What matters most when choosing an AI development company?
What matters most is that the provider protects your data, knows your technology and stands behind what it delivers after go-live. Then come product judgement, an architecture that can handle growth and a way of working that shows you value early, without asking for a large upfront investment.
These are the seven criteria we look at at Criterian, in order of importance, with the signal that would make us rule out a provider on each one.
1. Data security, compliance and confidentiality
An AI project handles personal data, company data and often its strategy. You need to know where the data lives, who has access, whether it is anonymised and whether anyone can use it to train third-party models. Access to and use of the data must be traceable and auditable.
At Criterian we build compliance into the project from the start. We have worked with banks, wealth managers, fintechs and investors, where it is a baseline requirement.
Red flag: the provider does not commit to your personal, corporate or strategic information not being used for other purposes or shared with companies you do not control.
2. Fit with your technology
The provider should adapt to what you already have. Ask which technologies they work with and how they fit with yours: languages, cloud, CRM and data systems.
We work mainly with React, React Native, Node.js, Python, Java, Salesforce, AWS and machine learning.
Red flag: no experience with your technology.
3. Maintenance, handover and written warranties
Before signing, find out how the system will be maintained, how knowledge will be handed over to your team and what happens if a defect appears months later.
At Criterian we give a 90-day warranty from go-live: defects in what we delivered, against what was agreed, are fixed at no cost. The warranty is separate from support and maintenance, which are agreed on their own terms.
Red flag: no development warranty for issues that appear in the following months.
4. Product judgement and measuring before and after
The provider has to understand the problem, the customer journey and the job your users are trying to get done, and challenge you when needed so that what gets built has business impact. And they have to measure: where things stand before starting and where they stand afterwards.
Red flags: they talk only about software and not about product, or they do not propose measuring the starting point.
5. Architecture designed to scale
A solution that works for ten users can break with a thousand. Ask how they design for growth: latency, load testing, what happens if usage grows a thousandfold and whether the architecture lets you replace a component, including the AI model, without rebuilding the rest.
Red flag: they cannot explain how the system will behave as usage grows.
6. Value early, without a large upfront investment
Asking for a large upfront investment in a waterfall build, with no interim deliveries, leaves the risk with the client. The alternative is small iterations that show value early and expand in modules.
That is how we work at Criterian. The senior professionals at our core have worked together for more than 8 years, with little friction between product, design, engineering and QA, which lets us iterate in short cycles. 98% of our clients stay with us.
Red flag: a very large initial outlay and a waterfall plan, with no value along the way.
7. A team that can join yours
Sometimes you do not need a whole project, but to reinforce your team, or your consultancy's, on a specific feature. Ask whether the provider can embed some of their people in your team and work with your processes. At Criterian we can set up squads that join and leave a project as it needs, agreed beforehand with the client.
The seven criteria in a table
| Criterion | What to ask | Red flag |
|---|---|---|
| Data and compliance | Where does my data live, who can access it and is it used for training? | No commitment to confidentiality |
| Technology | Which technologies do you work with and how do they fit with mine? | No experience with your technology |
| Warranty | What warranty do you give after go-live and how does handover work? | No development warranty |
| Product and measurement | How will you measure the before and after? | Software only, no product or measurement |
| Architecture | What happens if usage grows a thousandfold? | Cannot explain it |
| Way of working | What will I see in the first few weeks? | Large upfront investment and a waterfall plan |
| Team | Can you set up a squad that joins and leaves the project as it needs? | A fixed team that cannot be adjusted |
A real example
At TIFIN, a US fintech that uses AI to personalise investments, the Criterian team was embedded in the client's team. The challenge was to scale a multi-product platform with high availability and financial regulatory compliance. After migrating to microservices, API latency dropped from 320 ms to 112 ms, with average annual availability of 99.95%.
At APCOM, a multi-country technology distributor in Midis Group, the starting point was a manual forecast. With machine learning models, sales forecast accuracy improved by 30% compared with that forecast.
Common mistakes when choosing a provider
- Accepting a solution that is not designed to grow: no latency measurement, no load testing, no thought for what happens when usage grows a thousandfold and no way to replace a component without rebuilding the rest.
- Choosing a team made up only of data engineers or mathematicians, without senior software architecture experience. AI is a tool inside a well-designed backend and frontend.
- Letting judgement come from the model rather than from experience. A model tends to agree with you: what gets built that way can work in the demo and fail as usage grows.
Team culture
One criterion is harder to measure: how the provider treats people. At Criterian we look after our clients and our own team alike. Ask the provider how much turnover their team has and who will still be on your project a year from now.
Want to test your project against these criteria?
Tell us what you want to build and we will explain how we would approach it.
Let's talk