[
  {
    "startMs": 1309,
    "text": "[music]"
  },
  {
    "startMs": 12800,
    "text": "Well, first of all, hi everyone. I'm an"
  },
  {
    "startMs": 15679,
    "text": "engineer on the product growth team at"
  },
  {
    "startMs": 17600,
    "text": "Notion and now working on the GTM"
  },
  {
    "startMs": 20160,
    "text": "engineering team. A year ago, I would"
  },
  {
    "startMs": 22720,
    "text": "have told you that building a GTM system"
  },
  {
    "startMs": 25039,
    "text": "was a marketing ops problem. And today,"
  },
  {
    "startMs": 27439,
    "text": "I think it's one of the most interesting"
  },
  {
    "startMs": 29519,
    "text": "distributed systems problems that I've"
  },
  {
    "startMs": 31599,
    "text": "worked on."
  },
  {
    "startMs": 33520,
    "text": "GTM at most companies involve a"
  },
  {
    "startMs": 35920,
    "text": "spiderweb of tools like what you see"
  },
  {
    "startMs": 38079,
    "text": "here, and they're stitched together by"
  },
  {
    "startMs": 40719,
    "text": "customer notes, proposals, contracts"
  },
  {
    "startMs": 44000,
    "text": "that are passed back and forth. Over the"
  },
  {
    "startMs": 46960,
    "text": "last few months, a small team and"
  },
  {
    "startMs": 48719,
    "text": "myself, um, we've been trying to turn"
  },
  {
    "startMs": 51039,
    "text": "this spider web into a unified GTM"
  },
  {
    "startMs": 53760,
    "text": "system. We're still in the midst of it,"
  },
  {
    "startMs": 56000,
    "text": "but we've learned a ton that I'd like to"
  },
  {
    "startMs": 57760,
    "text": "share with you today."
  },
  {
    "startMs": 61039,
    "text": "So, this isn't really a new problem."
  },
  {
    "startMs": 63920,
    "text": "We've been wrestling with pieces of it"
  },
  {
    "startMs": 65680,
    "text": "for years. Life cycle messaging, product"
  },
  {
    "startMs": 68560,
    "text": "recommendations,"
  },
  {
    "startMs": 70080,
    "text": "sales automations, customer data, and"
  },
  {
    "startMs": 73280,
    "text": "onboarding. But the solutions were"
  },
  {
    "startMs": 75520,
    "text": "fragmented because the underlying"
  },
  {
    "startMs": 77520,
    "text": "technology forced them to be."
  },
  {
    "startMs": 80560,
    "text": "Then over the winter break, our CEO Ivan"
  },
  {
    "startMs": 83520,
    "text": "built spent it building a video game and"
  },
  {
    "startMs": 86320,
    "text": "he came back convinced that software"
  },
  {
    "startMs": 88560,
    "text": "engineering could be applied to many"
  },
  {
    "startMs": 90400,
    "text": "problems that were previously unwieldy"
  },
  {
    "startMs": 92720,
    "text": "or too costly. At the same time, our"
  },
  {
    "startMs": 96400,
    "text": "ability to execute skyrocketed with"
  },
  {
    "startMs": 99280,
    "text": "agentic technology and the breath of"
  },
  {
    "startMs": 101920,
    "text": "problems we could solve did too."
  },
  {
    "startMs": 104640,
    "text": "the costly, time-consuming, and"
  },
  {
    "startMs": 107200,
    "text": "previously unsolvable spaghetti could"
  },
  {
    "startMs": 109439,
    "text": "now be sorted. So, what we found is that"
  },
  {
    "startMs": 112799,
    "text": "GTM had become a systems problem. That"
  },
  {
    "startMs": 116240,
    "text": "made us realize we could chip away at"
  },
  {
    "startMs": 118159,
    "text": "this holistically."
  },
  {
    "startMs": 120560,
    "text": "In case you don't know, notion's"
  },
  {
    "startMs": 122560,
    "text": "platform is a collaborative brain for"
  },
  {
    "startMs": 124640,
    "text": "human and agents to think together. Over"
  },
  {
    "startMs": 127520,
    "text": "the years, we've evolved into a context"
  },
  {
    "startMs": 129920,
    "text": "layer for your company and AI agents can"
  },
  {
    "startMs": 132480,
    "text": "act on it. Notion's business moves"
  },
  {
    "startMs": 135200,
    "text": "between self-s serve growth and sales"
  },
  {
    "startMs": 137840,
    "text": "assist. Um, and customers move between"
  },
  {
    "startMs": 140560,
    "text": "these two motions all the time."
  },
  {
    "startMs": 143840,
    "text": "The problem is that customers experience"
  },
  {
    "startMs": 146160,
    "text": "one journey, but internally it is"
  },
  {
    "startMs": 149040,
    "text": "supported by disconnected systems that"
  },
  {
    "startMs": 151760,
    "text": "don't actually talk to each other very"
  },
  {
    "startMs": 153519,
    "text": "well. These processes were rife with"
  },
  {
    "startMs": 156319,
    "text": "human error and put a lot of cognitive"
  },
  {
    "startMs": 159120,
    "text": "burden on our teams. For example, sales"
  },
  {
    "startMs": 162160,
    "text": "reps are probably not the strongest at"
  },
  {
    "startMs": 164560,
    "text": "managing systems, but their strength is"
  },
  {
    "startMs": 166959,
    "text": "in sussing out human signals during the"
  },
  {
    "startMs": 169360,
    "text": "buying process. So, every time they had"
  },
  {
    "startMs": 172319,
    "text": "to context switch between after a call,"
  },
  {
    "startMs": 175120,
    "text": "doing research, drafting follow-up, they"
  },
  {
    "startMs": 177760,
    "text": "were spending less time with our"
  },
  {
    "startMs": 179120,
    "text": "customers and customer problems."
  },
  {
    "startMs": 182959,
    "text": "Most companies have separate systems for"
  },
  {
    "startMs": 184959,
    "text": "sales assist and productled growth. And"
  },
  {
    "startMs": 187519,
    "text": "this is actually also true for us."
  },
  {
    "startMs": 189920,
    "text": "Marketing run runs on one set of tools,"
  },
  {
    "startMs": 192720,
    "text": "sales on another, customer ops on a"
  },
  {
    "startMs": 195200,
    "text": "third. But all of them are looking at a"
  },
  {
    "startMs": 197760,
    "text": "customer independently and making"
  },
  {
    "startMs": 200239,
    "text": "decisions separately. So what we set out"
  },
  {
    "startMs": 203120,
    "text": "to build is a single decisioning system"
  },
  {
    "startMs": 206080,
    "text": "that spans self-s serve growth and sales"
  },
  {
    "startMs": 208560,
    "text": "assist. and it can help the customer"
  },
  {
    "startMs": 211440,
    "text": "decide the next step so that everything"
  },
  {
    "startMs": 214400,
    "text": "is cohesive."
  },
  {
    "startMs": 216400,
    "text": "Our vision is for this system to be"
  },
  {
    "startMs": 218159,
    "text": "programmable,"
  },
  {
    "startMs": 219840,
    "text": "proactive and continuous."
  },
  {
    "startMs": 223680,
    "text": "When we started, we were faced with some"
  },
  {
    "startMs": 226000,
    "text": "challenges. There was so no single"
  },
  {
    "startMs": 228319,
    "text": "source of truth. So customer data was"
  },
  {
    "startMs": 230560,
    "text": "spread across Salesforce, Gong,"
  },
  {
    "startMs": 233280,
    "text": "Outreach, Zoom Info and many more."
  },
  {
    "startMs": 236239,
    "text": "Product usage lived in Snowflake and a"
  },
  {
    "startMs": 238959,
    "text": "decade of the most important context"
  },
  {
    "startMs": 240799,
    "text": "lived in notes and meeting docs. Yes,"
  },
  {
    "startMs": 242799,
    "text": "our sales reps do use notion for that"
  },
  {
    "startMs": 244560,
    "text": "too. Notion employees were actively"
  },
  {
    "startMs": 248080,
    "text": "using MCPs and their own agents to solve"
  },
  {
    "startMs": 250799,
    "text": "problems already, but they were"
  },
  {
    "startMs": 253120,
    "text": "innovating within their own departments."
  },
  {
    "startMs": 255280,
    "text": "So it was single player mode or you"
  },
  {
    "startMs": 257759,
    "text": "could say here single department mode."
  },
  {
    "startMs": 260079,
    "text": "Marketing built tools for tool uh"
  },
  {
    "startMs": 261759,
    "text": "marketing sales built tools for sales"
  },
  {
    "startMs": 264479,
    "text": "and each tool served a tiny slice of"
  },
  {
    "startMs": 267120,
    "text": "that customer journey and it would make"
  },
  {
    "startMs": 269120,
    "text": "it really hard to create something that"
  },
  {
    "startMs": 271040,
    "text": "was more holistic."
  },
  {
    "startMs": 274240,
    "text": "When we tried to automate across all of"
  },
  {
    "startMs": 276240,
    "text": "that we hit some roadblocks. First data"
  },
  {
    "startMs": 279280,
    "text": "quality conflicting systems of records"
  },
  {
    "startMs": 282240,
    "text": "wrong contacts tied to different"
  },
  {
    "startMs": 283919,
    "text": "accounts. One bad mapping was enough to"
  },
  {
    "startMs": 286479,
    "text": "lose trust for sales rep."
  },
  {
    "startMs": 289040,
    "text": "Secondly, data latency. Every vendor"
  },
  {
    "startMs": 291759,
    "text": "added a hop and this lag was causing us"
  },
  {
    "startMs": 294720,
    "text": "to act on stale data and that meant we"
  },
  {
    "startMs": 297520,
    "text": "were automating on yesterday's world."
  },
  {
    "startMs": 300240,
    "text": "Third, and this is a big one, structured"
  },
  {
    "startMs": 302639,
    "text": "and unstructured data. The most"
  },
  {
    "startMs": 305360,
    "text": "important facts about a customer were"
  },
  {
    "startMs": 307360,
    "text": "left in notes like the champion just"
  },
  {
    "startMs": 310000,
    "text": "left or don't contact this customer"
  },
  {
    "startMs": 312320,
    "text": "again. Um or they're blocked illegal."
  },
  {
    "startMs": 315840,
    "text": "And so these are exactly the types of"
  },
  {
    "startMs": 317680,
    "text": "notes that help sales rep move forward"
  },
  {
    "startMs": 319840,
    "text": "and decide what to do next. And if an"
  },
  {
    "startMs": 322400,
    "text": "automation couldn't see it or process"
  },
  {
    "startMs": 324560,
    "text": "it, it could do something"
  },
  {
    "startMs": 326000,
    "text": "catastrophically wrong."
  },
  {
    "startMs": 329199,
    "text": "So our project team consisted of CX,"
  },
  {
    "startMs": 332080,
    "text": "RevOps, product, engineering, sales. And"
  },
  {
    "startMs": 336400,
    "text": "after brainstorming together, we all"
  },
  {
    "startMs": 338720,
    "text": "kept finding the same patterns"
  },
  {
    "startMs": 340320,
    "text": "underneath that complexity."
  },
  {
    "startMs": 343680,
    "text": "Whoops. Oh. Every workflow could be"
  },
  {
    "startMs": 346080,
    "text": "reduced to four questions. What do we"
  },
  {
    "startMs": 348479,
    "text": "know about the customer? What should"
  },
  {
    "startMs": 350800,
    "text": "happen next? How do we execute that"
  },
  {
    "startMs": 353440,
    "text": "safely? And did it work? That became our"
  },
  {
    "startMs": 357520,
    "text": "architecture."
  },
  {
    "startMs": 359520,
    "text": "So the system has four layers. Know a"
  },
  {
    "startMs": 362560,
    "text": "context layer we can trust about every"
  },
  {
    "startMs": 364720,
    "text": "customer. Decide, choose the single next"
  },
  {
    "startMs": 368000,
    "text": "best step for them. Third, fire act and"
  },
  {
    "startMs": 371840,
    "text": "fire a concrete action that could be a"
  },
  {
    "startMs": 374000,
    "text": "life cycle email um an inapp nudge or a"
  },
  {
    "startMs": 377120,
    "text": "task handed to a rep and then learn"
  },
  {
    "startMs": 380160,
    "text": "watch what happened and feed it back"
  },
  {
    "startMs": 382319,
    "text": "into the decisioning so that it's a"
  },
  {
    "startMs": 384080,
    "text": "loop."
  },
  {
    "startMs": 385840,
    "text": "But this architecture is missing"
  },
  {
    "startMs": 387919,
    "text": "something important."
  },
  {
    "startMs": 391520,
    "text": "The most important part is that humans"
  },
  {
    "startMs": 393840,
    "text": "and agents are operating on the same"
  },
  {
    "startMs": 396080,
    "text": "loop. Concretely, this means that the"
  },
  {
    "startMs": 399039,
    "text": "context needs to be displayed so that"
  },
  {
    "startMs": 401680,
    "text": "humans and agents can read and operate"
  },
  {
    "startMs": 403680,
    "text": "on it together."
  },
  {
    "startMs": 406000,
    "text": "So you can see that they're working in"
  },
  {
    "startMs": 407680,
    "text": "the same system, but they might have"
  },
  {
    "startMs": 409360,
    "text": "different roles. Agents do the"
  },
  {
    "startMs": 411440,
    "text": "repetitive work at scale like gathering"
  },
  {
    "startMs": 414080,
    "text": "context, researching, drafting"
  },
  {
    "startMs": 416639,
    "text": "recommendations, and writing artifacts."
  },
  {
    "startMs": 419680,
    "text": "Humans provide the judgment, adding"
  },
  {
    "startMs": 421919,
    "text": "nuance, deciding what to do next, and if"
  },
  {
    "startMs": 424560,
    "text": "a recommendation is correct. and owning"
  },
  {
    "startMs": 427120,
    "text": "the customer relationship."
  },
  {
    "startMs": 429520,
    "text": "We found that instead of building an AI"
  },
  {
    "startMs": 432000,
    "text": "layer on top of our business, we"
  },
  {
    "startMs": 434400,
    "text": "designed our architecture so that the"
  },
  {
    "startMs": 436240,
    "text": "agent can operate as another operator"
  },
  {
    "startMs": 438880,
    "text": "within the same system as humans."
  },
  {
    "startMs": 443039,
    "text": "Before we built the system, we made some"
  },
  {
    "startMs": 445120,
    "text": "choices about how we were going to"
  },
  {
    "startMs": 446639,
    "text": "implement this. Firstly, we deliberately"
  },
  {
    "startMs": 449680,
    "text": "chose not to let an agent talk directly"
  },
  {
    "startMs": 451919,
    "text": "to a customer. For sales assist"
  },
  {
    "startMs": 454639,
    "text": "workflows, humans stay in the loop by"
  },
  {
    "startMs": 457280,
    "text": "default and approve anything the agents"
  },
  {
    "startMs": 459759,
    "text": "do. The agents do the busy work. That"
  },
  {
    "startMs": 463039,
    "text": "decision that decision also has a"
  },
  {
    "startMs": 465199,
    "text": "security dimension too. If a prospect"
  },
  {
    "startMs": 468000,
    "text": "fills out a contact sales form online,"
  },
  {
    "startMs": 470639,
    "text": "we treat that as untrusted user input."
  },
  {
    "startMs": 473520,
    "text": "And so trust boundaries don't break"
  },
  {
    "startMs": 475440,
    "text": "down, especially because there is an"
  },
  {
    "startMs": 477360,
    "text": "agent in the middle."
  },
  {
    "startMs": 479520,
    "text": "Secondly, routing and eligibility became"
  },
  {
    "startMs": 482560,
    "text": "a first class primitive. Eligibility"
  },
  {
    "startMs": 485120,
    "text": "used to be scattered in all over the"
  },
  {
    "startMs": 487120,
    "text": "place. We had one check or rule in an"
  },
  {
    "startMs": 489440,
    "text": "email tool, another in sales and we"
  },
  {
    "startMs": 492720,
    "text": "pulled that all into one place so that"
  },
  {
    "startMs": 495199,
    "text": "these rules can be consumed across our"
  },
  {
    "startMs": 497520,
    "text": "codebase uh product sales engineering"
  },
  {
    "startMs": 501039,
    "text": "and these are like customer"
  },
  {
    "startMs": 502400,
    "text": "segmentations or signal signal"
  },
  {
    "startMs": 504879,
    "text": "definitions and then a single classifier"
  },
  {
    "startMs": 508000,
    "text": "will route what the customer should do"
  },
  {
    "startMs": 510319,
    "text": "and this will actually prevent double"
  },
  {
    "startMs": 512080,
    "text": "sends from our system and create very"
  },
  {
    "startMs": 514159,
    "text": "cohesive communication across"
  },
  {
    "startMs": 517039,
    "text": "And last but not least, we decided that"
  },
  {
    "startMs": 519440,
    "text": "it was very important to own the contact"
  },
  {
    "startMs": 521919,
    "text": "layer and we decided to rent everything"
  },
  {
    "startMs": 524240,
    "text": "else. Since we are a lean team, we will"
  },
  {
    "startMs": 527839,
    "text": "not build our own email vendor or"
  },
  {
    "startMs": 529680,
    "text": "enrichment services like Clay. We use"
  },
  {
    "startMs": 532080,
    "text": "Clay and we believe that we understood"
  },
  {
    "startMs": 534959,
    "text": "our customers the best. So we will not"
  },
  {
    "startMs": 538320,
    "text": "um give that away."
  },
  {
    "startMs": 541360,
    "text": "So let's get into what we built."
  },
  {
    "startMs": 544160,
    "text": "The first step was to gather a"
  },
  {
    "startMs": 545839,
    "text": "consolidated view of all of our"
  },
  {
    "startMs": 547760,
    "text": "customers."
  },
  {
    "startMs": 549279,
    "text": "Snowflake, which is our data warehouse,"
  },
  {
    "startMs": 551200,
    "text": "is where we compute this truth. We"
  },
  {
    "startMs": 553519,
    "text": "ingest data from all the vendors in our"
  },
  {
    "startMs": 555600,
    "text": "GTM stack to Snowflake. We run daily"
  },
  {
    "startMs": 558640,
    "text": "transforms and in some cases real time"
  },
  {
    "startMs": 561600,
    "text": "to produce a small set of modeled"
  },
  {
    "startMs": 563760,
    "text": "versioned entities and these are"
  },
  {
    "startMs": 565519,
    "text": "accounts, contacts, workspaces,"
  },
  {
    "startMs": 568480,
    "text": "eligibility, and facts. And this also"
  },
  {
    "startMs": 571040,
    "text": "has clear ownership of what teams or uh"
  },
  {
    "startMs": 573519,
    "text": "tools they come from and like"
  },
  {
    "startMs": 575040,
    "text": "timestamps."
  },
  {
    "startMs": 576720,
    "text": "Dynamob is our key value store and it's"
  },
  {
    "startMs": 579279,
    "text": "where we compute our truth or serve our"
  },
  {
    "startMs": 581760,
    "text": "truth. We publish a denormalized key"
  },
  {
    "startMs": 584399,
    "text": "addressable profile that agents can"
  },
  {
    "startMs": 587040,
    "text": "quickly query in milliseconds with no"
  },
  {
    "startMs": 589360,
    "text": "joins. We also persist agent uh"
  },
  {
    "startMs": 593040,
    "text": "persisted uh or generated artifacts and"
  },
  {
    "startMs": 595760,
    "text": "these are research snippets, summarized"
  },
  {
    "startMs": 597839,
    "text": "notes, rolling summaries and these"
  },
  {
    "startMs": 600560,
    "text": "unstructured data are also keyed by the"
  },
  {
    "startMs": 603200,
    "text": "same ids so that downstream systems can"
  },
  {
    "startMs": 605920,
    "text": "read all of this in one shot."
  },
  {
    "startMs": 610240,
    "text": "So this data was normalized and of"
  },
  {
    "startMs": 612880,
    "text": "course we brought it into notion so that"
  },
  {
    "startMs": 614959,
    "text": "we could work with structured and"
  },
  {
    "startMs": 616560,
    "text": "unstructured data at the same time. So"
  },
  {
    "startMs": 619040,
    "text": "some of the data I showed you in the"
  },
  {
    "startMs": 620480,
    "text": "boxes earlier, there's like product"
  },
  {
    "startMs": 622160,
    "text": "usage data, there's activity log from"
  },
  {
    "startMs": 624959,
    "text": "across our vendor stack, and then we"
  },
  {
    "startMs": 627200,
    "text": "also have like unstructured data that"
  },
  {
    "startMs": 629040,
    "text": "like I mentioned that is most important"
  },
  {
    "startMs": 631120,
    "text": "for sales context with research reports"
  },
  {
    "startMs": 634160,
    "text": "and notes."
  },
  {
    "startMs": 636240,
    "text": "And this turned out to be powerful for"
  },
  {
    "startMs": 638160,
    "text": "two reasons. First, our internal GTM"
  },
  {
    "startMs": 642000,
    "text": "teams didn't need to jump between many"
  },
  {
    "startMs": 644320,
    "text": "tools anymore. They could use notion"
  },
  {
    "startMs": 646720,
    "text": "itself, a tool they were already using"
  },
  {
    "startMs": 648959,
    "text": "to explore context, investigate"
  },
  {
    "startMs": 651519,
    "text": "investigate accounts, answer questions,"
  },
  {
    "startMs": 654000,
    "text": "and they could even take actions like"
  },
  {
    "startMs": 656240,
    "text": "sending to Nooks or um sending to"
  },
  {
    "startMs": 658880,
    "text": "outreach. This is a tool that they were"
  },
  {
    "startMs": 661200,
    "text": "very familiar with, and they didn't need"
  },
  {
    "startMs": 663040,
    "text": "to open seven tabs anymore. Secondly,"
  },
  {
    "startMs": 666160,
    "text": "because we weren't building an AI layer,"
  },
  {
    "startMs": 668720,
    "text": "um humans, workflows, and agents could"
  },
  {
    "startMs": 671040,
    "text": "all operate on the same source of truth."
  },
  {
    "startMs": 673600,
    "text": "In a very literal sense, we are using"
  },
  {
    "startMs": 675920,
    "text": "notion to grow notion."
  },
  {
    "startMs": 680480,
    "text": "The next primitive we decided that we"
  },
  {
    "startMs": 682800,
    "text": "needed was a way to turn customer events"
  },
  {
    "startMs": 685440,
    "text": "into actions. The unit here is a signal."
  },
  {
    "startMs": 689040,
    "text": "A signal is a single customer event"
  },
  {
    "startMs": 691680,
    "text": "that's important enough to change what"
  },
  {
    "startMs": 693519,
    "text": "should happen next for a customer. Some"
  },
  {
    "startMs": 696079,
    "text": "are userdriven like a customer hitting"
  },
  {
    "startMs": 698160,
    "text": "their a AI limit or maybe they reached"
  },
  {
    "startMs": 701360,
    "text": "out to contact contact sales. But some"
  },
  {
    "startMs": 704079,
    "text": "of them are not user initiated at all"
  },
  {
    "startMs": 706079,
    "text": "which are these external signal examples"
  },
  {
    "startMs": 708079,
    "text": "I listed here like company raising"
  },
  {
    "startMs": 710399,
    "text": "funding, hiring signals or shift in"
  },
  {
    "startMs": 713040,
    "text": "their tech stack. Those external signals"
  },
  {
    "startMs": 715680,
    "text": "are what allowed us to be proactive"
  },
  {
    "startMs": 717600,
    "text": "instead of reactive."
  },
  {
    "startMs": 720880,
    "text": "So this is the signal service that"
  },
  {
    "startMs": 723760,
    "text": "watches the customer profile, decides"
  },
  {
    "startMs": 726240,
    "text": "whether a single action is available,"
  },
  {
    "startMs": 729920,
    "text": "decides who should own that action, and"
  },
  {
    "startMs": 732399,
    "text": "then it emits a concrete task following"
  },
  {
    "startMs": 734880,
    "text": "the architecture I described before. And"
  },
  {
    "startMs": 737360,
    "text": "this task could be for a human or an"
  },
  {
    "startMs": 739680,
    "text": "agent. If there's a task for sales rep,"
  },
  {
    "startMs": 742480,
    "text": "that actually just lands in their notion"
  },
  {
    "startMs": 744240,
    "text": "database and they can quickly view it"
  },
  {
    "startMs": 746160,
    "text": "and act on it."
  },
  {
    "startMs": 749279,
    "text": "What's what's interesting about the way"
  },
  {
    "startMs": 750880,
    "text": "we built our GTM systems is that if"
  },
  {
    "startMs": 753279,
    "text": "there is no signal about a customer, the"
  },
  {
    "startMs": 755600,
    "text": "marketing component of our system kicks"
  },
  {
    "startMs": 757600,
    "text": "in. We have a predictive engine that"
  },
  {
    "startMs": 760560,
    "text": "will recommend product features most"
  },
  {
    "startMs": 762480,
    "text": "relevant for that customer and it will"
  },
  {
    "startMs": 764480,
    "text": "send out life cycle emails and inapp"
  },
  {
    "startMs": 766880,
    "text": "nudges or multi-channel communication to"
  },
  {
    "startMs": 770000,
    "text": "drive a customer towards adoption"
  },
  {
    "startMs": 771839,
    "text": "automatically."
  },
  {
    "startMs": 775279,
    "text": "So diving deep into a small slice of"
  },
  {
    "startMs": 778480,
    "text": "what happens when we decide what action"
  },
  {
    "startMs": 781519,
    "text": "should be emitted. Um this is for like"
  },
  {
    "startMs": 784320,
    "text": "the sales workflow and we shadowed our"
  },
  {
    "startMs": 787040,
    "text": "best reps to capture something that was"
  },
  {
    "startMs": 789279,
    "text": "the most repetitive part of our job and"
  },
  {
    "startMs": 791680,
    "text": "encoded it as a durable multi- aent"
  },
  {
    "startMs": 794480,
    "text": "workflow. Every signal becomes a"
  },
  {
    "startMs": 797200,
    "text": "workflow on temporal which is something"
  },
  {
    "startMs": 799200,
    "text": "we rent and a single run will touch"
  },
  {
    "startMs": 802160,
    "text": "enrichment web search draft generation"
  },
  {
    "startMs": 805360,
    "text": "and more. Each of these is a network"
  },
  {
    "startMs": 808240,
    "text": "call that could fail or rate limit. And"
  },
  {
    "startMs": 810800,
    "text": "so temporal lets us focus on writing the"
  },
  {
    "startMs": 813920,
    "text": "sequential logic for our GTM use cases"
  },
  {
    "startMs": 817279,
    "text": "while it will handle the retries, ddupes"
  },
  {
    "startMs": 820639,
    "text": "um and handling and going back to"
  },
  {
    "startMs": 822959,
    "text": "exactly where failures left off. and one"
  },
  {
    "startMs": 825760,
    "text": "malformed transcript can't take down the"
  },
  {
    "startMs": 828399,
    "text": "whole batch which was really important"
  },
  {
    "startMs": 830079,
    "text": "to us. So an example of a cold outbound"
  },
  {
    "startMs": 833839,
    "text": "signal for us will have a research sub"
  },
  {
    "startMs": 836399,
    "text": "agent do concurrent researches then"
  },
  {
    "startMs": 839680,
    "text": "it'll draft like an email and those"
  },
  {
    "startMs": 842560,
    "text": "emails uh there should be three of them"
  },
  {
    "startMs": 844639,
    "text": "so they're scored and then a review"
  },
  {
    "startMs": 847199,
    "text": "agent will pick the highest scoring one"
  },
  {
    "startMs": 849199,
    "text": "and make any updates if necessary. And"
  },
  {
    "startMs": 851760,
    "text": "this also operates on a loop um so that"
  },
  {
    "startMs": 854240,
    "text": "the email drafts are improved. And then"
  },
  {
    "startMs": 856639,
    "text": "when it's ready, this email draft will"
  },
  {
    "startMs": 858560,
    "text": "land in the sales task that is available"
  },
  {
    "startMs": 861600,
    "text": "for for them to act on. Um for more"
  },
  {
    "startMs": 864720,
    "text": "reactive signals after a follow-up call,"
  },
  {
    "startMs": 867519,
    "text": "the Gong transcript will come in. Our"
  },
  {
    "startMs": 869920,
    "text": "agent will parse the transcript and then"
  },
  {
    "startMs": 872959,
    "text": "um again it will extract the critical"
  },
  {
    "startMs": 875360,
    "text": "sales medpic data uh metrics economic"
  },
  {
    "startMs": 878959,
    "text": "buyer decision criteria plan and"
  },
  {
    "startMs": 882160,
    "text": "champion and draft a grounded followup"
  },
  {
    "startMs": 884880,
    "text": "for that. Every LLM step is traced so"
  },
  {
    "startMs": 888320,
    "text": "that we can evaluate quality and improve"
  },
  {
    "startMs": 890560,
    "text": "over time."
  },
  {
    "startMs": 893600,
    "text": "The third layer is what turns this"
  },
  {
    "startMs": 895519,
    "text": "automation into a system that"
  },
  {
    "startMs": 897600,
    "text": "self-improves. Every action is a"
  },
  {
    "startMs": 900560,
    "text": "decision log and every outcome threads"
  },
  {
    "startMs": 903519,
    "text": "back to the decision that caused it. So"
  },
  {
    "startMs": 906000,
    "text": "the naive version of this is a data"
  },
  {
    "startMs": 908160,
    "text": "analyst coming in and trying to"
  },
  {
    "startMs": 910240,
    "text": "understand if the output of this could"
  },
  {
    "startMs": 912320,
    "text": "be better. The rebuilt version of this"
  },
  {
    "startMs": 914560,
    "text": "is wiring our engagement history back"
  },
  {
    "startMs": 916880,
    "text": "into the decision layer so that the"
  },
  {
    "startMs": 919440,
    "text": "system decides whether or not to"
  },
  {
    "startMs": 921440,
    "text": "continue a thread, advance to the next"
  },
  {
    "startMs": 923600,
    "text": "step or pivot. The system will continue"
  },
  {
    "startMs": 927360,
    "text": "to do that with the life cycle message"
  },
  {
    "startMs": 929279,
    "text": "performance history as well. So these"
  },
  {
    "startMs": 931680,
    "text": "verification loops are really critical"
  },
  {
    "startMs": 934240,
    "text": "so that the system can self-heal and"
  },
  {
    "startMs": 936320,
    "text": "continuously improve."
  },
  {
    "startMs": 938959,
    "text": "Let's see how an agent and human work"
  },
  {
    "startMs": 940959,
    "text": "together in this shared customer view."
  },
  {
    "startMs": 943199,
    "text": "In the customer view, a rep can come"
  },
  {
    "startMs": 945440,
    "text": "here and see the product usage, the"
  },
  {
    "startMs": 947519,
    "text": "recent activity and get an answer using"
  },
  {
    "startMs": 950000,
    "text": "that same data. They used to find all of"
  },
  {
    "startMs": 952880,
    "text": "this across many different tabs and now"
  },
  {
    "startMs": 955279,
    "text": "they can just come here each day. The"
  },
  {
    "startMs": 957680,
    "text": "rep can ask an agent and the agent will"
  },
  {
    "startMs": 960160,
    "text": "reply uh querying our context layer. We"
  },
  {
    "startMs": 964399,
    "text": "can also use notion custom agents which"
  },
  {
    "startMs": 966959,
    "text": "are sharable across companies to access"
  },
  {
    "startMs": 969519,
    "text": "this data context for recurring"
  },
  {
    "startMs": 971440,
    "text": "automated workflows."
  },
  {
    "startMs": 974800,
    "text": "In the task view, a rep starts their day"
  },
  {
    "startMs": 977040,
    "text": "with an already prioritized task box and"
  },
  {
    "startMs": 980800,
    "text": "they already know how to move forward"
  },
  {
    "startMs": 982800,
    "text": "with accounts and contacts. And the"
  },
  {
    "startMs": 985519,
    "text": "email draft for an outreach task is"
  },
  {
    "startMs": 987839,
    "text": "already pre-ressearched and available"
  },
  {
    "startMs": 989920,
    "text": "for them to review. The human is still"
  },
  {
    "startMs": 992720,
    "text": "in the loop and actually adds their own"
  },
  {
    "startMs": 994800,
    "text": "judgment and taste um and sales secret"
  },
  {
    "startMs": 997600,
    "text": "sauce, but they're no longer starting"
  },
  {
    "startMs": 999519,
    "text": "from a blank sta slate. And this does"
  },
  {
    "startMs": 1002880,
    "text": "more than one help one rep be"
  },
  {
    "startMs": 1005360,
    "text": "productive. Um our goal is actually to"
  },
  {
    "startMs": 1008399,
    "text": "raise the floor for the entire team. So"
  },
  {
    "startMs": 1011120,
    "text": "a Neil sales rep coming in, they can"
  },
  {
    "startMs": 1013920,
    "text": "learn the notion sales process,"
  },
  {
    "startMs": 1016320,
    "text": "understand what signals are important to"
  },
  {
    "startMs": 1018480,
    "text": "look for, um, understand which playbooks"
  },
  {
    "startMs": 1021519,
    "text": "are effective, and basically know what"
  },
  {
    "startMs": 1024400,
    "text": "good followup looks like. Reps who can"
  },
  {
    "startMs": 1027679,
    "text": "are ramping can still learn from the"
  },
  {
    "startMs": 1029520,
    "text": "patterns of the strongest reps without"
  },
  {
    "startMs": 1031600,
    "text": "needing every lesson to be passed down"
  },
  {
    "startMs": 1033760,
    "text": "manually."
  },
  {
    "startMs": 1036480,
    "text": "So, one of the questions that we came"
  },
  {
    "startMs": 1038720,
    "text": "across along every step of the way is a"
  },
  {
    "startMs": 1041120,
    "text": "classic question. Do we build or buy?"
  },
  {
    "startMs": 1045039,
    "text": "And it's very tempting and trendy to say"
  },
  {
    "startMs": 1047679,
    "text": "build everything. But what we found is"
  },
  {
    "startMs": 1050160,
    "text": "that there are still key areas to build"
  },
  {
    "startMs": 1052559,
    "text": "and rent access to at every single"
  },
  {
    "startMs": 1055760,
    "text": "layer. Internal agents are actually"
  },
  {
    "startMs": 1058640,
    "text": "cheaper and faster to build than most"
  },
  {
    "startMs": 1060720,
    "text": "people assume. And so since we have the"
  },
  {
    "startMs": 1063360,
    "text": "most data on our con on our data model"
  },
  {
    "startMs": 1066000,
    "text": "um we build it there first and then we"
  },
  {
    "startMs": 1068799,
    "text": "uh rented the generalizable parts later."
  },
  {
    "startMs": 1071679,
    "text": "So for us the build versus buy as a per"
  },
  {
    "startMs": 1073919,
    "text": "layer decision. We will not build a lot"
  },
  {
    "startMs": 1076320,
    "text": "of these tools like orchestration,"
  },
  {
    "startMs": 1078480,
    "text": "email, CRM. Um vendors do that really"
  },
  {
    "startMs": 1082080,
    "text": "well. We refuse to outsource the context"
  },
  {
    "startMs": 1084880,
    "text": "layer because that's where our edge is."
  },
  {
    "startMs": 1087360,
    "text": "a generic tool can't capture all of our"
  },
  {
    "startMs": 1089600,
    "text": "esoteric data models or workflows and we"
  },
  {
    "startMs": 1092880,
    "text": "do not want that context layer to be"
  },
  {
    "startMs": 1094640,
    "text": "something we can't um debug"
  },
  {
    "startMs": 1098799,
    "text": "and so as I mentioned before that"
  },
  {
    "startMs": 1101200,
    "text": "context layer is a notion it's built off"
  },
  {
    "startMs": 1104000,
    "text": "of plain markdown a language that agents"
  },
  {
    "startMs": 1106960,
    "text": "are fluent in and we have databases and"
  },
  {
    "startMs": 1110160,
    "text": "hierarchies that they can navigate"
  },
  {
    "startMs": 1111679,
    "text": "easily at the same time this is well"
  },
  {
    "startMs": 1114160,
    "text": "designed for human um so this is what"
  },
  {
    "startMs": 1117039,
    "text": "lets our engineers, agents, and GTM work"
  },
  {
    "startMs": 1120240,
    "text": "off the same context. And this has all"
  },
  {
    "startMs": 1122960,
    "text": "the data synced across sources."
  },
  {
    "startMs": 1128400,
    "text": "Ultimately, the reason to see if we can"
  },
  {
    "startMs": 1130559,
    "text": "do all of this is to see if we could get"
  },
  {
    "startMs": 1132320,
    "text": "a better throughput on deals. It's very"
  },
  {
    "startMs": 1135039,
    "text": "early for us. We're still building this"
  },
  {
    "startMs": 1136960,
    "text": "out, but the initial signs are"
  },
  {
    "startMs": 1138400,
    "text": "promising. In the last 13 weeks, we are"
  },
  {
    "startMs": 1140960,
    "text": "already seeing enterprise reps have"
  },
  {
    "startMs": 1143120,
    "text": "increased qualification or qualified"
  },
  {
    "startMs": 1145440,
    "text": "opportunities. And on the life cycle"
  },
  {
    "startMs": 1147440,
    "text": "marketing side, users who received"
  },
  {
    "startMs": 1149600,
    "text": "contextaware recommendations were 63%"
  },
  {
    "startMs": 1152640,
    "text": "more likely to take the next step. This"
  },
  {
    "startMs": 1155039,
    "text": "is the early days with a lot more"
  },
  {
    "startMs": 1156480,
    "text": "features we want to build, but our"
  },
  {
    "startMs": 1158400,
    "text": "thesis that us building a single system"
  },
  {
    "startMs": 1160799,
    "text": "on no, decide, act, and learn seems to"
  },
  {
    "startMs": 1164480,
    "text": "be right so far."
  },
  {
    "startMs": 1166960,
    "text": "A few key takeaways. um from entering"
  },
  {
    "startMs": 1169919,
    "text": "entering this world as an engineer in"
  },
  {
    "startMs": 1171840,
    "text": "the last six months is that before you"
  },
  {
    "startMs": 1174080,
    "text": "build shadow your best human. I talked"
  },
  {
    "startMs": 1177039,
    "text": "to many sales reps and when I opened uh"
  },
  {
    "startMs": 1179679,
    "text": "when they opened their computers I saw"
  },
  {
    "startMs": 1181360,
    "text": "how many tabs and tools they were"
  },
  {
    "startMs": 1183200,
    "text": "navigating between and that was a chaos"
  },
  {
    "startMs": 1185919,
    "text": "but it was also the spec and so if you"
  },
  {
    "startMs": 1188240,
    "text": "encode a mediocre process you get a"
  },
  {
    "startMs": 1190400,
    "text": "mediocre agent. Start with the most"
  },
  {
    "startMs": 1193039,
    "text": "legible workflow. That's the one that's"
  },
  {
    "startMs": 1195440,
    "text": "documented and repeated. And let humans"
  },
  {
    "startMs": 1198080,
    "text": "stay in the loop on where there are"
  },
  {
    "startMs": 1199520,
    "text": "risky possibilities."
  },
  {
    "startMs": 1202320,
    "text": "Model GTM as primitives, entities,"
  },
  {
    "startMs": 1205440,
    "text": "context, triggers, actions, eligibility"
  },
  {
    "startMs": 1208559,
    "text": "rules, and the alien world becomes a"
  },
  {
    "startMs": 1210880,
    "text": "system you can engineer. Last but not"
  },
  {
    "startMs": 1213679,
    "text": "least, be headless by default and design"
  },
  {
    "startMs": 1216080,
    "text": "for agents as operators and not just"
  },
  {
    "startMs": 1218240,
    "text": "co-pilots. If humans and agents can't"
  },
  {
    "startMs": 1221039,
    "text": "read from the same substrate, you're"
  },
  {
    "startMs": 1223760,
    "text": "basically building two systems that will"
  },
  {
    "startMs": 1225600,
    "text": "eventually drift apart. For us, that"
  },
  {
    "startMs": 1228159,
    "text": "layer is notion."
  },
  {
    "startMs": 1230640,
    "text": "Right now, humans are the primary"
  },
  {
    "startMs": 1232240,
    "text": "consumer of GTM data and agents are"
  },
  {
    "startMs": 1234559,
    "text": "helping at the edges. Soon, agents will"
  },
  {
    "startMs": 1237360,
    "text": "become primary first class consumers"
  },
  {
    "startMs": 1239919,
    "text": "within the system, moving from drafting"
  },
  {
    "startMs": 1242240,
    "text": "to acting within guard rails by creating"
  },
  {
    "startMs": 1245600,
    "text": "the best context and substrate for"
  },
  {
    "startMs": 1247600,
    "text": "humans and agents to collaborate"
  },
  {
    "startMs": 1249280,
    "text": "together. Now, you're setting up your"
  },
  {
    "startMs": 1251200,
    "text": "team to sprint faster."
  },
  {
    "startMs": 1253600,
    "text": "Um, yeah, and feel free to contact me if"
  },
  {
    "startMs": 1256320,
    "text": "you guys want to ask more questions."
  },
  {
    "startMs": 1272679,
    "text": ">> [music]"
  }
]
