How to Identify Opportunities and Risks with Chat Agent

Email News Update: If your bounce rate has elevated somewhat lately (my client went from their regular 0.1% to 0.5%) and it's mostly being driven by Yahoo soft bounces due to the mailbox being over quota. This is most likely due to a recent change in account storage allocation. Depending upon your ESP this should clear up automatically without causing you any issues if they scrub emails that repeatedly soft bounce. Example bounce message one of my clients on Klaviyo is seeing: "failed after I sent the message. Remote host said: 552 2 Requested mail action aborted, mailbox is over quota" As a side note Klaviyo, I find it interesting that your mail server said "after I sent the message" you'd tell us if you imbued the MTA with AI until it gained sentience right? - Insights from our Lead Technical Consultant, LoriBeth Blair

Email News Update: If your bounce rate has elevated somewhat lately (my client went from their regular 0.1% to 0.5%) and it's mostly being driven by Yahoo soft bounces due to the mailbox being over quota. This is most likely due to a recent change in account storage allocation. Depending upon your ESP this should clear up automatically without causing you any issues if they scrub emails that repeatedly soft bounce. Example bounce message one of my clients on Klaviyo is seeing: "failed after I sent the message. Remote host said: 552 2 Requested mail action aborted, mailbox is over quota" As a side note Klaviyo, I find it interesting that your mail server said "after I sent the message" you'd tell us if you imbued the MTA with AI until it gained sentience right? - Insights from our Lead Technical Consultant, LoriBeth Blair

Is this normal? Our team keeps asking each other for the same logs over and over. The other day, I noticed a pattern: “Can you pull the auth logs?” “Can you share the trace for that request?” “Where did you find that error?” I realized this isn’t a people problem. It’s a systems problem. If working together means running the same searches again and again, and passing screenshots around, something’s broken. Observability isn’t just about data. It’s about seeing the same picture. 👉 When everyone shares the same view, work flows. 👉 When they don’t, you get repeats, delays, and frustration. Does your team face this too? How do you avoid “log déjà vu”?

I’m a big believer that researchers should own the data quality process and tools but it’s not all roses and sunshine.  It’ll cost you time, money, and make you question the sampling ecosystem.   Here's why 👇 After testing 7 different fraud detection tools over the past few months, I’m sharing the 𝐭𝐨𝐩 𝐜𝐡𝐚𝐥𝐥𝐞𝐧𝐠𝐞𝐬 I’ve run into. I’d love to hear about your experiences—DMs are welcome! 1️⃣ 𝐓𝐢𝐦𝐞-𝐂𝐨𝐧𝐬𝐮𝐦𝐢𝐧𝐠 𝐃𝐚𝐭𝐚 𝐑𝐞𝐯𝐢𝐞𝐰 Before you can trust a new tool, you have to test it. That often means running a trial where you flag, but don’t block fraudulent respondents. I’ve spent countless hours manually reviewing data, row by row, to understand 1) if it’s effective at all 2) if the default settings are a good fit or need fine-tuning. It’s a massive time sink ⏳. 2️⃣ 𝐈𝐧𝐭𝐞𝐠𝐫𝐚𝐭𝐢𝐨𝐧 Getting a tool up and running isn’t always simple. Some require API integration, which means you either need to know how to code 👨💻 or have a programmer handy. Others demand an understanding of links and redirects 🔗. Either way, setup can easily add a day to your project timeline the first few times. 3️⃣ 𝐂𝐨𝐬𝐭𝐬 💸 These tools aren’t cheap and require a huge mindset shift: You’re being charged per 𝘴𝘶𝘳𝘷𝘦𝘺 𝘦𝘯𝘵𝘳𝘢𝘯𝘵, not per 𝘤𝘰𝘮𝘱𝘭𝘦𝘵𝘦.  If your incidence rate is low—or your sample size is large—it can get pricey fast. Many (most?) tools are designed for high-volume, subscription-based use, which can be tough for small businesses. 4️⃣ 𝐓𝐨𝐨𝐥 𝐋𝐚𝐲𝐞𝐫𝐢𝐧𝐠 🛠️ You need to layer multiple tools to cover all bases, and then you have to consider how they complement each other. No point paying for two tools that catch the same people. 5️⃣ 𝐓𝐡𝐞 𝐇𝐚𝐫𝐝 𝐓𝐫𝐮𝐭𝐡 😭 I hate to say it but no tool will solve all the problems. No matter what you choose, some data cleaning will 𝘢𝘭𝘸𝘢𝘺𝘴 (𝘢-𝘭-𝘸-𝘢-𝘺-𝘴!) be necessary on the back end. Sorry, but it’s true! Does that resonate with anyone else? #mrx #dataquality #surveys

📊 𝗪𝗵𝗮𝘁 𝗶𝗳 𝘀𝘁𝗼𝗰𝗸 𝗮𝗻𝗮𝗹𝘆𝘀𝗶𝘀 𝘄𝗮𝘀𝗻’𝘁 𝗷𝘂𝘀𝘁 𝗔𝗜 𝗴𝘂𝗲𝘀𝘀𝗲𝘀... 𝗯𝘂𝘁 𝗮𝗰𝘁𝘂𝗮𝗹 𝗱𝗮𝘁𝗮 + 𝗰𝗵𝗮𝗿𝘁𝘀? I’ve been tinkering with n8n lately, and I built something I’m pretty excited about:  a workflow that does both fundamental AND technical analysis — but with a twist.  What makes it different from “𝗷𝘂𝘀𝘁 𝗮𝗻𝗼𝘁𝗵𝗲𝗿 𝗔𝗜 𝗯𝗼𝘁”? ✅ Uses specialized tools (not only a generic LLM) — e.g., charting APIs + connectors.  ✅ Generates real chart images with the indicator you choose (MACD, RSI, etc.).  ✅ Gives buy/sell recommendations based on those charts.  ✅ Modular workflow → you can swap data sources, indicators, or even models. 𝗧𝗵𝗲 𝗰𝘂𝗿𝗿𝗲𝗻𝘁 𝗰𝗵𝗮𝗹𝗹𝗲𝗻𝗴𝗲: The chart images show up perfectly in the sub-workflow, but they’re not yet rendering inside the final chat interface. 👉 If anyone has solved this embedding issue in n8n, I’d love your input!  𝗪𝗵𝘆 𝘁𝗵𝗶𝘀 𝗺𝗮𝘁𝘁𝗲𝗿𝘀 𝗠𝗼𝘀𝘁 𝗔𝗜 𝘀𝘁𝗼𝗰𝗸 𝘁𝗼𝗼𝗹𝘀 𝘀𝗽𝗶𝘁 𝗼𝘂𝘁 𝘁𝗲𝘅𝘁.  𝗧𝗵𝗶𝘀 𝗼𝗻𝗲 𝗽𝗮𝗶𝗿𝘀 𝗮𝗻𝗮𝗹𝘆𝘀𝗶𝘀 + 𝘃𝗶𝘀𝘂𝗮𝗹𝘀, 𝘀𝗼 𝘆𝗼𝘂 𝗴𝗲𝘁 𝗰𝗼𝗻𝘁𝗲𝘅𝘁 + 𝗰𝗼𝗻𝗳𝗶𝗿𝗺𝗮𝘁𝗶𝗼𝗻 𝗯𝗲𝗳𝗼𝗿𝗲 𝗺𝗮𝗸𝗶𝗻𝗴 𝗮 𝗰𝗮𝗹𝗹.

🚨 SELF-SERVING POST ALERT! 🚨 Are you costing your business more than you realise by using non-technical internal staff to manage your IT and security? Let's skip straight to the solution, so there's no beating about the bush. Get proper IT and security support for your business. It will almost certainly save you money, especially in the long term. We can look at why... You have someone in your business who is "good with tech" - words that often sent shivers down my spine! 😱 That's great, it's certainly helpful. But should you put them in charge of your IT? No! Definitely not. Should they be responsible for securing your business? Hell no! 🙅♂️ First of all, they don't have the experience, knowledge or skills to make sure things are set up properly. You have a ticking time bomb on your hands right there. Well-meaning hobbyists seem great, but they will be leaving gaps all over the shop. Or office... or... well... you get it. Your business. But also, how much is it actually costing you? You hired them to do a specific job, right? The one you're paying them for. Meanwhile, they're getting distracted left, right and centre by tech issues. Other team members will be bothering them too. All the while, the poor soul is barely getting any of their actual work done. They'll be spending time asking ChatGPT for answers on how to do things that they have no business getting involved with, whilst their own workload lags behind. Instead, you could be getting it all taken care of, proactively, by a team who has invested in the right tools, the right skills and has the time to get things done, fast. Get in my DM's, let's get the ball rolling!

Government Use Case of the Week- Thanks to Dr. Nicol Nicola, DBA -- Reconciling messy datasets? There’s a faster way! ✅ Nicol was prepping for a multi-state job exchange conference and had to reconcile two Excel files—each with different structures—to report totals by category. At the same time, she was building a healthcare workforce presentation for New Jersey using American Community Survey data. Instead of manually aligning rows or searching for code online, she uploaded both files to ChatGPT with a prompt to match categories and generate totals. Then, she asked for R code to calculate the total number of healthcare workers by type.  What would’ve taken hours took seconds—with results that powered two high-stakes briefings. ⏱️ “It saved me a lot of time. Normally, it would take much longer to find the right code online.”  ⚠️ As always, validate the outputs—especially when generating code. 👉  Have a prompt you’re proud of? DM us or drop it in the comments. Thanks to Dr. Nicol Nicola, DBA for sharing! OpenAI

Here's the latest in the My Protocols series; this one on My Personal Agents. In writing that I stumbled onto a pattern that was always there in the logic of personal agents, but I think now points to the path to scale deployment. That scale comes from the same pattern underpinning thousands of use cases for briefing/ deploying personal agents. The pattern is that Agents programmatically apply verbs to nouns within a defined context. For example: - I (identifier and context) want to buy (verb) a widget (noun) - I (identifier and context) want to renew (verb) my passport (noun) - We ((identifiers and context) want to compare (verb) our service contract(s) (noun) with other options - I (identifier and context) want you to fix my broken XYZ (noun) So all that changes to enable those thousands of use cases are the verbs and the nouns. More here https://lnkd.in/er_36A4G

🖌 Demystifying RAG (Retrieval-Augmented Generation) 📌 RAG = an LLM that first retrieves facts from your own knowledge (docs, wikis, DBs, tickets) and then generates an answer grounded in those facts—with citations. 📌 Building Blocks of RAG Prepare data: clean ➜ chunk (~300 tokens, 10–20% overlap) ➜ embed (dense + sparse) ➜ store with metadata & ACLs. Retrieve: hybrid search + reranking, with filters (recency, tags, permissions). Augment: LLM uses only supplied context; no guessing. Generate: answers with citations, confidence, and optional structured output. 📌 Why it’s required / why it matters 1️⃣ Accuracy: Grounding reduces hallucinations 2️⃣ Freshness: Answers reflect the latest policies/prices 3️⃣ Security: Enforces user/role permissions (ACLs) 4️⃣ Cost & speed: No heavy re-training for every doc update 5️⃣ Auditability: Citations for compliance & internal reviews 📌 Real-world use cases 1️⃣ CS/CX Helpdesk: IT/HR FAQs, refund/warranty policies with links to the exact clause 2️⃣ BFSI: KYC/AML & card/loan policy Q&A by region + effective date 3️⃣ Contact Center & WhatsApp: Micro-journeys grounded in CRM/ERP data; better Window Message Ratio 4️⃣ Sales Enablement: Instant battlecards, case studies, competitor intel by vertical #RAG #LLM #GenerativeAI #InformationRetrieval #VectorSearch #EnterpriseAI #CustomerSuccess #BFSI #ContactCenter #WhatsAppAI #DataProducts #AIinProduction