In this article, we explain the different approaches to AI implementation; from business case analysis to go-live. Every AI company has its own approach or methodology that addresses customer questions. The consequences of these differences are often difficult to estimate initially, but have a huge impact on the final ROI and even the chance of success of the entire project. Finally, we outline how Trendskout differs from other AI companies.
Let’s start with the biggest challenges facing successful AI applications, the customer and the AI company.
- AI solutions are only as effective as the quality of the questions asked. The customer often realizes that AI can play a crucial role in the optimization of a certain process, but the question is not yet clearly defined. For example, AI can indeed have an impact on the top line, but in which functional domain this must be done – sales forecasting, digital conversion, etc. – can only be determined by smart managers and leaders within the customer’s organization in consultation with the AI company and not by the AI itself, of course.
- Creating the right innovation culture within an organization is a step-by-step process. AI not only has an impact on IT-related services in an organization, but also on the way other services work. Mapping the various stakeholders and a good communication plan are crucial in this.
- Stating that data quality is an important condition for the technical success of an AI project is an understatement. Just like “data is never perfect”, this is also a cliché. At the beginning of each project, any problems with data, inconsistency or incompleteness must be properly identified. Subsequently, the impact of any problems on the course of the project must be evaluated together with the collaborating AI company.
Of course there are other challenges, but we will continue these in a later post.
The attentive reader will have noticed that the above challenges are common in almost every IT project. It mainly revolves around sound analysis and communication. Below we explain how the choice of an AI company, and their corresponding approach, determines the quality of the organization’s response to such challenges.
A Waterfall or an Iterative Process?
The approach of almost all AI companies can be seen as either a Waterfall or an Iterative Process. We explain briefly what this means, and then indicate the impact on the success of an AI application.
The Waterfall process has long been the standard in software development. In short, it is based on the assumption that the needs of a software project could be determined prior to implementation, after which the actual programming was completed in a straight line towards the finish. While the simplicity of the approach exudes a certain elegance, there are some important pitfalls that come with it.
Requirements or assumptions may change throughout the project due to new insights. By definition, these changes cannot be accommodated during the pre-implementation analysis. This scope creep often ensures that the programming work does not solve the real emergency, but a virtual emergency that only lives on in the analysis document.
Examples of this with AI applications are:
- additional data sources need to be added,
- the accuracy based on the first implementation is insufficient and another AI algorithm has to be used,
- the AI model has to control additional software or generate new information,
- the integration interface with external systems changes,
- drift lowers the accuracy of the model so that the original data transformations, algorithm selections and hyprtuning are no longer sufficient,
The second consequence stems directly from the first, and concerns either the impact on ROI or budget. If it is decided to still stick to the original analysis and approach, the added value or ROI of the project will be uncertain. After all, what is the value of a project that no longer represents reality, or where buy-in from stakeholders is missing. A second option is to partially restart the project with a new analysis, which of course derails the budget of the project and even more when additional changes are needed during the second run.
According to many, the answer to these risks is an Agile approach that starts with the assumption that requirements cannot be known in advance and that these should be elaborated during an iterative process in which development and analysis alternate, or take place in parallel. Scrum, Extreme programming, Pair programming … are just some of the technical terms that come under this umbrella.
Because of the realism embedded in this Agile method, scope creep is no longer a problem in itself, it is even almost a requirement that is necessary to determine what must be developed in different sprints (time blocks in which programming is done). Unfortunately, this approach is not without risk.
Often, changing requirements or simply not getting them right leads to delays and out-of-budgets due to a much longer development process.
Almost all AI companies are focused on a project approach for implementation and use one of the above methodologies. When choosing an AI company, it is therefore crucial to ask which of the two options are used during development.
How does Trendskout work?
Because Trendskout offers an AI software platform as a product and is not a classic project agency, our approach differs significantly. We review the differences in terms of budget and impact of changes.
- Our customers subscribe to our software and support, and know in advance what it costs. Our software supports a variety of ready-to-use AI, input and output options and an entire data transformation engine to improve data quality. If the needs of the customer do not match our platform, this will also be made clear during the first conversations. This guarantees the customer knows what the necessary budget will be.
- Requirements changes are also welcome in our methodology, and are an inherent part of our vision that a few small-scale AI experiments should be done before an AI application is ready for the real thing. Customers can do this themselves, without necessarily having to be AI engineers, and learn in a user-friendly way how an experiment can be improved into a production-ready model. Changes in requirements therefore have no impact on budget, but lead to a better understanding of the business case. Of course Trendskout offers support in this, but without the associated budget impact of a classic implementation process.
This iterative method with budget certainty also provides sufficient time to tackle the three challenges in AI projects (business case analysis, communication and data quality). After all, the success of the AI business case should not be overshadowed by delays or unforeseen budget increases.
If you also want to get more out of your data and your sales process, you can contact us here.