Video: AI Agents in Action: Unlock Smarter Selling | Duration: 2404s | Summary: AI Agents in Action: Unlock Smarter Selling | Chapters: Welcome and Introduction (0s), AI-Driven Sales Trends (118.24s), AI for Sales (286.375s), Scaling Gen AI (582.84s), Dynamic Selling Assistant (842.13s), Dataiku's Dynamic Selling Assistant (1212.965s), Solution Implementation Details (1807.32s), Q&A Session (1923.18s), Multimodal Data Integration (2027.115s), Personalization and Quality (2163.74s), Conclusion and Thanks (2372.34s)
Transcript for "AI Agents in Action: Unlock Smarter Selling":
Hello, and welcome to today's discussion on agents and unlocking smarter selling. Good to have all of you here with us today. We'll be yes. So we'll be going through, key trends impacting the retail CPG and broader landscape with a particular focus on how these trends and topics linked to to sales teams and sales performance before we move into a live demonstration of the dynamic selling, agent. So I'm Carla Winston. I'm a solutions consultant supporting the retail and CPG sector here at Dataiku, and I'm joined by my colleague, Quentin Quentin. Yeah. I'm, I'm working as a social data scientist in, written in CPG for data. Alright. Thank you. So let's get, started. So as as a further intro into who we are, we have had the opportunity to support over 80 retailers and CPGs in their journeys, helping advance them along the the AI maturity scale, with built in flexibility and capabilities, that accelerate their their path to business results through industry relevant, use cases. This has meant the realization and implementation of of use cases spanning various domains, and with our focus today on sales teams, it's important to note that activities around selling more and selling better are transversely important. And this has has shown up in the number of sales focus use cases that companies are prioritizing. And we can expect the number of of sales focus use cases to remain significant, as today's trends, intersect with GNAI and agentic, applications. There are, a few primary trends that stand out, that touch really every industry and are bat in importance by the different types of companies that are tackling the challenges that go hand in hand with these trends. Genai, remains at the forefront of conversations, especially as it relates to is integration into powering everyday insights, and the support it can provide in helping teams respond to, an evolving landscape, behaviors and make adjustments to workflows in response to things like global trends, economic factors, and, changing customer expectations by providing access to your sales performance information and getting you the intelligent insights you need for for actioning, and to make those needed adjustments possible with with greater ease. So we can sum up these trends into to three areas, with one being that we are seeing industry players narrow in on smarter operations by leveraging AI to automate supply chain networks in order to to drive efficiency and ultimately reduce, calls, directly impacting, bottom line profitability. And top line growth also remains, of course, a a key priority for companies. Companies are driving their top line with with strategies that include the creation of hyper personalized experiences that differentiate them from a crowded marketplace and accelerate their sales revenue, coupled with building in agility to adjust to the customer's response to sales and marketing efforts by enabling their teams to unlock customer insights on the fly, and take the right actions to enhance customer engagement. And then lastly, we're seeing companies mitigate and eliminate impediments to their decision making through projects like AI generated summaries and trend analysis to, and in order to really reduce latency in the decision making process by providing teams access to synthesized information, which further enables the business Thio react quickly, to those market changes and work through, any bottlenecks. So what's next and what does this kind of mean for for sales? As we focus a little bit more in on GenAI, agenda use cases for the sales domain. I want to first kind of re ground us in a Gentic, which is not only a popular subject, but now a very tangible reality across industries. When we talk about agents, we're referring to an LLM powered system designed to achieve objectives through, multiple steps, with the idea of going beyond single retrieval. So, in other words, going beyond, more bare bones chat box to agents that plan and execute a number of steps across different business systems, and tools. So when we look ahead with this context, it's easy to imagine, AgenTek AI playing a key role in use cases tied to several business areas. And more than that, GenAI and agents are already powering insights in various areas for sales teams. But we'll focus, in on three. One being, the sales planning process, two, driving conversion, and and three, capturing loyalty. And these three areas, in in mind, agents can assist, pulling those critical growth driving levers for sales teams. So we can first take sales planning and augmenting the sales planning process through a competitive signaling agent. So what does this look like practically speaking? A competitive signaling agent can be designed to, monitor digital signals like social media post, announcements of new initiatives or product segments, etcetera. Another test can can be performed is, flagging customer feedback topics and reporting customer sentiment. Think, unstructured text and making sense of it through natural language. Then we could also, imagine the agent, automating outreach and personalizing messages based on the different competitive signals it has detected, after which kind of drafting a message, highlighting your product features and special offers, and to engage your customers before losing them to the competition. Then the agent could, perform administrative tasks like routing items and email drafts to the right approvers before any content reaches, your customer. And also coordinating emails and scheduling those personalized messages for more timely engagements, across your your different platforms. The competitive signaling agent is is one use case that sales teams can embrace with agentic AI, while another real way that we'll highlight is around, driving conversion through automated AB testing. We can apply agenda to AB testing to supercharge sales by, first automating variation generation and their deployment, with the creation of numerous email subject lines or message tones, for instance, and and testing them simultaneously to identify the preferred option. Secondly, I would say the the agent can analyze customer data like intent to generate tailored recommendations and content for customers, while another task can involve, learning and adapting to, real reallocate efforts to more high potential segments. And we can also imagine this agent, forecasting the, the right variations that will work best for your your business. So these two agentic examples, and the dynamic selling example you'll see live, momentarily, demonstrate that the latest trends with hyper personalization really standing out across all use cases, as well as demonstrating the the access to the right tools are critical to setting teams up for success and building business ready agentic, use cases, which leads us, into the the building blocks. Right? And and foundation needed, for this. It's really, cemented that, GNAI is is here to to stay. And what the past year or so has taught us with these technologies is that, they can't be ignored. They're already being used in your industry and probably even by people in your own organizations. And what that tells us is that you can leverage Gen AI and have valuable, generative AI use cases going going without having a plethora of machine learning in place or following a predetermined and flexible path in order to extract value with, generative AI. In fact, there are sales teams getting value today from large language models and accessing foundational models and analytics that help them fuel their success by taking advantage of existing projects and pre built use cases that we can offer, with the the ease of of building this all out in in one platform, which is doable in in Dataiku. A path that sales teams can naturally take is dynamic selling, which at its core leverages agent to to orchestrate the automation of pricing insights, customer personalization, product and location information to both, reactively and proactively make customer centric, sales decisions, without having to to wait to get, data pipelines perfect. Teams can effectively bring in a combined, unstructured customer data and leverage clustering models in conjunction with data mining techniques and rule based analysis, and build on these elements and connect them with the aim of having an agentic solution to support the sales persona with unified information and guided recommendations through an interactive, easy to use interface, helping bring that that speed to insight when it comes to those common everyday, sales responsibilities. So how does this come to life, in in Dataiku? What we we have learned, is that executing one or two, generative AI POCs or projects is relatively straightforward, but where it gets a little murky for organizations is scaling. That is scaling the number of end users consuming, an AI application or scaling the number of use cases delivered across the full spectrum of the business and scaling a number of complex actions, from a an AI system that can reliably perform both a single action and multiple actions as well as various types of actions completing those tasks, autonomously. So overcoming these obstacles is critical to moving, gen AI projects into production and scaling both the delivery and adoption of these projects by equipping both your data experts and your business subject matter experts. So both your your coders, no code users and, in between, with a shared collaborative space that offers teams visual and coding capabilities while simplifying tool management and easing reuse across different projects, across use cases and across, different teams. So our Gen a I and agenda capabilities are a part of the Dataiku Universal AI platform. And this means you can cleanly connect your your agents to your data, your analytics, and machine learning where appropriate. And we understand that agents are really truly valuable when they can be integrated with your data and analytics, and different models, which means that they must be governed, and controlled. And Dataiku enables this kind of connectivity and the creation of agents at scale by providing the right controls within your your organization and in a very collaborative, way. So we can get a little bit more into, Gen AI and Agintic AI and their relationship, to, the to sales processes. So the the Dataiku platform is broad and collaborative in nature, that enables teams to to build and structure, their workflows in a transparent way across use cases, fueled by their data, their tools at NASA to where, this data may live. So what this means for teams and sales teams, no matter the, the industry we're referencing, is the ability to automate, to augment, and to optimize its scale. And in a way that can be easily integrated into various use cases, directly, affecting your top line and bottom line. And while also driving efficiencies and more impactful work because the individual is envelope, making sense of the the output and turning that into something meaningful, to impact the business. The AI agent in the the picture takes on that manual heavy lifting and the more rote task, freeing up these sales persona to focus on the bit picture and actions that drive performance derived from from relevant, use cases. When we talk about, use cases, there is inherent connectivity benefit to employing complimentary use cases in a single platform, to tangibly realize this results in a transparent, reusable, and accelerated manner. What this boils down to is accelerating sales processes and analysis that will support a sales analyst or demand planner, for instance, with, generative AI, coming alongside that the individual to efficiently and effectively deliver smart suggestions that feed into, and inform internal go to market strategies, trade plans or a localized product nets, which directly ties back to what your customers see, in the market and how they interact with your brand. The actions your teams take, with a customer centric approach takes us back to those critical sales levers that impact your your margin, your sales, and how you drive conversion and ultimately build, loyalty. The focus of this discussion is selling and more specifically the dynamic selling assistant prebuilt solution with a Gentic, which will be demonstrated again in detail, but sits at the center of these sales focus use cases, which our platform allows you to to power and automate now, and interact with the insights, underscoring the collaborative capabilities of the the platform. So this can be brought to life through unifying your data, using GenAI on top of it and side by side with the sales analysts or the demand planner to vet the insights and then using agents to orchestrate and automate that process. Sales teams find success in different areas, but the most common being, three main buckets that, that are familiar to any sales team, which are pricing, product and promotion. We we see, teams bring it in their sales platforms, first party data, customer tools, analysis, with with or without a machine learning layer, or manual forecasts and projections that may live in spreadsheets or existing in house models or models in external systems that inform their sales planning, their customer engagement, and how they go to market. They're bringing in these elements that underpin their processes, and enhance those processes, again, using generative AI to power the insight and agentic AI to automate how they dynamically sell better, in the way that is illustrated on this page and across other sales related topics that, may not be captured here. So a key element, to unlocking dynamic selling and an important item to note is the power that comes from combining your sales use cases and combining them in a way, again, where they're no longer stand alone, but complementary building blocks in a single platform that facilitates, collaborative components like the use of chat bots to easily interact with and consume insights as well as dashboarding to quickly review and monitor, sales performance and integrating GenAI and Aginta to provide those additional layers for automating the extraction of the insights from your interconnected, sales projects and use cases. So now to kind of continue right that our discussion around dynamic selling and how teams can, right now, elevate their processes and move the needle, on on sales results by using large language models, both in the the setup of the project and consumption of the data to uncover those latent insights at the the click of a button. Leveraging your pricing, promotion and product information to drive those critical sales levers. I'll turn the floor over to, Quentin, to walk us through how this can be realized right now, in Dataiku. Quentin. Thanks, Carla. So first, let me show you how Dataiku as the universal, data platform, AI platform integrated sent into the LMM landscape landscape. So first, everything is built around the LLM mesh, which is the way to interact with every provider and, LLMs that, is possible. So that's a great way to allow your teams to switch easily between, LLMs. So it can be ever LLM hosted on your own instance on premise or things that are running through API. And we all know that's, landscape is very fast changing. And so that's a great tool to be able to switch really from one model to another without having to share a lot of sensible, information on private keys to everyone. On top of that, we have a control layer, which is basically to ensure full compliance of all that process, but make sure that cost, quality, and content are monitored and everything is signed off properly by the, responsible, people. And first, the agent is nice, but we need to be able to use it. And so that's the last layer that we see, completely today. So the agent or the tech will answers that are two ways for everyone to be sure to access the built agent on that I could instance as we make sure that they have all the tool, content, and information to use them as, efficiently as, possible and be sure to improve quality of, the full switch. So today, we are going to have a, in-depth look into the dynamic selling assistant. So this solution built by Dataiku is really, targeting sales managers. So that could be people that are running one or several stores, around a given location. And that's people that's really under, shop floor and they have a lot of different responsibilities. So they can really they really have to make a lot of, fast decisions, very concrete decisions that can be very broad. So to make logistical concerns and need to make sure that they have the right people, give the right product at the right time for the right customers. And that can be complicated. So, obviously, with Dataiku, we can pre build a lot of, insights and a lot of, information. But for a lot of, ad hoc, content and, very fast questions that may be, different from what we have built in the past, it's not really something that you can do on the fly. So that's really where, the dynamic selling assistant is putting itself. So at the bottom, we have all the data that we have, using and like Carla, presented, it's really, the integration of a lot of different use cases. So we have a bunch of data that is whole data that is coming from directly, the store information. So you can have inventory, product characteristics, past sales, but we can also have informations that come from other processes, other projects under Dataiku. Here we showcase a few other business session, the Dataiku business solutions that are here to accelerate, your process and directly use the data that you already have with the Dataiku to get a clear insight. So for example, we have several segmentation topics regarding customer segmentation, customer lifetime value forecasting. We also have product related solution, for example, product recommendation. And lastly, I would say last but not the least, definitely one of the topics that is most relevant for store manager is demand forecasting to be sure that they have clear view of what products, are needed in the future. And on top of that, we have, price optimization with, Dataiku optimization to make sure that the price is, correct for everyone. So all of that data, can consider them as data products so they can be cleaned and maintained by the different data teams. But it's to make sure that they are available for the business people, we have created a set of agents that will all communicate with the different datasets and the different models and, run queries to make sure to transform basically the, questions that the manager has into, actionable insights. So we have, different agents that are all focused on the specific vertical. So customer agent, product agent, etcetera agents to be sure that they all have the relevant information so that they don't confuse things. And they are all aggregated into the main sales assistant agent that will be the ones that we will interact with, into agent type. And let me show you that right away. Here we see agent hub. And so that's as I showed it a bit earlier, it's a way to interact with all the agents that we've built, on the tech group. And so here, we can see that we have, our access or store manage store assistant agent, but, in agent hub, we are not limited to one agent. So we could have a set of agents that would serve different use case, and the agent hub will be able to automatically, drive which agent is needed and maybe use several agents. Or here in that case, we are specifying which agent we we want to use. And then we can just start chatting with it and, as different type of questions we want. So we have prepared a few different topics so we can have, supply topics where we ask what are the current status for our best selling products. And the agent will explain what he wants to do, current context of information, disable to ask, clarifying questions, make sure that it's running the right queries, presenting its topic, showing which different tools and which different, agents it will use and what its process. We can review it and, ensures that it's performing the right thing and then give you the right answers. So here you can, shape that into directly information, accessible, in the chat, but it can also be, sent by email directly through the agent and, in-depth. Here we answer one topic, but it can something that we run into, we can deep dive into a specific question starting with one question going to another and so on. Monitoring is also very important. So on the side, we have clear information of which tool was used and, which agent was used. So that way you can show that the people the agent is running the right thing. In this example, everything is prepared, but if you have a large family of agents with a lot of different tools, it can be sometimes a bit confusing. So it's nice to have everything at one, in one place and be sure that everything is, seen properly. And if we want to go into a lot more details, we have the Trace Explorer tell all you to see every code that was made by the agent and be sure that everything was run, exactly as you as you wished. On other topics, I say we, we have demand forecasting models that is used also for pricing information. So is it the agent is able to query the info the features from a dataset and run, live inference on the, markdown optimization model to be to find the right price optimization. And so in that case, we can find which product, we want to study. It is still the same best selling product. Like, we can see what is the impact in terms of sales, in terms of, profit, can ask a lot of different information. And you can go into more details. So we can ask for different products at the same time and give the agent a lot more, freedom. Yeah. It gives, the agent a range of price that they are allowed to to explore and see what would be the impact for the different, for for the different products. And the agent is able to to cross combine information. So here we know, what will be the, predicted sales. But if I find which one. Sorry. You can combine that with the inventory information and know which would be the supply left for, specific products. And that lets you, as a store manager run the full process. You can see which product I want to optimize, which product I want to sell more. Do we have enough supply to satisfy the future demand? And if I don't have enough supply, then send an email to, my procurement team to be sure that I have the right product at the right time. We can be also doing things that are a lot more precise. Here, we can look at cross selling operations based on our product recommendation and our market basket analysis solution. We combine all of this information to find which product we need to sell together. And, here is able to look at what are the best selling product and look at which, which product we want to sell as a bundle for cross sell or upsell. And, here we have a global approach, looking at all products for all customers, but we can also have strategies that will be very, focused if you work on a very small scale and look at specific customer information and write, automated email so that you can go into the realm of ePAP personalization and sending directly the right, marketing email to the right, people. Yes. I think that's it. We can continue into, the solution itself. So here we can see oh, sorry. It's very fast. We can see that even if it's, all packaged into a web app for the consumer, everything is accessible, in Dataiku with a full detail. So we assure that we have all the customer data and, all the product data, sales data that are accessible. So it's not a black box. It's not sent at, somewhere else. It's on your own instance managed by you. Same for the agents. It's on, on the instance connected to the different AI connections that I showed in the LMM mesh as it can be switched very easily from one connection to another. So if you want to compare different, LMM performance or if you just want to have the last, last version of the current LMMs, set of the LMMs. And same, we have access to the models themselves, and we can, validate their performance, past training information, if you will be needed. Lastly, I said that we can interact with the agent for the web app. We can send emails, but we can also have free, package solutions that would be then run automatically with a time trigger on scenarios. And so we can imagine having, questions that are already prepared for for you to send a weekly reminder, monthly, summary for your team and so on. And so you don't have to build everything each time. You just ask the agent to build the insight and send it, for the to the right person. So you can have a very integrated system, forever, quick information, live or more prepackage question if you want to have a repeat topic, for the agent. And, that's it for, the agent presentation. So I think we can have a bit of a q and a. If you have any question, feel free to ask them in the chat. Yes. We're happy to answer any questions you may have. That's a good question. Let me, show you, I'll share back my screen. So like I mentioned, everything is monitored inside of Dataiku. So you can be sure to monitor, cost. And if I go back to the address explorer, you can see how many tokens were used for a specific, specific query. You can see the cost, impacted with that. And so we can be sure, here, just a monitoring aspect, but then you can put, constraints on the connection to be sure that, you can put budgets, and so on, both obviously on the APIs themselves, but then for a specific user for specific application. So you have full, monitoring and control, onto those aspects. For AOI, I don't have numbers, for that. I don't know, Carla, if you have industry, numbers on the topic of, sales, improvement. Would you know? I I actually don't have, any quantified numbers that, we've we've captured. No. Don't have that at this time. Okay. For the multimodal, question, yes. So we we do support basically, it's, we we support everything that the l l, the LLM connections themselves supports, but you can, upload images. You can have, video processing. It's not necessarily done in the same interfaces. I know that, it's still a work in progress on a lot of the web interface, but, who what I showed the the tech who recipes was a process itself inside of the the flow. It's possible to integrate multimodal, data. And so as long as the model the LLM models themselves supports multimodal inputs, it can be done. So Dataiku is not selling, LNM model. So, just like for data connection and so on, it's your own it's basically you bring your own connection. So, whether you have your own API keys and you can exempt through that echo or you, post your own relay major and then you will plug it inside of that echo, but there's no surcharge, the data equivalent. And lastly, yes, for the charts, I cannot demo it right now for the the process that I have, but, yes, we're able to generate charts. So there's a you can generate charts live in, agent hub, and there's also, stories. It is another way to generate reports with live, charts based on datasets. So you can set your, datasets, your prepared dataset in a place and then, ask as the agent to build a few slides based on those, datasets, and it will be, build the the the chart themselves, but the chart will be, dynamic based on the dataset. So once the that the charts and the slides are created and the dataset is updated, so charts will automatically adapt itself. And I I noticed the question. I don't know if we we answered it around, hyper personalized offers. So, I think personalized email messages, and if we have experience there, supporting customers, with those types of use cases, the the answer, is yes. And, Kent, I don't know if you wanna share anything, regarding, how that links to the dynamic selling assistant when we think about personalized offers, content, those areas. Yeah. We have other applications, regarding specifically product recommendation. We're able to generate specific recommendation, what I mentioned in, in the presentation where let me, prepare that so I can share it. It will be nicer where we can generate specific emails. We've basically, based on the specific customer with, information on a specific product and other characteristic like their own segmentation, we're able to generate specific emails. So it's a bit small in the presentation, but, you can basically build an email with placeholder, and then we'll integrate, product characteristic. We also tried it with the specific promotion. So you can really combine all those application and do really one to one process, but it's integrated inside of the data workflow. So you don't have to manually create everything, and then you can export this kind of dataset into an ML tool. Regarding relevance and variations, that's a very important topic. We have, like I mentioned in the, compliance, layer, we have specific tools. If you have golden datasets and you can monitor the quality of the process, of the quality of answers, to be sure that you don't have any, discrepancies. And you can also use patterns that are LLM as a judge where we'll use it. For example, for the LLM, we will, for the compensation content. We want to be sure that the content generated is sensitive is not sensitive. And so we ask the LLM to validate this process and to flag any contents that see that's, inappropriate. And that way, you can, send it for manual review. And so we can address that for, like, live quality control. But, yeah, for more precise information, you need a golden dataset so you can be sure that for specific queries, you get specific type of answers. And, again, you can ask the agent to compare the the answers from the tool itself and from, your golden dataset and to be sure that, answers are matching. And, as many things in the tech, you can monitor that full time and, have specific process at hand and send you a warning if the query is, too low. Any other questions? Okay. So, thank you, everyone for, watching, this presentation and, all for your, good questions. And I hope, we show you some relevant information that give you a lot of, interest for Dataiku and, what we can do, as a AI platform and for agent application. Yes. Thank you for joining us.