GenAI Tools and Risk Management

From Hallucinations to Copyright: GenAI Risks Under Control

GenAI is amazingly powerful, but the technology is not perfect. High-profile cases are constantly being reported in the media where chatbots have made mistakes or caused problems. Recently, for example, the AI-based support chatbot of a global logistics service provider suddenly became foul-mouthed and insulted its own company. The user had to deliberately manipulate it to say these things by making targeted entries, so this was not a standard situation. And in B2B marketing, such direct customer interactions with chatbots are the exception rather than the rule.

Nevertheless, this is a fundamental problem for all GenAI use cases that must be taken into account. Otherwise, you run the risk of reputational or financial damage, legal sanctions and regulatory penalties. We therefore recommend comprehensive risk management for GenAI across three dimensions.

Avoid Hallucinations and Logic Errors

GenAI does not generate answers on the basis of real understanding, but on the basis of probabilities within the framework of large language models (LLM). That is why it is possible for a chatbot to freely invent "facts" and present them in a very confident and believable way ("hallucinations"). What is also astonishing for a computer-based technology is that these systems can sometimes have considerable difficulties with logic, and even with mere counting. Answers and results must therefore always be checked by a human. Appropriate processes have to be set up for this.

Secure Data Protection

Strict regulatory provisions such as GDPR apply to the collection and processing of personal data. However, many GenAI offerings do not fulfil these regulations. In principle, GenAI acts as a "black box" whose behaviour cannot be predicted. For example, it is possible that protected data processed by GenAI could be included in the model's training and then be put out again in response to requests from third parties. In some cases, such behaviour can also be deliberately induced (via prompt injection). It is therefore best not to enter protected data into GenAI technology at all. Alternatively, you can look to special GenAI chatbots that have guaranteed GDPR compliance.

Protect Intellectual Property

A similar issue arises in relation to intellectual property and copyright. If users enter trade secrets or copyrighted content into a chatbot, this could also be incorporated into the model's training and thus potentially be accessible to third parties. For this reason, sensitive content should be handled very consciously and carefully. One solution is to offer services where the user input is not used to train the language model. However, users have no way of verifying whether these safeguards are being adhered to. Intellectual property of critical importance should therefore not be processed by external GenAI services.

Which Chatbot for Which Purpose?

GenAI comes with a whole range of risks. However, marketing teams should not be put off by this. The good news is that there is also a whole range of offerings on the market that can be used to implement individual use cases securely and compliantly.

For example, the well-known ChatGPT offering from OpenAI is available in several versions, some of which do not utilise user input for training. There are also offerings that are specifically geared towards the needs of marketing teams. Or there is the option – albeit a rather demanding one – of running your own language model, which makes it easier to manage risks.

Marketing managers can derive the criteria for their ideal tool from their designated GenAI strategy and risk management requirements. Here is an overview of the basic options:

ChatGPT

The chatbot from OpenAI exists in several versions (as of January 2024).

• The free version uses the less powerful GPT-3.5 model, which is based on outdated information and is only recommended for private users.
• ChatGPT Plus brings GPT-4 and GPT-4o, plugins, custom GPTs, image analysis, the graphics GenAI Dall-E and other useful functions for companies. Particularly valuable: for some time now, GPT-4 and GPT-4o have also been able to access the current internet independently.
• ChatGPT Team adds team collaboration capabilities to Plus and offers higher limits for the volume of requests. Above all, Team offers the promise of not using your own input to train the model, which is particularly important for companies in terms of data protection. However, sensitive data should still be handled extremely carefully and restrictively.
• ChatGPT Enterprise, on the other hand, includes unlimited access to GPT-4 and GPT-4o, priority support and many other useful features, especially for larger organisations. The Enterprise version also has additional security features. Unlike Plus and Team, Enterprise access is priced individually.

Integrated GenAI

The major advantage of offerings such as Bard or Gemini (Google) and Bing or Copilot (Microsoft) is their seamless workflow integration into popular office suites. They also offer up-to-date web access for internet searches, in some cases without additional plug-ins.

Jasper

Jasper offers a GenAI copilot that is specifically designed for marketing teams. The tool is suitable for creative content creation, data analysis and insights. Other capabilities include project management and communication support as well as the management of company knowledge. Predefined templates make work easier and reduce the need for prompt engineering.

Claude Pro

Based on the Claude.AI offering from Anthropic. The chatbot offers advantages in marketing integration thanks to real-time analytics and seamless CRM embedding. However, the application is limited to certain regions. This includes the UK, but not Germany and other EU countries.

Operate your own model

This option, mentioned at the beginning, is very interesting, especially from a data protection perspective. However, it’s also complex and requires a lot of expertise.

Holistic AI Strategy: Selecting the Right Use Cases

With a holistic GenAI strategy, risk management concept and appropriately selected tools, marketing managers have fulfilled the most important requirements for successful GenAI implementation. Now they can get started and roll out GenAI in their teams. Of course, this is where the real work begins – the development and application of use cases that boost productivity, take pressure off the team and increase the effectiveness of B2B campaigns. We know from our own experience at andweekly that the possibilities of GenAI in B2B marketing are as diverse as human creativity itself.

To get you started, we have created a comprehensive GenAI guide that describes over 20 attractive use cases in all areas of B2B marketing – from strategy optimisation, content and campaigns to technology, automation and CRM. All use cases are strategically categorised in terms of the expertise required, the effort involved and the potential productivity gains. Our guide also provides further practical details, such as tips on the different strategies for GenAI input (prompts).