2823 - Building an event marketing intelligence solution powered by watsonx.pdf

《2823 - Building an event marketing intelligence solution powered by watsonx.pdf》由會員分享，可在線閱讀，更多相關《2823 - Building an event marketing intelligence solution powered by watsonx.pdf（38頁珍藏版）》請在三個皮匠報告上搜索。

1、Does this sound familiar?PAIN POINTSSWEAP LEAD COMPANION10/22/2024Imagine.SWEAP LEAD COMPANION10/22/2024Imagine.Every SDR having an AI Assistant to capture,assess and rate leads.SWEAP LEAD COMPANION10/22/2024Imagine.Saving hours of work while you are able to focus on your most important customer int

2、eraction.SWEAP LEAD COMPANION10/22/2024Imagine.Badge Scanning and real-time enrichment with all data accessible online to give you a complete contact profile.SWEAP LEAD COMPANION10/22/2024IBMxSweap:AI Lead CompanionResult of a six week Pilot Development in joint collaboration between IBM and Sweap10

3、/22/2024SWEAP LEAD COMPANION10/22/2024What you will learnTAKEAWAYSSWEAP LEAD COMPANION10/22/2024Vision models in a image processing workflow.Using WatsonX Governance to monitor AI systems in production.Agentic Systems and how to employ them.How to use WatsonX AI to build a user-centric AI applicatio

4、n.SWEAP LEAD COMPANION10/22/2024Our Goals for the PilotRecap of the kick-off Workshop,goals for the MVP,success metrics and Project team.AGENDABuilding the PilotProduct design,data science,backend and frontend development efforts to bring the AI Companion to life.PILOTRefining the PilotThe future is

5、 bright with the AI Lead Companion.A look at the next steps and business case for the solution.MVPGOALSSWEAP LEAD COMPANION10/22/2024Our Initial GoalsRecap of the Workshop,goals for the MVP,success metrics and Project team and timeline.GOALSSWEAP LEAD COMPANION10/22/2024Workshop RecapDesign Thinking

6、 WorkshopBerlin JULY 9Enterprise Design Thinking Workshop To define value&Pilot Scope8 hoursOutcome?Use Case identified&aligne?Data and solution explore?Scope and success criteria define?Develop and align on MVP development planApproach/Activitie?Workshop coverin?Use Case Alignmen?Solution Discover?

7、Pilot ScopingSweap x IBMSWEAP LEAD COMPANION10/22/2024Envisioned SzenarioPainkillers and GainsAn AI mobile application which enables SDRs to scan the event badge and enrich the contact with external data in real time.Further recording voice conversation notes,which are analysed and scored by the pre

8、define categoriesCurrent SituationPains and ChallengesTo collect sales leads at an event or industry trade fair,you often have to rely on the organisers scanning system.This solution is expensive,provides data with a time delay and fragments the call documentation.SCENARIOSWEAP LEAD COMPANION10/22/2

9、024Meet the teamSweap x IBMChristophAnnalisaArisChrisSafaltaDavidSaraPaulaSvenJakobWissamDevelopmentData ScienceDesignFrontendSalesSWEAP LEAD COMPANION10/22/2024PILOTBuilding the PilotProduct design,data science,backend and frontend development efforts to bring the AI Companion to life.SWEAP LEAD CO

10、MPANION10/22/2024Badge ScanningExtract lead data(name,surname,role,company)Search for and access user information from one third-party providerProvide information about the lead:interest,fit,lead profile,technical maturity,decision-making authority,etc.)Based on 5 Dimensions:Product Fit Score,reason

11、 for action,Competitive mention,interest matchingProvide Next Best actions suggestion on how to approach the leadProvide a checklist while recording the audio as guidance for the user Use more third party data to enrich the profile even further(e.g.CRM)Data EnrichmentRecord Conversation SummaryLead

12、Scoring134AI Lead CompanionSweap x IBMTo-beMVP Scope2SWEAP LEAD COMPANION10/22/2024AgentWatsonXAIPostgresDBFrontendCOSVPC(+GPU)EnrichmentBackendArchitectureSWEAP LEAD COMPANION10/22/2024AGENTSAgents are a new paradigm in building AI applications.In them,a LLM is driving decisions along defined paths

13、 and constraints using a set of tools.What are AI agents?SWEAP LEAD COMPANION10/22/2024AGENTSAgentic systems are modular by design and easy to monitor.Consisting of nodes and paths,we can define behaviour on a granular level.This makes integrating external data sources safer and faster.Why AI agents

14、?SWEAP LEAD COMPANION10/22/2024Design of the Agent First PartSecond PartCan run multiple timesImage ProcessingEnrichmentAuto ScoreAudio ProcessingExtraction,ProcessingLLM ResponseWaiting for user input.SWEAP LEAD COMPANION10/22/2024Design of the Agent First PartSecond PartAgent takes a decision here

15、Agent takes a decision hereCan loop multiple timesWe interrupt the agent halfway through to way for the user to record an audio note.SWEAP LEAD COMPANION10/22/2024Image Processing&External DataLLaVa on VPCAnthropic Claude Sonnet 3.5Person and Company Enrichment via ClayQuery Customer CRM*Coming soon

16、SWEAP LEAD COMPANION10/22/2024Auto Score+=Ideal Customer Profile(ICP)connected to Event“SME,50 employees,contact in engineering lead position or CTO,marketing sector”Auto Score generated by Llama3 70B via WatsonX AI“auto_score”:“A”Person and Company Enrichment via Clay“John Doe,engineering lead Acme

17、 Inc.marketing sector”SWEAP LEAD COMPANION10/22/2024Convert speech to text using AI-powered speech recognitionAI-poweredSWEAP LEAD COMPANION10/22/2024Tailored Customise large speech models to your domainFast&Accurate10 x Faster,42%more accurate,and 5x smaller than OpenAI Whisper*https:/ AI Technolog

18、ySpeech-to-TextMaps important information from the transcription to predefined categories(product-fit,competitive mention,etc.).Extract InsightsSWEAP LEAD COMPANION10/22/2024Auto ScoreCompares the created contact to the ideal customer profileConversationGenerates text in a conversational way for an

19、engaging user experienceIBM AI TechnologyWatsonX AIAt runtimeRun as a batch job and helps monitoring the performance of the LLMs leveraged in the applicationBatch JobsSWEAP LEAD COMPANION10/22/2024LLM-as-a-judgeDetect hallucinations and evaluates completeness of extractionIBM AI TechnologyWatsonX AI

20、Asynchronous?Code Engine(cost based on traffic but need one instance for prod?larger instance for backen?tiny instance for fronten?VPC(for LLaVA image captioning,high fixed cost)Server:SWEAP LEAD COMPANION10/22/2024Data Storage?Postgres DB(permanent scalable fixed cost?Object Storage(usage based pri

21、cing)IBM InfrastructureData Storage&Server?Docker image repository(usage based pricing?Classic deployment pipelines(backend and frontend)user based pricin?Deploy on each commit on dev branch(build docker image and deploy?Log Analysis(usage based pricing)DevOps and ObservabilitySWEAP LEAD COMPANION10

22、/22/2024IBM AI Tools(all usage based pricing)?Speech to Tex?Watson Studi?Watson Machine LearningIBM InfrastructureAI Tools,Dev Ops&ObservabilityMonitoring&TracingSWEAP LEAD COMPANION10/22/2024Monitoring of LLM results via Custom LLM-as-a-judge cron jobRegular MonitoringAgent Tracingvia Langsmith(par

23、t of the Langchain family)Frontend DevelopmentKey Features Responsive,cross-platform UI Instant contact capture via camera Integrated audio recording interface Dynamic lead scoring display Offline capabilities for reliable useTechnical Highlight?Built with Ionic Reac?Cross-platform(iOS&Android?Custo

24、m React components for reusabilit?WebSocket for real-time update?Integration with AI backend servicesWireframe state of the frontend in SeptemberSWEAP LEAD COMPANION10/22/2024FUTURERefining the pilotA look at the next steps and the branding of the Lead Companion in regards to the vision for the app.

25、SWEAP LEAD COMPANION10/22/2024Branding00:18Based on what youve shared,this looks like a promising lead.Heres what stands out:1 minute agoSWEAP LEAD COMPANION10/22/2024Warm,clean and friendly design.We want to stand out from other business applications in the sales and marketing space.SWEAP LEAD COMPANION10/22/2024Demo Time SWEAP LEAD COMPANION10/22/2024Lead Companionv0.2(alpha)Built bySWEAP LEAD COMPANION10/22/2024Want an early look?Sign up for the waitlist now!SWEAP LEAD COMPANION10/22/2024Questions&AnswersLead CompanionSWEAP LEAD COMPANION10/22/2024