Welcome to the world of AI Excellence!

Generative AI can be used to enable the creation of efficient, consistent, and personalized content across a range of modalities.

Issue / Opportunity

Companies face a significant challenge in managing and optimizing marketing content. With hundreds of websites for brand portfolios, each in dozens of languages, companies struggle to allocate enough time and resources to create customer group-specific product descriptions, images, video, and even audio. Enterprises also wrestle with consistency across descriptions, imagery, ads, and other media, and the materials may not always be optimized for the necessary purposes (e.g., product descriptions for search versus e-mail). Companies need a method to provide a seamless and personalized brand experience across different ecosystems and touchpoints.

How Gen AI can help

Next-gen content generation
With Generative AI, the enterprise can create product descriptions, imagery, video, and more much faster and more consistently than with existing tools and processes. Personalization at scale. Generative AI models can draw from multimodal data (e.g., text, image, geospatial data) to create personalized and contextually relevant content. The model can be used to catalog content and adapt content and user flow based on language, region, and customer behavior trends.

Assisting compliance
Due to the consistency Generative AI enables across modes, languages, and contextual factors, the enterprise can enhance regulatory compliance for materials across different geographies, cultures, and topics.

Managing risk and promoting trust

Reliability
While tasked with producing superior marketing materials, Generative AI systems may invent inaccuracies, which will lead to poorer customer engagement and outcomes.

Fair and impartial
Biases in the data (e.g., due to incomplete datasets) could lead to unequal quality of content in the face of different geographical or cultural factors.

Generative AI in asset maintenance planning can improve equipment uptime, reduce maintenance costs, and enhance operational efficiency.

Issue / Opportunity

In mining and oil and gas operations, maintenance planning helps prevent premature equipment failure, costly repairs and replacements, and extends the life of an asset. Facing near- and long-term constraints and factors, maintenance plans and the subsequent downstream processes can be changed to align with production, in response to resource availability, or because of unexpected events. Making maintenance plan alterations, however, can be costly and labor intensive.

How Gen AI can help

Continuous improvement
Generative AI can be used to reconcile lessons learned from prior shutdowns, identify opportunities for maintenance alignment, provide planners with the information needed to challenge assumptions on maintenance alignment, and develop strategies to minimize the impact across the system.

Optimal maintenance scheduling
Generative AI helps optimize maintenance schedules by weighing operational factors (e.g., equipment use, production requirements, and maintenance costs), recommending the most efficient and cost-effective schedules, and analyzing equipment use and performance data to minimize downtime and maximize equipment availability.

Simulation and optimization
Generative AI can simulate maintenance scenarios and evaluate the impact of maintenance strategies on equipment performance, productivity, and operational efficiency. This helps reveal the most effective maintenance approaches and optimizes resource allocation for maintenance activities.

Managing risk and promoting trust

Robust and reliable
Generative AI applications for asset maintenance planning depend on the quality of the data. Data that is incorrect, incomplete, or is not representative of the current operational environment or maintenance practices can lead to a suboptimal and potentially inappropriate maintenance plans that may even be detrimental to asset health management and future maintenance planning activities.

Accountable
There is no machine substitute for a human asset maintenance planners' knowledge, experience, and expertise. Overreliance on AI-generated outputs without critical human review may lead to important contextual factors and valuable insights being overlooked.

Safe and secure
Generative AI models may struggle to account for the uncertainties inherent in asset maintenance planning, like unexpected equipment failures or changing production requirements. Suboptimal or unrealistic Generative AI recommendations due to overfitting can lead to inaccuracies or poor performance when applied to real-world maintenance scenarios. The degree of human intervention and oversight needed must be considered in the design phase of the solution. This is especially true in complex maintenance scenarios with interdependent systems or intricate operational constraints may also prevent Generative AI from providing accurate and feasible solutions.

Generate synthetic data for model training, anomaly detection, and identifying cyber and deception attacks.

Issue / Opportunity

Missing data is a significant challenge for FSI organizations. Datasets may be incomplete, data transfers may be restricted, and potential anomalies are underrepresented in the data. Using synthetic data can help overcome these challenges. In cloud transformation, data transferals may be delayed due to the risks associated or regulations around data governance, and using synthetic data first enables a smoother and more efficient transformation. Meanwhile, machine learning anomaly detection systems (such as for identifying fraud, waste, and abuse) are trained on data from previous events. Their rarity and the dearth of data around them can make anomalies harder to assess.

How Gen AI can help

Improve model training
Generative AI can be used to quickly create synthetic data to supplement machine learning model training data, which is then used to aid and accelerate digital and cloud transformations. In this way, Generative AI complements the enterprise's wider AI initiatives, fueling (rather than replacing) other AI deployments. Amplify anomaly event detection. The rarity of anomaly events can make it difficult to train machine learning systems to detect instances of fraud, waste, and abuse, but by creating synthetic data with Generative AI, ML systems have a larger suite of examples that lead to a greater capacity to find patterns and anomalies in the data.

Harden the organization's cyber posture
Just as synthetic data can be used to train models to identify fraud, adversarial synthetic data can be used to train models to detect and mitigate cybersecurity risks, as well as user deception of virtual assistants.

Managing risk and promoting trust

Fair and impartial
A significant risk when generating synthetic data is that historic biases can creep into the generated data, perpetuating those biases. This bias is not necessarily intentional, such as in the case of certain communities or socio-economic groups being underrepresented in the data because those groups have conducted fewer banking business in the past.

Reliability
Synthetic data created with Generative AI can be limited in its scope and scale, and it should not be presumed to be accurate or perfectly reflective of real-world data. An over-reliance on synthetic data may inject problems with data reliability, which can hamper the validity and usefulness of the outputs and model training.

Content creation can be facilitated and enhanced with Generative AI tools that minimize the need for manual editing and time-consuming content management.

Issue / Opportunity

Content creators and managers are faced with large volumes of data that require considerable time to generate, edit, and oversee. There are significant time and resource investments needed for video and image editing, and the volume of content creates challenges around data management and finding the right content at the right time. Amid this, content creators face tight deadlines requiring high levels of efficiency for content management and editing.

How Gen AI can help

Creative assistant tool
Generative AI can be used to create imagery and apply edits using descriptive commands. Conversational editing, text-to-template, text-to-image, and more allow users to expedite the editing phase of the content creation process.

Picture editorial
Producers can automate footage management with video-to-text Generative AI to evaluate and create tags for scenes and content. Text-to-video commands (e.g., “add more rain to this scene”) can be used to enhance and accelerate the editing process.

AI “reshoots”
Content creators can use scripts and 3D scans of actors to generate new content, alter footage to create more realistic special effects, and allow studios to make edits without the need for reshoots.

Managing risk and promoting trust

Responsible
Generative AI tools may be trained with large databases of media and content, some of which may be copyright protected. As a result, the model outputs may include aspects of a creator’s or studio’s work or style that are not attributed to them, which raises legal and civil risks for the organization.

Reliable
Noticeable changes in style and brand quality due to Generative AI content creation and editing may erode consumer trust in the brand and content.

Privacy
If bad actors access the underlying models or applications, it could contribute to the spread of fake content on behalf of the organization, leading to misinformation. Model owners should ensure strong privacy and access controls to mitigate this risk.

Reducing avoidable animal suffering with AI technology

AI4Animals: an intelligent surveillance system

AI4Animals is an innovative camera surveillance system to better monitor how animals are handled in slaughterhouses. The key objective is to better identify, address and avoid welfare and animal handling issues.

Situation

Over the last years, most major slaughterhouses in the Netherlands have implemented camera monitoring systems. Every day, this results in hundreds of hours of video footage. Although current camera systems can help identify animal handling issues, there are significant limitations. In practice, slaughterhouses review a random selection of the many hours of video footage generated every day. As a result, most video footage remains unseen.

To address these shortcomings, a unique group of organizations consisting of animal welfare organizations De Dierenbescherming and Eyes On Animals, meat producer Vion and Deloitte teamed up to jointly develop an innovative camera surveillance system.

Mission statement

To significantly reduce avoidable and unnecessary animal suffering through innovative technology and its effective adoption, in close collaboration with committed organizations and people.

Efficiently monitoring animal welfare in slaughterhouses

The newly developed video software uses artificial intelligence to monitor how animals are being handled. Video recording that potentially contain animal handling issues are automatically selected by the AI4Animals algorithm and presented through a dashboard to be reviewed and, if required, escalated. This enables slaughterhouse employees to assess the video images and take corrective actions in order to improve animal welfare.

The system has proven itself in slaughterhouses across western Europe where employees are watching the highest animal welfare risk events selected by the system on a daily basis. This allows slaughterhouses to take control of their animal welfare situation, either with a single slaughterhouse or across slaughterhouses.

How it works

AI4Animals consists of AI models and a dashboard that significantly improves camera surveillance by automatically detecting animal handling issues, in near real-time. It does so in 4 steps

1. AI4Animals uses streaming camera footage and detects animals, people, and objects and how they interact.
2. Business rules and tracking are applied to the detections in the frames to find potential handling issues.
3. The system then selects and aggregates all video fragments containing potential issues to be reviewed.
4. The results are reported in trend reports outlining deviations over time and per slaughterhouse.

Privacy
If bad actors access the underlying models or applications, it could contribute to the spread of fake content on behalf of the organization, leading to misinformation. Model owners should ensure strong privacy and access controls to mitigate this risk.

IoT and the benefits of smart manufacturing

Savola Foods leads the way

Savola Foods is keen to make a positive impact on society and health, in line with Saudi Arabia’s Vision for 2030. With that in mind, they reached out to us to introduce smart factory manufacturing solutions, incorporating exponential technologies like Internet of Things (IoT) and cloud into their operations to capture digital data and turn it into actionable insights.

Situation

Savola Foods is a leading innovator in the food industry in the Middle East, with innovations such as non-hydrogenated and immunity boosting edible oils, but also robotic automation and cutting edge digitisation of manufacturing processes. Facing pressure from a competitive market, the company was looking for ways to increase market share, efficiency, product quality, and customer satisfaction.

Taking a foundational approach to Industry 4.0, Savola Foods wanted to create a connected production facility, beginning with the implementation of a smart factory pilot. They approached us to help develop and implement digital solutions that would drive value to be scaled to their global production network.

The Deloitte difference

Our cross-functional team piloted digital solutions by baselining performance to identify process bottlenecks and critical business challenges that could be addressed by real-time asset monitoring and a digital maintenance collaboration platform.

• A tailor-made AWS IoT platform based on Deloitte’s Smart Factory Fabric accelerator was configured to capture real-time data from the PLC/SCADA of connected equipment and sensors that enabled the team to co-develop features and dashboards to both visualise operational KPIs and provide insights into the production process.
• Key process and machine parameters were captures, with business rules applied to generate alerts in case of abnormalities, enabling key stakeholders to address issues more efficiently and with more certainty
• Digital libraries of asset manuals, equipment drawings, and spare parts lists were developed to allow technicians ease of access to information, improving the repair time




The cloud provides the necessary computing power to process and visualise all the data that is being captured, allowing Savola to work faster and with more accuracy than before.