Contents (7)
Today almost every software company says “AI”. Seeing an impressive demo is easy; the hard part is finding a solution that works correctly every day with your data and your customers. When choosing an AI company, look at evidence rather than presentations, and at written terms rather than promises.
#First, be clear about what you want
“We want to use AI” isn't a need; AI is a tool. Go into the meeting knowing which of these you want:
- An assistant that talks to customers: A chatbot that answers questions on your website or on WhatsApp.
- Document and recording analysis: Extracting information from meetings, call recordings, contracts or emails.
- Classification and routing: Sorting incoming requests by topic and passing them to the right person.
- An internal knowledge assistant: Letting employees search company documents and get answers.
#Evaluation criteria
| Criterion | What to look for | Warning sign |
|---|---|---|
| Live examples | A product or project used by real users; something you can try | Only a presentation and a demo video |
| Testing with your data | A small pilot with your real questions, with the results measured together | A promise that it “does everything” |
| Data and KVKK (Türkiye's personal data protection law) | Where the data is processed, which service it goes to, how long it is kept | Vague or verbal-only answers |
| Cost structure | Development, model usage fees and hosting listed separately | A single all-in price; usage costs never discussed |
| Ownership | Who owns the source code, the prompts and the data | Permanent dependence on the company |
| Maintenance | Model updates, error tracking and improvements | No support plan after delivery |
#An existing model or your own?
Most businesses don't need to train their own AI model from scratch; it is expensive and slow. The common and effective approach is to use existing large language models through an API and feed them with your own documents, price list or systems (RAG). If a company says “let's train a model just for you” in the first meeting, ask in detail why and what it will cost. If your data must not leave the company, we compare on-premises options in on-premises AI and data protection.
#10 questions to ask before getting a quote
- Can you show us an AI project of yours that is live and being used by real users right now?
- Can we run a small pilot with our data, and how will we measure success?
- Which AI service, and which country, will our data be sent to?
- How long will the data and conversation logs be kept, and how will they be deleted?
- What will the assistant do when it doesn't know the answer, and how will it hand over to a person?
- How much will the model usage (API) fees come to each month, and what makes them go up?
- Are development, hosting and maintenance listed separately in the quote?
- Will we own the source code, the prompts and the configuration?
- What happens if the model you use changes or is discontinued?
- After launch, who will monitor the quality of the answers, and how?
#A local firm or a large agency?
Large agencies offer big teams and a broad scope; with small, specialist teams you talk directly to the person running the project and decisions are made quickly. Which one suits you depends on the size of the project. For choosing a software company in general, the comparison table in how to choose a software company will also help.
#Our approach
At Averis Soft we develop and run our own AI product, Transify AI, and we build web and WhatsApp assistants and document and recording analysis solutions for businesses. On every project we make clear in writing from the start where the data goes, what usage will cost and who owns the source code. The details are on our AI solutions page; the initial consultation is free.
Thinking about an AI assistant for your business?
Let's work out which questions can be answered automatically and where the assistant should run.
The discovery call is free. Same-day reply on weekdays 09:00–18:00 (GMT+3), otherwise within 24 hours. · See the service
Frequently asked questions
What should I look for when choosing an AI company?
Look for live working examples, a pilot using your own data, where the data is processed and kept, a cost broken down into separate items, and ownership of the source code.
What does the cost of an AI project depend on?
The development scope, the systems to be connected, the usage fees of the model (per message or per operation) and hosting and maintenance costs determine the cost. Usage fees grow with the volume of operations.
Do we need to train our own AI model?
For most businesses, no. Existing large language models are used through an API and fed with the company's own documents and data (RAG). Training your own model is expensive and only needed in special cases.
How do AI projects comply with KVKK, Türkiye's personal data protection law?
You determine which service and which country the data is processed in, how long it is kept and who can access it, and you prepare a privacy notice. If the data must not leave the company, an on-premises set-up is preferred.
How long does an AI project take?
It depends on the scope; the exact timeline is set out in writing in the scope document prepared after the analysis. Starting with a narrowly scoped pilot reduces both time and risk.