Generative AI is a powerful new technology that enables the creation of text, images, and various types of content. But how can businesses utilise these advanced tools to address their specific needs? This article will explore seven different examples of how to use generative AI for business.
What does AI stand for in business?
According to the recent Accenture study, 42% of companies want to make a large investment in ChatGPT this year. However, it is important to distinguish between hype and facts when evaluating products like ChatGPT and the models they rely on (GPT-3.5 and GPT-4, Stable Diffusion, DALL·E 2, and others).
Before delving into the details, it is crucial to make some acknowledgements. The significant advancements in Generative AI were made possible by large language models (LLMs) and foundation models.
These breakthroughs have deciphered language complexities and empowered machines to understand context, infer intent, and even exhibit independent creativity. Furthermore, these models can be fine-tuned to perform a wide range of tasks, which can help to streamline processes and improve efficiency.
However, they have limitations and may not cover all needs independently. So, integration with complementary tools is often necessary to perform all required tasks.
Here are some limitations of using AI for business:
1. Today, Generative AI for companies is not ideally suited for real-time tasks such as phone conversations, which require quick response times. Therefore, developing and training more lightweight and faster models that can be optimised for real-time performance is necessary.
2. GPT-4 and some other cloud-based models have a limited context window, posing challenges for few-shot learning. Few-shot learning refers to the ability of a model to learn from a small number of examples or tasks.
Various techniques can be employed to overcome this limitation and effectively navigate through large amounts of information. One common approach is to use intermediate steps such as indexing and embeddings. Alternatively, enterprises building their own models can employ the fine-tuning approach.
3. Most LLMs typically rely on the creators’ cloud infrastructure, which poses another significant issue. This limitation can restrict their applicability to enterprises that handle sensitive data.
However, Tovie AI offers a solution that addresses this concern by enabling the deployment of LLMs on-premise. The AI solution allows organisations to create offline smart bots within their infrastructure, ensuring data privacy and security.
As Generative AI solutions for business are relatively new, comprehensive real-world use cases within corporations are scarce, creating a fair and equal opportunity for everyone involved. While these models are still in the early stages of scaling, we are beginning to witness the initial applications across various business functions. Let’s look at this in more detail!
Banks can use ChatGPT to analyse customer reviews from various online platforms. It can identify patterns and trends in customer satisfaction levels and areas where the bank needs improvement.
For example, ChatGPT could be used to identify customers’ most common complaints about the bank’s website, mobile app, or customer service. This information will then help improve the bank’s products and services.
ChatGPT can also help evaluate customer service replies to reviews and recommend how to correct them. See the use case below.
|
Prompt for ChatGPT:
Below I will provide bank customer comments. Your tasks are:
1. Determine the tone (negative, neutral, positive).
2. In five words, express a summary of the review.
3. Answer whether there is anything critical for the bank in the comment (yes/no).
4. Provide recommendations to the bank’s customer support (if applicable, no more than ten words).
5. Suggest a possible response to the bank’s customer.
Comment text: …
|
|
Comment 1:
I spoke to two or three people at the bank who helped me through the questions and what to do next with my enquiries about car insurance renewal/prices. The phone lines are busy but able to get through quickly. The team I spoke to reassured me and helped throughout.
|
LLM response:
1. Sentiment: Positive
2. Summary: Helpful customer support for insurance renewal
3. Critical for the bank: No
4. Recommendations for customer support: Keep providing reassurance and assistance.
5. Possible response: “Thank you for your positive feedback! We’re glad our team could assist you with your insurance renewal. We appreciate your patience during the phone lines’ busy periods. If you have any further questions, please contact us anytime.”
|
|
Comment 2:
I believe it is important to be contacted immediately if someone has obtained my personal details without authorisation, rather than discovering it two weeks later when my card fails twice during a short vacation. Additionally, I would prefer to receive paper statements as I value the ability to review all my card transactions.
|
LLM response:
1. Sentiment: Negative
2. Summary: Delayed notification, preference for paper statements
3. Critical for the bank: Yes
4. Recommendations for customer support: Improve the immediate notification system, offer paper statements
5. Possible response: “We apologise for the delayed notification and any inconvenience caused. We understand the importance of timely communication regarding personal data security. We recommend enabling immediate alerts for suspicious activities. While we encourage paperless statements, we understand your preference and can assist you in receiving paper statements. Please get in touch with our customer support for further assistance.”
|
Next steps:
1. Automated parsing of websites with comments and reviews
2. Incorporating banking context, automatic classification, and auto-replies for positive, neutral, and non-critical reviews.
3. Automated identification of common topics and rating reviews based on those topics.
4. Automatic generation of daily/weekly summaries highlighting key issues and benefits and generating recommendations.
5. Assessment of the quality of responses from bank employees and providing recommendations for improvement.
6. Ensuring a consistent communication style across the bank.
ChatGPT can analyse transcribed conversations between clients and a contact centre to provide a summary and offer recommendations for improvement. It will extract relevant information, identify key points, and analyse the language and sentiment used in the conversations to understand the overall positive, negative, or neutral tone.
Based on this analysis, the AI model can provide recommendations to enhance customer satisfaction, improve communication strategies, or address any identified issues to enhance the overall contact centre performance.
|
Prompt for ChatGPT:
I will cite a conversation between a bank client and a call centre operator. Your tasks:
1. Highlight its tone (negative, neutral, positive).
2. In five words, express a summary of the review.
3. Indicate if the employee was rude.
4. Answer what problems the client had.
5. Write what recommendations should be conveyed to the bank operator (if applicable, no more than ten words).
|
|
Transcription fragment:
– … Did you carefully read the contract, Mr. Davies?
– What did I read?
– Did you read the contract carefully?
– Of course, carefully. And what?
– So, Mr. Davies, according to the contract provisions, the presence or absence of a cash loan from our bank should be fully settled by February 15th…
|
LLM response:
1. Tone: Neutral
2. Summary: Contract clarification regarding cash loan repayment
3. Employee rudeness: Indeterminate (insufficient context)
4. Client problems: Unclear from the given transcription fragment
5. Recommendations for a bank agent: Provide a clear explanation of loan repayment terms
|
Use LLMs to create a document FAQ, for example, for a paper describing bank card tariffs as a data source. The model parses and processes the document to extract key information and create a structured knowledge base. When a user poses a question, it consistently provides relevant information in a chat format after some clarifying questions.
|
Prompt for ChatGPT:
Provide the information: “What should be done so that the interest on the balance is more than 5%?”.
Note: The document does not explicitly reflect these conditions.
|
Result:
1. The bot independently compared the possible values of interest accrued on the balance in the document.
2. Defined values greater than 5.
3. Analysed the combined requirements necessary to obtain the desired percentage.
4. Provide the correct answer.
|
Case No 4. Compliance analysis of candidate CVs with company vacancies
AI language models can be used for compliance analysis of candidate CVs with company jobs and help streamline recruitment. ChatGPT can be efficient enough to analyse resumes, match candidate qualifications with job requirements, and provide valuable insights to recruiters, enabling them to make informed decisions and identify the most suitable candidates for specific roles within the organisation.
| Prompt for ChatGPT: The document’s first part indicates a legal adviser’s job description. Below are two CVs: Candidate 1 and Candidate 2. 1. Compare the job requirements and the Candidate’s 1 work experience 2. Compare the job requirements and the Candidate’s 2 work experience 3. Write which candidate best fits the job. 4. Recommend to the HR officer whether it is worth inviting candidates for an interview. |
LLM response:
1. Candidate 1 doesn’t meet the job requirements, as they don’t have at least three years of experience in law.
2. Candidate 2 meets the job requirements, as they have a degree in law and more than three years of experience in law.
3. Candidate 2 is a better match for the position, as they have the required education and work experience and knowledge of the various areas of law required for this position.
4. Inviting only Candidate 2 for an interview is recommended, as they meet the vacancy’s requirements, unlike Candidate 1.
|
ChatGPT is also great for crafting job descriptions. For optimal accuracy, make the bot conduct a short interview with the “customer” (the prospective manager) and review sample job descriptions from the company.
During the interview, the bot will ask about the anticipated salary range, key requirements and metrics, desired location (city/office or remote), and other pertinent details. The outcome is a complete job posting that can be shared on HR platforms.
|
Prompt for ChatGPT:
Your task is to prepare a job description for an employee needed by a bank’s HR department. To do this, you need to conduct a brief interview with the client—the future supervisor—and review examples of other job descriptions from the bank (I will provide them below).
During the interview, you should clarify the expected salary range, possible remote work, and the location (city, office) where the employee is required. You need to prepare the description for the job position: Data Scientist.
Ask questions and seek clarifications from the client in a dialogue format. It is preferable to ask no more than 6-7 questions. Ask one question at a time, and you can ask the following question after receiving an answer to the previous one, taking that answer into account (in each iteration, you will see all the previous questions, answers, and tasks).
Here are two existing job description examples from the bank:
Job 1: ………….
Job 2: ………….
|
Next steps:
1. Provide information on general working conditions, and ensure the final document is styled and formatted appropriately.
2. Share an interview link and contact details.
3. Gather requirements from various stakeholders involved.
4. Analyse and monitor CVs for compliance with requirements, automatically seeking clarification on missing data from prospective candidates.
5. Generate a shortlist of qualified candidates.
6. Conduct a preliminary interview with the candidate via chat.
Case No 6. Follow-up and meeting minutes
One of Generative AI’s most amazing capabilities is creating meeting follow-ups by summarising key discussion points, action items, and decisions made. Additionally, it can generate comprehensive meeting minutes by capturing essential details, including attendee names, agenda items, and any relevant attachments or links shared during the meeting.
Moreover, for the AI model to function effectively, it requires just the text of the meeting, transcribed through a speech recognition system. Even if some terms are recognised incorrectly, or there is no explicit division into roles within the text, the chat will attempt to comprehend the inaccurate words and form hypotheses about the participants. It will then generate a concise follow-up summarising the key discussions, resolutions, and assigned tasks.
Next steps:
1. Analyse the semantic structure of text segments and draw intermediate conclusions while preserving essential details when dealing with substantial amounts of data (e.g., meetings lasting over 30 minutes).
2. Conduct further training in Automatic Speech Recognition (ASR) on company-specific terms, create a dictionary for parsing hypotheses, and enable autocorrect functionality within the LLM.
3. Integrate standard meeting minute templates for generating well-structured documents.
4. Identify and clarify any unclear issues in the transcript.
5. In the long term, explore the possibility of extracting speaker voices directly from the audio stream.
To create images, you will need a tool like Midjourney or other similar. By providing Midjourney with a brief and some inspirational images, you can generate realistic illustrations and then use them to promote your products or services. Let’s see an example of a bank advertising campaign.
|
Prompt for Midjourney:
Objects in the illustration: a young man, a laptop, a modern and stylish apartment
The text for which the illustration is being created:
Become a happy owner of a new apartment with the Bank.
To apply for a mortgage, all you need to do is prove your income. If you are an employee of an IT company, then your rate will be only 2%. Young people in a big city should live in a beautiful, comfortable, modern place. Feel the pulse of life. Be at the centre of events.
|
Image generated by Midjourney with this prompt
Generative AI offers companies immense opportunities to simplify, personalise, and automate numerous business processes in our dynamic world. In today’s fast-paced environment, customers demand prompt assistance.
Leveraging advanced language models like ChatGPT, organisations can efficiently manage high volumes of customer queries and retrieve essential information in real-time, enabling seamless scalability.
Below is a comprehensive compilation of the primary use cases for Generative AI for business.
| Application scope | Scenarios |
| Recruiting and mass hiring | CV screening, job description creation, cold email outreach to candidates, and initial web chat interviews. |
| Press release preparation and communications | SEO text for websites |
| Text rewriting based on multiple sources |
| Writing professional emails |
| Gathering information through interviews |
| Design | Creating design references for illustrations |
Preparation of regulatory documents and instructions
| Document analysis and assistance in approvals |
| Creating documents based on templates |
| Transcribing meetings and writing minutes | Transcribing and analysing meetings |
| Unified knowledge base: answering questions | Question-answering system on large datasets |
| Customer feedback analysis | Customer feedback analysis and generating responses |
| Analysis of customer complaints and inquiries | Analysis of customer service recordings with call centre agents |
| Writing code | Preparing code snippets |
| Code testing | Identifying potential errors in the code |
| Prompter for client managers and assistant for clients | Supporting existing corporate and retail customers, answering inquiries |
What every CEO should know about Generative AI is that it is changing the business world right now. It is revolutionising industries, and its potential is still expanding. As technology rapidly advances, new use cases continue to emerge.
Nevertheless, integrating generative AI models like ChatGPT requires careful consideration and planning. If you need guidance or wish to develop a strategic approach to implementing AI within your organisation, please get in touch with us.
There has never been a better time to embark on your Generative AI journey and unlock its benefits for your company