch if you'd like to go through this diagnostic in more detail for your specific case.