Artificial intelligence (AI) is transforming the way we connect with and access information. Justdial, a long-time leader in connecting businesses with consumers, has recognized this shift and is harnessing the power of generative AI to enhance its platform and provide a more intuitive and informative user experience.
In an exclusive interview with Business Today, Shwetank Dixit, Chief Growth Officer at Justdial, spoke at length about the company's AI strategy and outlined how the technology is being used to overcome user pain points, create richer content, and ultimately shape the future of online business directories and local search.
PD: Justdial has a long history of connecting businesses with consumers. Why was generative AI the next logical step in Justdial's evolution? What specific user pain points and market opportunities did you identify?
Shwetank Dikshit: Justdial's mission has always been to efficiently bridge the gap between businesses and consumers. We have a large database of over 148 million reviews and 43 million listings. This is one of the largest number of listings on a single platform in a single location, so effectively leveraging this vast amount of information became a priority. We have been a leader in local search since Web 1.0, and Generative AI was the next logical step to better connect users with businesses.
AI helps provide users with all the necessary information about a business in the simplest and most intuitive format. This includes summarizing reviews, guiding users to write effective reviews, generating concise business summaries, providing detailed B2B product information on B2B platform JD Mart, and more. It also ensures correct categorization of businesses and products. These features not only improve user experience but also increase brand value by providing the best information about a business available online.
PD: Can you elaborate on how Justdial uses AI to help users write reviews?
Shwetank Dikshit: At Justdial, we understand that writing detailed reviews is difficult for many users. Often, users tend to write short comments like “Amazing” or “Good service” which are positive but lack details. We also discussed with our users the pain points they face while writing reviews. Most users feel that they are unable to come up with words while writing and end up abandoning the page. This causes us to lose potential and genuinely interested reviewers.
Our AI assistant addresses this issue by providing users with attribute-based keywords that highlight different aspects of their experience, including service quality, amenities, and offerings. For example, when reviewing a restaurant, keywords might include “great ambiance,” “zero wait time,” and “budget-friendly.” Users select keywords along with questions related to their experience and use the magic wand feature in the review section to instantly generate a rich text review. This feature allows users to express their thoughts more comprehensively, making their reviews more informative. Additionally, users can edit these AI-generated reviews to add additional information, ensuring the final review is unbiased and personalized. Our goal is to make the review writing process easier and more effective, resulting in better, more useful reviews for everyone.
PD: Is it to encourage questions? Does it help improve the quality of the review (e.g. grammar, detail)?
Shwetank Dikshit: Yes, our AI system suggests relevant keywords and questions to users, helping them consider different aspects of their experience in their reviews. This results in more comprehensive and detailed reviews. With around 170 million unique users every quarter from different parts of India, their understanding of spelling and grammar varies. Our AI enables everyone to write high-quality, clear and helpful reviews, irrespective of their language skills. This makes the review writing process easy and effective, and maintains the authenticity of reviews, resulting in better and more helpful reviews for everyone.
PD: How have you addressed concerns about potential bias and manipulation in AI-generated content?
Shwetank Dikshit: We prioritize fairness and accuracy in our AI-generated content. We employ a variety of rapid engineering strategies to ensure that content is free of bias and duplication. All content on our platform undergoes rigorous testing before publication. Even after publication, content is regularly audited by human moderators for balance and accuracy. User feedback is also continuously monitored with our sales and customer support teams to ensure any concerns are addressed promptly. This comprehensive approach ensures that our reviews and summaries remain reliable and unbiased, providing users with information they can trust.
PD: Since implementing the AI-driven reviews feature, what impact have you seen on user engagement? Have you seen a quantifiable increase in reviews posted, time spent on listings, etc.?
Shwetank Dikshit: Since implementing the AI-driven reviews feature, we've seen a significant increase in user engagement. Review submissions have increased by 37%, with notable increases in categories where users had rarely written reviews before, indicating that AI assistance is making it easier and more engaging for users to leave detailed feedback. AI-assisted efficiency has reduced the average time spent on the review creation page, but increased the number of views and time spent on pages displaying summary and detailed reviews.
PD: The concise explanations generated by AI seem promising. Can you explain the technical process? What data sources does the AI use to create these summaries?
Shwetank Dikshit: We have conceptualized summarizing business offers in under 20 words, providing users with a quick snapshot created using Generative AI. We leverage internal data such as category, highlights, location, years established, and other relevant business information. Our primary tool is GPT 3.5, which is ideal for our use case. We have developed prompting techniques specific to different industry verticals, ensuring that the information is relevant and accurate for each sector. For example, hotels ensure that amenities and perks are highlighted.
PD: How can you ensure accuracy and avoid overly general or potentially misleading statements?
Shwetank Dikshit: We defined parameters for explanation generation and a minimum number of business details to avoid generic output. We also ensure that results are relevant and dissimilar across industries to avoid generic output across industries. For example, a restaurant and a hotel have very different structures.
We ensure accuracy by using verified business information collected over 20 years. Regular audits by human moderators catch inaccuracies and continually update descriptions with the latest data to keep them relevant and accurate.
PD: What specific strategies and technologies are you employing to minimise Justdial’s environmental impact whilst leveraging AI?
Shwetank Dikshit: We do not use on-premise AI infrastructure, but leverage energy-efficient cloud infrastructure with Azure and Google Cloud. Wherever possible, we use small language models (SLMs) such as BERT (Bidirectional Encoder Representations from Transformers) to avoid the increased carbon footprint associated with large language models (LLMs). While LLMs such as GPT 4 are useful for complex content generation, open-source SLMs are highly effective and efficient for data classification, segmentation, and text entity recognition. These SLMs are easier to fine-tune for our data and improve the accuracy of many of our internal data operations, allowing us to remain efficient while reducing our environmental impact.
PD: Are there any trade-offs that need to be considered regarding AI model complexity and data processing?
Shwetank Dikshit: Yes, we have carefully considered the range of complexity of our AI processes. Our models range from the simplest linear regression to advanced LLMs. LLMs like GPT are powerful but can be overkill for simple tasks and come with high computational and monetary costs.
For many tasks, we use SLMs such as Phi or BERT, which are more efficient for many internal tasks. Additionally, we leverage distilled versions of larger models. These distilled models maintain accuracy while reducing the model size and computational requirements. They are trained with fewer data parameters, but trained on large amounts of data, making them suitable for most tasks.
We also use open source models for AI image generation. The open source community has evolved significantly over the past two years, providing curated and efficient models on a daily basis that meet most of our needs. This approach allows us to effectively balance performance, cost, and computational efficiency.
PD: How do you think generative AI will shape the future of online business directories and local search?
Shwetank Dikshit: Generative AI has great potential to enhance online business directories and local search. However, we believe the fundamentals of local search remain the same: AI makes business search easier by providing users with more detailed and rich content. AI makes business search easier for both business owners and users by giving them the tools to express themselves more clearly from their own perspective. AI makes business search easier for both business owners and users by providing them the tools to express themselves more clearly from their own perspective. …
Clearly, India is evolving rapidly in adopting various AI technologies. There is huge potential in Tier 2 and Tier 3 cities, which have seen a massive increase in web usage over the last decade. Leveraging AI in these areas will provide the most granular data on listings, helping to bridge the geographic gap and improve how users search for businesses.
We also expect to see a rise in AI-based solutions for Indic languages, allowing SMEs to create online listings in their native language, which will onboard a large number of SMEs and increase their online presence and competitiveness.
Additionally, the rise of voice-based AI allows businesses to become more conversational and overcome the inertia sometimes seen in text-based interactions. Voice-based AI empowers users to add details about their business, products, and services in their own native language, giving them more freedom online and giving them a competitive advantage.
As more data comes online, users are able to make decisions faster. Increased web usage leads to more online reviews and business details, ultimately improving the quality and transparency of inquiries to businesses. Businesses can also use AI insights from data in easy-to-understand language to identify seasonal trends to plan marketing and inventory. This leads to more strategic decisions and more efficient operations.
We are committed to maintaining our leadership in local search technology while leveraging AI to deliver the best user and business experience. The future holds exciting possibilities, and we aim to make online business directories and local search more dynamic, personalized and inclusive.
