Agentic AI: Automate digital ad sales workflow

In my previous article, “Agentic AI x SaaS: Making Things Work Better, Together” I made the case that: SaaS gives you structure. Agentic AI gives you motion. That line resonated with people based on their work experiences. Many PMs messaged me and asked how to apply it practically. Here's what I'm hearing from product managers: "Leadership wants AI in the product. We see a few use cases. But everything we come up with feels either too basic or way too ambitious." The basic ones like notifications, automated reminders etc., feel like table stakes. The ambitious ones like fully autonomous AI agents that make workflow decisions feel exciting, but out of reach for most teams today for various reasons. So progress gets stuck in brainstorming cycles. I think the reason is that we are trying to apply feature thinking to something that isn't a feature. Product managers are trying to apply agentic AI using the same thinking patterns that work for traditional feature development. And that's exactly why most AI initiatives in SaaS products feel forced or fail to gain traction. Why Most PM Mental Models Don't Work for Agentic AI Traditional product thinking is: Identify user need Build a feature Launch it Measure adoption But agentic AI doesn't live in features. It lives in the gaps: between an approval request and an escalation between a resignation and an offboarding checklist between a goal-setting module and the actual progress update Forget process diagrams. Agentic AI starts where your users pause, not where your flows end. To find those gaps, PMs need to do something uncomfortable: unlearn the old roadmap-first model and learn to listen for the micro-frictions users don't even log as bugs. If your workflow assumes people follow through, that's your first agent use case. While there's plenty of literature on agentic AI, workflow design, and AI for product managers, very few resources focus on how PMs can practically discover real use cases in their workflow-based products. Here's my attempt to help PMs do just that: uncover opportunities for agentic AI using a structured, usable framework. MOMENT: A Framework for Unlearning and Reframing This took me way longer to figure out than I would like to admit. I kept trying to build the perfect discovery process, mapping every possible AI use case. That was exactly the wrong approach. Then I started just... listening. Really listening to user frustration. Not feature requests or bug reports, but the sighs. The moments where people said "ugh, not this again." That's when I stumbled onto what I now call MOMENT. It's not fancy. But it works. M.O.M.E.N.T. = Missed cues, Overhead, Misalignment, Effort drag, Nags, Thread-hopping Stop designing for the ideal flow. Start building for the real friction. Use this like a scanner: the more MOMENTs you find in a process, the more it's begging for an agent. What Agentic AI Can Actually Do We talk a lot about what AI can do. But let's pin it to this framework: Read messy inputs (emails, notes, chats) Parse context and know what's missing Nudge with intelligence, not spam Act across systems (Slack + Calendar + Jira + HRMS) Propose next steps before users even think to ask Agents don't automate steps - they unlock stuck moments. Here's how they match up: Missed cues → Agent sees the delay, triggers action Overhead → Fills forms, auto-suggests info, simplifies steps Misalignment → Summarizes current state for all stakeholders Effort drag → Removes redundant thinking ("what do I do next?") Nags → Timely, relevant nudges where users already are Thread-hopping → Brings everything into one message Making It Real: Where You Start as a PM You don't need a platform overhaul. Start small. Sit with a user and watch them work. Tag friction with MOMENT terms. Ask: "Would a smart assistant help right here?" Sketch the agent's role: what it sees, says, and does. Pilot with a shadow agent (human mimicking the AI). You won't find agent use cases in your spec doc. You'll find them in user sighs. Examples of Use Cases in Common SaaS Systems CRM ERP Supply Chain Customer Support Being from HR Tech domain, I would like to list out some of the potential use cases for HR Tech platforms: How You Can Roll This Out Stage 1: Shadow agent. Collaborate with actual users and ask them to do what the AI would've done. Stage 2: 1 agent, 1 team. No platform launch - just stalled reviews + reminder. Stage 3: Expand. Slowly. Trust builds over months, not weeks. Pitfalls (Because They'll Happen) I'm still learning what works and what doesn't, but here's what I've seen go wrong: Too many nudges → Agent gets ignored No context in messages → People don't act Bad early guesses → Users lose faith fast How to fix: Add thresholds. Show confidence scores. Always explain "why this matters." What You Track Time saved (per user, per workflow) Drop in Slack/email reminders Increase in on-time completions Sentiment: "This saved me real time" Final Thought Don't chase AI. Chase friction. That pause where your user says "Ugh, not this again"? That's your entry point. Because what users remember isn't your models or tech stack. They remember the day the system quietly helped them instead of making them chase it. That's your product's sweet spot. Still figuring this out? Same here. Try the MOMENT framework on one workflow in your product and see what you discover. Would love to hear what friction patterns you find that I missed. #AgenticAI #B2B #SaaS #VertcialSaaS #AIInSaaS #SaaSProducts #HRTech #PMFramework #ProductThinking #UseCaseDiscovery #BuildBetterTools #EnterpriseSoftware #WorkflowInnovation #FutureOfWork Vishal Saha Vineet Pandita Mrigank Tripathi Vipul Mathur Akash Agrawal Vaibhav Goel Nityanand Gopalika Vivek Pandey Sumit Kumar Singh Naman Goel