Implementing Customer Experience Software

Today, we're publishing the AI Agent Engine! 🙌 It's a distillation of our learnings from many successful deployments of AI agents at enterprises. Ultimately, this is what's required for a successful implementation of AI agents. Of course, this is specific to our space (customer service & experience), but the themes will carry over to any vertical. 1. First, you have the "AI agent", defined as a software system that can autonomously do the work of a human agent, such as looking up data, taking actions, making complex decisions, and writing personalized responses. This is the holy grail that everyone wants to get to. 2. Around it, the rest of the engine is designed to reinforce the AI agent and allow it to continuously improve. This starts with a mechanism (i.e. "Routing") that determines when the conversation should be escalated to a human in the loop. This is key because it allows you to roll out your AI agent incrementally. 3. Next, you have the AI tooling for your human agent to use that automates away mundane tasks, like drafting an answer, finding relevant information, polishing tone, etc. We call this "Agent Assist", and it's akin to a copilot. 4. Then, the conversations all feed into a central data platform, our "Admin Dashboard", that allows the leaders of the team to use LLMs to analyze the conversations. This will surface themes, trends, and anomalies in the data easily. It'll also identify gaps in your knowledge, for example, and proactively tell you how to fix them. 5. Finally, you need a way for human staff to "QA" the conversations to constantly give feedback. We've built this directly into the product. These components form the AI Agent Engine, a helpful framework for thinking about AI implementations. The full post written by Bihan Jiang, Kaylee George, and Cynthia Chen is below! 👇

Yesterday I explored to key types of SaaS agreements and today I want to explore two other topics within the SaaS industry: Data Processing Addendums (DPAs) and Compliance Requirements. Let's break them down. A) Data Processing Addendums (DPAs) These often form a key part of many SaaS Agreements and are essential in our privacy-conscious era, especially with regulations like GDPR and CCPA. Key Parts Include: -Data processing roles: DPAs clearly define who's the data controller and who's the data processor, establishing responsibilities. -Processing limitations: They specify allowed data uses, preventing misuse and ensuring compliance with privacy laws.== -Security measures: DPAs outline required technical and organizational safeguards to protect personal data. -Sub-processor management: DPAs outline the terms of engagement for both engaging and for the managing and adding/removing of sub-processors, which are entities which conduct further processing of a customer's data (e.g. for customer support, website hosting, and the like.( -Data subject rights: DPAs address how to handle requests from individuals about their personal data, ensuring regulatory compliance. B) Compliance Clauses Example Clauses Include: -Industry-specific standards: Agreements may reference HIPAA for healthcare, PCI DSS for payments, or SOC 2 for general data security. -Audit rights: Customers often require the right to audit the SaaS provider's compliance, either directly or through third-party assessors. -Certifications and reports: Clauses may require SaaS providers to maintain certain certifications or provide regular compliance reports. -Breach notification: Specific timeframes and processes for reporting security incidents or data breaches are typically mandated. Understanding and carefully negotiating DPAs and compliance clauses is crucial in today's incredibly data-driven and increasingly regulated business environment. #legaltech #innovation #law #business #learning

There is an implementation trap with AI that few are talking about and I wanted to bring it up. Specifically, how two teams experience the same tech, but in two realities. Here’s the scenario that leaders need to look out for: Let’s say your company’s CX team rolls out AI orchestration with clear goals: unify data, automate touchpoints, and reduce friction across the customer journey. (I’d go more specific but let’s keep it high level for the sake of simplicity.) From the CX Team’s view, this is a strategic win. “We’re finally connecting the dots.” “This gives us real-time visibility and faster resolution.” “We can drive proactive service, not just reactive triage.” But in the contact center? “This thing is pulling tickets off our queue.” “I have to bounce between more tools now.” “Why weren’t we part of this decision?” Same rollout. Completely different realities. That’s the miss in most change efforts. Context is everything. The CX team sees orchestration as an upgrade. The contact center sees it as disruption. And unless you understand both perspectives, you don’t get transformation. Instead, you get tension. An important point here is that emotional reactions aren’t resistance. They’re system feedback. They tell you where the friction lives. Where the incentives don’t line up. Where the design skipped the people doing the work. If you want AI implementations to be successfully implemented, it starts with this: Co-design the change with those it impacts. Build for team workflows impacted by the tech, not just those customers see. Use the emotional signals like frustration, anxiety, and hesitation as diagnostic tools. Because when the context is ignored, the system will definitely let you know and push back! I see this all the time. And while AI Is exciting and people are starting to lean into the tech, the simple reality is that it’s still about people. People > Tech #customerexperience #ai #changemanagement #contactcenter #changemanagement #changeleadership

Conversational AI is transforming customer support, but making it reliable and scalable is a complex challenge. In a recent tech blog, Airbnb’s engineering team shares how they upgraded their Automation Platform to enhance the effectiveness of virtual agents while ensuring easier maintenance. The new Automation Platform V2 leverages the power of large language models (LLMs). However, recognizing the unpredictability of LLM outputs, the team designed the platform to harness LLMs in a more controlled manner. They focused on three key areas to achieve this: LLM workflows, context management, and guardrails. The first area, LLM workflows, ensures that AI-powered agents follow structured reasoning processes. Airbnb incorporates Chain of Thought, an AI agent framework that enables LLMs to reason through problems step by step. By embedding this structured approach into workflows, the system determines which tools to use and in what order, allowing the LLM to function as a reasoning engine within a managed execution environment. The second area, context management, ensures that the LLM has access to all relevant information needed to make informed decisions. To generate accurate and helpful responses, the system supplies the LLM with critical contextual details—such as past interactions, the customer’s inquiry intent, current trip information, and more. Finally, the guardrails framework acts as a safeguard, monitoring LLM interactions to ensure responses are helpful, relevant, and ethical. This framework is designed to prevent hallucinations, mitigate security risks like jailbreaks, and maintain response quality—ultimately improving trust and reliability in AI-driven support. By rethinking how automation is built and managed, Airbnb has created a more scalable and predictable Conversational AI system. Their approach highlights an important takeaway for companies integrating AI into customer support: AI performs best in a hybrid model—where structured frameworks guide and complement its capabilities. #MachineLearning #DataScience #LLM #Chatbots #AI #Automation #SnacksWeeklyonDataScience – – –  Check out the "Snacks Weekly on Data Science" podcast and subscribe, where I explain in more detail the concepts discussed in this and future posts:    -- Spotify: https://lnkd.in/gKgaMvbh   -- Apple Podcast: https://lnkd.in/gj6aPBBY    -- Youtube: https://lnkd.in/gcwPeBmR https://lnkd.in/gFjXBrPe

5 Mistakes People Make in Their First CRM Implementation (and what to do instead if you want to scale): Implementing a CRM isn’t hard. But implementing it well? That’s where most companies fall short. If you're setting up your CRM for the first time, avoid these: 1) No clear plan Most teams jump straight into setup mode. No blueprint. No long-term vision. Just “get Salesforce in place.” You wouldn’t build a house without a floor plan. Treat your CRM the same way. 2) Building for today, not for scale Your CRM might feel fine at $2M ARR. But what about when your team doubles? What happens when your pipeline triples? If you don’t build for scale early, you’ll end up rebuilding later. 3) Ignoring the humans CRMs are about systems... But people still have to use them. If your reps don’t adopt it, it doesn’t matter how good it looks. Build workflows that match how your team actually works. And make the process easy to follow. 4) Tracking the wrong things Just because Salesforce can track everything… Doesn’t mean it should. Too many fields = confusion. Confusion = bad data. Focus on the fields that drive results. 5) Treating it like a one-time project Your CRM isn’t something you “set and forget.” It evolves with your business. Revisit it. Refactor it. Keep it aligned with your goals. The best CRM setups aren’t perfect. They’re maintained. Thanks for reading. Enjoyed this post? Follow Jordan Nelson And share it with your network. Want more ideas like this? Subscribe to my newsletter. Sign up here: https://lnkd.in/gBukTtJN I share practical systems founders can use to fix your backend and stop drowning your team.

7 ways to seamlessly integrate AI into your users journey 1. The core purpose of AI directly shapes the user’s journey. 
 Conduct user research to identify key pain points or tasks users want AI to solve. ↳ if the startup’s AI helps automate content creation, what’s the user’s biggest friction in the current workflow? 2. Where will the AI interact with users within the product flow? Map out where AI should intervene in the user journey. For instance, ↳ does it act as an assistant (suggesting actions)
 ↳ a decision-maker (making recommendations)
 ↳ a tool (executing commands) 3. Simplify feedback loops help build trust and comprehension
 Focus on how users will receive AI feedback. ↳ What kind of feedback does the user need to understand why the AI made a recommendation? 4. Build a modular, responsive interface that scales with AI’s complexity. Visual elements should adapt easily to different screen sizes, user behaviors, and data volume. ↳ if the AI recommends personalized content, how will it handle hundreds or thousands of users while maintaining accuracy?
 
 5. Use layers of transparency At first glance, provide a simple explanation, and offer deeper insights for users who want more detailed information. Visual cues like "Why?" buttons can help. For more on how layered feedback can improve UX, check out my post here 
https://lnkd.in/eABK5XiT 6. Leverage Emotion Detection patterns that shift the tone of feedback or assistance. ↳ when the system detects confusion, the interface could shift to a more supportive tone, offering simpler explanations or encouraging the user to ask for help. For tips on emotion detection, check this https://lnkd.in/ekVC6-HN 7. Prototype different AI patterns ⤷ such as proactive learning prompts ⤷ goal-based suggestions ⤷ confidence estimation based on the business goals and user needs Run usability tests focusing on how users interact with AI features. ↳ Track metrics like user engagement, completion rates, and satisfaction with AI recommendations. Check out the visual breakdown below 👇 How are you integrating AI into your product flows? #aiux #scalability #designsystems #uxdesign #startups

"How should I think about the security and privacy of customer data if I use ChatGPT in my product?" We get this question a lot at Vanta. If you’re planning to integrate a commercial LLM into your product, treat it like you would any other vendor you’re onboarding. The key is making sure the vendor will be a good steward of your data. That means: 1. Make sure you understand what the vendor does with your (= your customers'!) data and whether it may train new models. Broadly speaking, you don't want this, because in the process of training a new model, one customer's data may show up for another customer. 2. Remember that if your LLM vendor gets breached, it's leaking your customers' data, and you'll need to let customers know. In my experience, your customers are unlikely to care that it was another provider's "fault" – they gave the data to you. As with any other vendor, you'll want to convince yourself that your LLM vendor is trustworthy. However, if you’re using the free version of ChatGPT (or any free tool), you might not be able to get the same contractural assurance or even be able to get specific questions answered by a person (not, you know, an LLM-powered chatbot.) In those cases, we recommend: 1. Adjusting settings to ensure your data are not shared or used to train models. 2. Even them, understand there's no contractural guarantee. We recommend keeping confidential, personal, customer, or private company data out of free service providers for this reason. As ever, ymmv. Matt Cooper and Rob Picard recently hosted a webinar, answering common questions about AI, security, and compliance. Link in comments if you're curious for more.

We spend months interviewing to find the “perfect” CSM… and then set them up to fail. Here’s the reality I see too often: ❗ New hires are thrown customers after 1–2 weeks. ❗ Product training is rushed or nonexistent. ❗ SOPs are thin, outdated, or missing. ❗ Leaders don’t invest the time to set expectations or coach. ❗ And then KPIs are handed down that even seasoned CSMs struggle to hit. The issue isn’t the talent, it’s the lack of enablement. But here’s the good news: you don’t need a dedicated L&D team or endless resources to onboard well. You need intention. A simple enablement plan for new CSMs (even with limited resources): 1️⃣ Onboarding Buddy - Pair new hires with an experienced CSM for shadowing, Q&A, and feedback. 2️⃣ 30-60-90 Plan - Outline clear goals and expectations for their first 3 months. (Focus on learning before doing.) 3️⃣ Product Deep Dives - Host weekly “lunch and learns” where Product, CS, or Support walks through one feature in detail. Have them shadow customer onboarding or watch recordings. 4️⃣ Playbook Starter Pack - Even if you don’t have full SOPs, document 3–5 repeatable workflows (renewals, QBR prep, escalation handling). 5️⃣ Mock Meetings – Run practice customer calls internally before they ever face a real customer. 6️⃣ Leader Time - Block weekly 1:1s focused not just on performance but on coaching, context, and confidence-building. These aren’t heavy lifts, they’re discipline and focus. If you want your CSMs to succeed (and your customers to stay), stop spending all your energy on hiring the “perfect” candidate and start spending more on enabling them once they walk through the door.

Here's the truth: You don't need a $250K CDP to get a unified customer view. Your customer data is trapped in silo'd platforms...CRM, email, loyalty, reviews, ecom/oms. But none of them talk to each other. You understand the value of unifying and OWNING your customer data, but implementation feels impossible with your limited budget and tiny tech/data teams. Start here: 1.) Document what you already have. List every platform containing customer data. You'll feel better making this first little step. 2.) Pick which system will be your foundation. Often your CRM, but could be your email platform if it has better engagement data. Don't try connecting everything at once. 3.) Use ETL tools, not a full CDP. Tools like Fivetran or Funnel let you authenticate with just your admin credentials. They'll pull your data into a warehouse YOU own for $1-2K monthly. 4.) Own your data. That means storing data in your warehouse. Google Cloud or Snowflake are user-friendly options for under $1k/mo for most mid-market brands. No more data trapped in vendor platforms. 5.) Start basic, get wins. Daily batch syncs of core customer data will solve 80% of your problems. Setup simple pipelines or (eek) scheduled queries to bring data together. Most companies overcomplicate this. They try integrating 20+ data sources simultaneously or request hourly syncs they don't need. Start with 3-5 critical sources syncing daily. 6.) Get help if you need it. There are plenty of partners and data consultants that do this for a living. You'll pay consulting rates, but this can be up and running in weeks vs. lengthy 6/12/18mo CDP timelines. The ROI on the data... - Truly understand customer LTV and retention - Build segmented audiences that sync across channels - Make decisions without having to "trust your gut" I've seen mid-market brands who've abandoned $500K CDP implementations for this approach and gotten better results with 80% less frustration. CFO will be happier. Your marketing team will be happier. What customer data sits trapped in your platforms today? Ps... Yes, I'm aware CDPs do a lot more than 1P data merging. Yes, there tons of other data vendors besides those mentioned. Yes, I know the costs above are wildy vague (iT dEpEnDs). #cdp #customeranalytics #moderndatastack

How should you structure your customer 360? Option 1: Create one row per customer with all attributes (e.g. name, age, address) and computed features (e.g. total page views, num_login_last_7_days, last_5_products_clicked, total_revenue_in_last_6_months) as columns. Option 2: Separate dimensions (customers) and facts tables (login_events, product_click_events) and let downstream users compute features ad-hoc. There’s no universal answer, but here are some considerations: 💾 Storage is cheap, compute is costly If you're referencing the same feature (e.g., last_5_product_clicked) multiple times in dashboards or marketing segments via rETL, it’s better to compute it once and store (cheap) than do a JOIN (costly) on every query. ⚡ Optimize with batch processing Computing features in batch instead of one at a time allows data teams to run multiple SQL queries in parallel, share intermediate results, and significantly reduce costs. 🛠️ Self-serve is great - if the team has the right skills Enabling business teams to self-serve features works only when they are tech savvy enough to do so. Feature computation can get tricky, particularly if ID stitching is required. 🧹 Handling dirty data is a universal challenge With messy data (like having multiple login events, e.g., login_ios_v1, login_android_v2), it's better to have data teams compute aggregates like total_login_last_7_days and make them available to business stakeholders. The ideal customer 360 structure balances efficiency, accessibility, and data quality – and empowers your organization with smart, fast decision-making capabilities.