Artificial intelligence (AI) has the power to transform how organisations operate, but not every problem is suitable for an AI-driven solution. In the first blog of our three-part series, we explored how AI can automate repetitive tasks, support complex decision making and generate insights from unstructured data. If you missed it, click here to read it and catch up.

In Part 2, we explore three further key "problem lenses" to help determine whether AI can address your business challenges: The Lens of Process Automation & Optimisation, The Lens of Risks & Ethical Constraints and The Lens of Multilingual Translations.

By applying these lenses, you can gain a clearer understanding of how AI might streamline workflows, reduce bottlenecks, and handle ethical complexities in your business processes.

The Lens of Process Automation & Optimisation

What is this lens?  

With large language models (LLMs), you can save time and increase efficiency by automating parts of your operations that require reading, writing, understanding or summarising large amounts of information. AI can optimise these processes by reducing manual intervention, increasing speed, and maintaining compliance with complex rules or regulations. Unlike repetitive task automation, this lens looks at entire end-to-end workflows and processes.

Practical steps to apply this lens:

1. Map the current workflow: Identify workflows involving repetitive text processing or decision-making based on structured and unstructured information. Start by mapping out the process you want to evaluate, including all steps, stakeholders, and decision points. For example, when processing planning applications, steps might include receiving forms, verifying compliance, routing for approval, and issuing permits. Identify processes where people are spending too much time on repetitive text-based work and look for bottlenecks, or inconsistencies. 

2. Evaluate bottlenecks: Look for steps that rely heavily on manual effort, require repetitive reviews, or create significant backlogs. Focus on areas that are predictable and repetitive, but still require too much time. Look for predictable, rule-based tasks such as document verification, data entry, or compliance validation, as these are the easiest to automate. For example, a team handling customer service enquiries could benefit from an LLM that drafts responses to frequently asked questions, which staff can review and send.

3. Understand what success means: Once you’ve identified areas for improvement, ensure you’re clear on how you can identify what success looks like. Capture key metrics to demonstrate this before making the improvements. This baseline of data will enable you to perform a before-and-after comparison once you implement the changes.

4. Test on one step of the process: Start small by introducing LLMs to one part of the workflow rather than overhauling everything. Look for processes that have minimal customer impact or are internally focused. This helps minimise risks as you experiment with AI-driven automation. For example, use an LLM to automatically generate summaries of lengthy client reports, leaving your team to refine the summary instead of writing it from scratch.

5. Test on low-risk processes: Begin with processes that have minimal customer impact or are internally focused. This helps minimise risks as you experiment with AI-driven automation.

6. Measure the impact: Track improvements such as how much faster tasks are completed or how much time your team saves. The accuracy of the AI outputs should also be periodically reviewed along with user feedback to determine if it is performing effectively. As with any digital service, information is only beneficial if it’s used. Ensure the insights are fed back into the development process so they are used to continually improve the service. 

Examples:

• Planning applications: Local councils often face delays in processing planning applications. AI can automate initial reviews, checking for completeness and regulatory compliance and flagging applications that need further human inspection. This reduces bottlenecks and frees up staff to focus on more complex cases.

• Contract management: Government procurement teams could use AI to generate standardised contracts that meet legal requirements. AI systems can also track contract renewals and flag potential risks, reducing manual oversight.

By applying this lens, organisations can streamline their operations, improve service delivery, and enhance productivity. Automating processes doesn’t just save time; it reduces errors and improves consistency across workflows.

The Lens of Risks & Ethical Constraints

What is this lens?

LLMs can generate remarkable results, but they aren’t perfect and can make up convincing but untrue information, known as hallucinations. This lens encourages you to think about the risks associated with deploying these tools, particularly when their outputs could impact people’s lives or reputations. Concerns like bias, data privacy, and harmful or incorrect information must be addressed. Emphasis must be put on careful assessment of risks, transparency, and ethical considerations right from the start of an AI project.

Using this lens helps ensure that AI implementations align with organisational values, comply with regulations, and maintain public trust.

Practical steps to apply this lens:

1. Identify ethical risks: Evaluate whether the process involves sensitive decisions, such as those affecting individuals' welfare, financial outcomes, or legal rights. Consider the potential for bias in training data or AI outputs.

2. Maintain human oversight: Ensure humans remain accountable for decisions, especially in areas where AI outputs could significantly impact people's lives.

3. Evaluate privacy concerns: Ensure data used by AI is anonymized, securely stored, and compliant with privacy regulations like GDPR. Avoid sharing sensitive or personally identifiable information with third-party AI systems. Even when using LLMs deployed in your own public cloud environment, cloud providers can still log out prompts (the text you send to the LLM) for monitoring so you may need to explicitly opt out of such conditions in their terms of use.

4. Monitor for bias: Regularly audit AI outputs to identify patterns of bias. For example, does the system disproportionately flag certain groups or produce skewed recommendations?

5. Be transparent: Let users or customers know when they’re interacting with an AI tool. Transparency builds trust.

6. Define clear accountability: Establish who is responsible for oversight and decisions when AI is part of a workflow. This is especially critical for processes that could have significant direct impacts on people’s lives, such as eligibility assessments or hiring.

Examples:

• Hiring bias: An AI tool used to screen job applicants must be monitored to ensure it doesn’t favour certain demographics or perpetuate existing biases in recruitment. Regular audits and human oversight are crucial for maintaining fairness.

• Benefit assessments: AI might assist in evaluating applications for benefits or subsidies. Left unmonitored to make assessments, biases in LLM training data can lead to unfair judgements with catastrophic affects. Final decisions should involve human intervention to ensure nuanced cases are handled ethically and transparently.

By applying this lens, businesses can responsibly deploy AI while mitigating risks. Thoughtful governance ensures AI contributes to efficiency without compromising ethics or accountability.

The Lens of Multilingual Translations

What is this lens?

If your organisation communicates with a diverse audience or operates across regions, LLMs can help translate content into multiple languages, breaking down barriers and making your services more inclusive.

Practical steps to apply this lens:

1. Identify where language is a barrier: Look at areas where communication is hindered because of language differences. This might include customer support, internal communications, or public-facing materials. Consider if customers struggle to understand your service offerings because they’re only available in one language.

2. Decide what needs translation: Focus on high-impact areas like customer FAQs, key marketing materials, or onboarding documents. For example, translating your “Help Center” articles into the most common languages spoken by your users.

3. Integrate translation into existing processes: Use LLMs as part of your workflow to reduce delays. For example, a multilingual chatbot can instantly translate and respond to customer queries in their preferred language, improving the customer experience. 

4. Test for cultural appropriateness: Ensure translations align with the cultural norms of your audience to avoid misunderstandings. Particularly when integrating such a capability into a workflow, it is imperative that the responses are carefully monitored for accuracy and bias. Using a native speaker to review translated text can be helpful to ensure it resonates with the target audience.

Examples:

• Public services: Local governments can translate their websites into multiple languages to ensure all residents have access to important information.

• Global marketing: Companies can launch campaigns in multiple regions by quickly translating ads and social media posts into the local language.

By using LLMs for multilingual translations, you can improve accessibility, reach new audiences, and strengthen relationships with diverse communities.

Conclusion

By applying these lenses to your business, you can begin to identify opportunities to save time, improve accessibility, and enhance customer satisfaction - all while ensuring responsible use.

In the next and final post of this series, we’ll cover the lenses of thought exploration and knowledge management, helping you use LLMs to innovate and unlock creative solutions for your organisation.

Caution Your Blast Ltd's Librian AI Solution can revolutionise your customer service by providing fast, accurate, safe and ethical responses to user questions - it is proven to reduce in-bound written enquiries by 80%. Request a demo or get in touch here