These are some examples of data & AI use-cases that we've partnered on with customers.

One of the largest construction companies in The Netherlands wanted to create a single source of truth for their teams. Data was scattered among an different software tools the various teams use. There were big challenges in discrepancies between important metrics from the different tools, leading to discussions and uncertainty in decision making.

We started created a design for a reliable and well structured single source of truth. Based on the used tools like the ERP system, a CRM and a large selection of proposal and design files and drawing, a recommendation was made on the best suited, simple and easy to use technology for a data warehouse and supporting infrastructure. One of the key aspects here was to create one data stack that could be used seamlessly for decision making in Business Intelligence, but also to develop AI initiatives with LLMs.

Based on the design, a data pipeline was developed that collects daily data from all key tools, clean and structures this. The data warehouse was setup in Azure and structured in a way that it suits both BI and AI purposes. Based on a top-down analysis of business challenges, a set of data models was develop that creates daily actionable insights for the various teams to make decisions with, fully integrated in the existing BI.

A SaaS enterprise was experiencing market-driven pressure to adopt AI, as emerging competitors introduced innovative solutions and customers began inquiring about their AI roadmap.

We decided to keep the traditional document search product and build a stand-alone system based on conversational AI next to it. This is now capable of reasoning about all their internal documents. Instead of delivering hundreds of search results to users to select, review, and combine, the AI generates complete answers to questions based on the most relevant content available. 

We embraced a 'glass box' approach, offering professionals who use our software a clear view of its workings. Contrary to most Large Language Models which operate as 'black boxes'—providing answers without insight into their origins—our glass box model allows users to see exactly which documents and specific sections the AI utilized to generate responses. 

This transparency not only fosters trust in the system but also equips professionals with a convenient tool to delve deeper and refine the LLM's answers when necessary. After the initial trial, the feedback from professionals was that in 88% of the cases, the provided answer was on par with what they would have written themselves.

A small to medium-sized business with a large support ticket database wanted to enhance its support team's responsiveness and efficiency, aiming to improve the overall quality of responses.

We used their historical helpdesk questions and answers to create a co-pilot for their support team. This replaces manual answer generation, where the AI drafts an answer based on all historically similar questions. Based on several variables, it introduces a certainty indicator (high-medium-low). With this indicator, the support team decides on their level of review of the pre-generated answer before sending it to the customer.

Since the helpdesk was managed through a legacy system and helpdesk content is typically unstructured and filled with problematic data, such as out-of-office responses and one-line replies, the core focus of the project was data cleaning. This involved first identifying the various issues, then devising strategies to clean the data, and ultimately constructing a pipeline capable of handling the diverse steps required for data cleaning. Helpdesk content is also very often full of personally identifiable information, so we set up an anonymization pipeline that removes things like person names, organizations,  addresses, email, and other contact information.

A SaaS organization reached out to us because they wanted to optimize their sales process and also predict the optimal moment when leads are ready for direct outreach. They were pretty convinced that AI could help them become more efficient and therefore more profitable.

We discovered that their challenges weren't primarily AI-related. The lack of comprehensive user analytics and integrated sales signals from various platforms, such as email interactions, made AI deployment less effective. Our first step was foundational: we developed a pipeline to combine diverse data streams into a unified storage system. This included syncing data with their existing sales infrastructure and expanding data collection networks, like user analytics.

The first result was actionable signals for the sales team that predict which lead is ready for direct outreach at a given moment. These signals were implemented in the team’s CRM tool as tasks and used for 6 months to validate this data-driven way of working.

The team now is very convinced of this path forward, and we are now working with them to use AI to automate this data-driven sales workflow and help their teams be as efficient as possible in closing new leads.

A government institution wanted to test if AI could improve their workflows without a specific idea of which part would be best. The focus was really on delivering a better experience to the citizens and less on efficiency.

After some initial sessions, we decided to focus on augmenting document creation with AI, for applications of certain benefit packages. This creates a completely new workflow that previously was fully manual to combine sets of application documents into one central application format, submit the application, and communicate with the involved citizen.

We created an AI document creation copilot, that fetches all relevant data from application documents and combines this in an easy-to-review format for the government user to approve in a conversational interface. Once approved, the application is automatically generated and sent to the correct government system with an API integration. The application is also sent to the citizen by email, together with an explanation letter in simple, non-legal, terms. This letter can also translated into different languages.

Introducing this application (in beta) seems to lead to a significant reduction in the processing time of applications, but it is pending iterative testing and improvements, to ensure quality and consistency before this can go into production.

A manufacturing company is experiencing increased competition in the job market and as a result, has shorter tenures and higher employee turnover. Therefore they wanted to make their onboarding process more efficient.

We proposed to reduce the onboarding time of new employees with an onboarding duet. This is an AI system that has access to the company’s internal working instructions and knowledge bases and uses this data to dynamically generate onboarding guides, training new employees to be up to speed quicker. 

New employees are using a conversational AI interface that first walks the user to a learning session, and then creates dynamic QAs. It is using chain of thought reasoning to make concepts more clear and take the employee through the desired workflow step-by-step. If comprehension is not at the desired level the system can keep questioning the same ideas in different ways. We created a pipeline to clean and standardize all internal working instructions and knowledge bases, and developed the AI logic to transform this into guides and dynamic Q&A.