AI implementation in a company starts with one business process. It does not start with choosing a model or a tool. A well-run project has five stages: choosing the process, a data audit, a pilot, production rollout and maintenance. I am writing this for the board, the COO and the IT manager, the people who are looking for a vendor and want to judge whether that vendor’s approach makes sense. Below you will find the stages, the questions to ask a vendor and a first step you can take right away.
Table of contents
Where to start with AI implementation in a company: pick one process
First, point to one repetitive process with a problem you can measure. That could be handling time, the number of errors or a ticket queue that grows faster than the team can clear it. A good candidate meets four conditions:
- it happens often and follows a similar pattern,
- it relies on data the company already has,
- it has a clear owner on your side,
- it lets you measure the result before and after the change.
In practice, three areas come up most often: answering customer enquiries, searching internal documents, and classifying and processing documents. And “AI for everything” projects? Steer clear. They have no owner and no measure of success, so nobody can say afterwards whether they worked. This is where AI consulting for businesses helps, because it narrows a general idea down to one specific use case.
Process and data audit: what to check before the first line of code?
An AI process audit answers a simple question: does the company have the data and system access without which the tool simply will not work? Among other things, it checks:
- where the data lives: CRM, ERP, document repositories, email,
- the quality of that data, meaning how complete, current and consistent it is,
- permissions: who has access to it today and on what basis,
- legal constraints, such as personal data and trade secrets.
Then there is the law. GDPR (RODO in Poland) and the obligations under the AI Act decide what information may be processed and where. The audit ends with a “yes” or “no” decision for the chosen process, a list of required integrations and a description of the risks. Money is a separate topic, so we covered it in an article on the costs and risks of implementation.
AI pilot project: how to test the tool on a small scale
A pilot is a stripped-down version of the tool for one team, with a success criterion set in advance. Not after the fact. Before you start, you need to define the scope, the user group, the duration and how answer quality will be assessed. Then the team works with it and reports feedback, and the vendor improves the prompts (the instructions for the model), the knowledge sources and the integrations. At the end, one of three decisions is made: scale up, refine or stop. Stopping is fine too. It is better to do it after a pilot than to sink the budget into a solution that did not hold up in day-to-day work.
Production rollout and integration with company systems
In production, three things happen: the tool is connected to the company’s systems, permissions are assigned and the team is trained. Typical integrations are CRM, document workflow systems, messaging apps and the website. Security? There must be clear rules here: who sees which data and how the actions of users and of the tool itself are logged. Here is how it looks for us: Ragable answers based on company documents, and Botino works as a chatbot for customer conversations. We follow the same pattern when we build custom AI applications for businesses. Data and permissions first, the interface last.
Maintaining an AI tool after rollout: what happens after launch?
Launch is not the finish line. An AI tool needs ongoing quality monitoring, regular knowledge updates and adjustments as models change. Maintenance is mainly about:
- updating the data sources the tool uses,
- reviewing wrong or incomplete answers,
- moving to new model versions and testing after every change,
- keeping usage costs under control.
Risk works the same way: it is an ongoing process, not a one-off check before launch. As a reference point, I recommend the AI Risk Management Framework from the US institute NIST. It is voluntary and covers the design, use and evaluation of AI systems. Oh, and someone on the company side has to be responsible for the tool. Usually that is the process owner, working together with IT.
How to choose an AI implementation vendor: questions to ask
A good vendor starts with questions about the process and the data. Not with slides showing a ready-made tool. Before you sign a contract, ask:
- Do they propose an audit and a pilot before full rollout?
- Where will the company’s data be processed?
- How will answer quality be measured?
- Who maintains the tool and on what terms?
- Will the company keep access to its data and configuration if it changes vendors?
- What will integration with existing systems look like?
Red flags? Promises made before the audit, no pilot, no maintenance plan. If an offer skips these elements, all the risk lands on the buyer. In other words, on you.
A successful AI implementation in a company starts with one process and a thorough audit, not with buying licences. Web Systems is a software house from Łódź that has been operating since 2006 and builds AI applications and custom software. Want to find out which process in your company is right for a first project? Book a free consultation with our team. If you would like to start with a conversation about one process, see how we approach AI application development for businesses, and book a free consultation.
Frequently asked questions
How long does AI implementation in a company take?
It depends on the process, the state of the data and the number of integrations. A project built on well-organised documents runs differently from one that has to connect several systems. A pilot lets you estimate the scope of work before the company commits to a full project.
What is an AI process audit and do you need one?
It is a review of processes, data and legal constraints before the tool is built. It shows whether a process is suitable for automation and which integrations it will need. In short, it reduces the risk of a misguided investment.
Does AI consulting for businesses commit you to an implementation?
No. The consultation is for assessing the idea and pointing to the process worth starting with. The company decides on next steps, such as an audit or a pilot, only afterwards.
