Dave Killeen is Field CPO at Pendo, a $2.6B company. He monitors 45 active enterprise deals. He doesn’t attend pipeline reviews. He built a system with Claude Code that reads meeting transcripts, checks CRM health scores, scores each deal, and sends him a Slack message with what needs his attention. Aakash Gupta documented the case in detail.

One command. 45 deals. Zero review meetings.

Killeen didn’t automate a team process — he automated himself. And that’s what makes this different from nearly every enterprise AI case we’ve covered so far.

45 Deals, One Command, Zero Pipeline Reviews

Killeen’s system has three pieces working together.

Input: Granola transcribes every sales meeting automatically. This isn’t passive recording — Granola generates structured notes including commitments, objections, and next steps. Each transcript feeds the system without manual intervention.

Intelligence: Clary provides health scores for each account — engagement metrics, sales cycle progress, churn risk. Claude Code takes these two sources (transcripts + health scores) and applies custom scoring that Killeen defined according to his own criteria: which signals indicate a deal at risk, which patterns predict close, which accounts need CPO intervention vs. ones the sales team can handle alone.

Output: The system generates two things. First, a prioritized list of deals requiring attention — not all 45, but the 5 to 7 that have something Killeen needs to see. Second, draft Slack messages — not generic ones, but contextualized with the specific information from each deal.

The result: Killeen reviews the output, adjusts what needs adjusting, and acts. The weekly pipeline review — that 90-minute meeting where each deal manager reports status — stopped being necessary for him.

He didn’t eliminate the meeting for his team. He eliminated it for himself. His team still does their work. Killeen simply already knows what he needs to know before anyone tells him.

The Architecture Behind the System

What Killeen built can be described in three layers that apply beyond his specific case:

	Traditional pipeline review	Personal AI system
Frequency	Weekly or biweekly	Continuous (whenever new data arrives)
Data sources	What each person remembers to report	Transcripts + CRM + automated scoring
Latency	5-14 days between signal and action	Same day
Bias	Filtered by who reports and how	Direct data, criteria defined by the executive
Time cost	60-90 min/week + preparation	Minutes of review per day

The input layer captures data that already exists — meetings that already happened, metrics the CRM already tracks. It doesn’t generate new information. It transforms it from scattered formats (audio, dashboards, emails) into a format the intelligence layer can process.

The intelligence layer is where the executive’s judgment lives in code. Killeen didn’t tell Claude Code “analyze these deals.” He told it “a deal at risk looks like this, a deal that needs my intervention looks like this, and a deal my team can handle has these characteristics.” Defining those criteria is the real work — and it’s work only the executive can do.

The output layer isn’t a dashboard. It’s prepared action. Messages ready to send, not charts to interpret. The difference matters: a dashboard requires the executive to spend time reading and interpreting. A draft Slack message requires them to decide whether to send or modify. The second is orders of magnitude faster.

Not Team Automation — Intermediary Elimination

When we talk about “AI for business,” we usually mean automating processes that teams execute: customer onboarding, report generation, ticket classification. Those are valuable projects. But they’re team projects.

What Killeen did is different. He removed the chain of intermediaries between raw data and his own judgment. The traditional pipeline review works like this: data lives in the CRM. Account managers interpret it. They prepare a summary. They present it in the meeting. The CPO listens, asks questions, and forms an opinion. There are at least three layers of human filtering between the original data point and the executive’s decision.

Killeen’s system compresses that to zero filtering layers. Data flows directly from the CRM and transcripts to the CPO’s system, processed according to criteria he defined himself.

He didn’t replace anyone. His account managers still do their jobs. What he eliminated was the information telephone game — that chain where each link adds interpretation, omits details, and filters based on what they think is relevant.

In our experience at IQ Source, the biggest signal loss in B2B companies doesn’t happen in the systems — it happens in the layers of human summarization between systems and decisions. An executive who receives data filtered through three people has one version of reality. An executive who defines their own filtering criteria and applies them directly to the data has another.

Killeen Isn’t the First — But He’s the Most Specific

The pattern of executives using AI directly didn’t start with Pendo. But each prior case solves a different problem.

Tobi Lutke at Shopify ran autoresearch on the Liquid engine — production infrastructure code optimization. It’s direct AI use by an executive, yes, but applied to engineering, not recurring business decisions. Lutke used an agent for a specific project with a start and end. Killeen built a permanent system that runs every day.

Jason Calacanis with Ultron automated team operations — deal flow, due diligence, portfolio management at his investment fund. It’s closer to Killeen’s case, but Ultron operates as the team’s system, not the leader’s personal system. The difference: Ultron replaces tasks that analysts used to do. Killeen’s system replaces the meeting where he received the information.

And the evolution of business intelligence dashboards created better ways to visualize data for the entire organization. But a dashboard is a shared tool. Killeen’s system is personal — calibrated to his criteria, connected to his channels, optimized for his decisions.

What makes the Pendo case special is its specificity. It’s not “I use AI to be more productive.” It’s “I built a system connecting Granola + Clary + Claude Code to monitor 45 deals against my criteria and send me Slack drafts.” That concreteness is what makes it replicable.

What Kind of Executive Can Do This Today

This doesn’t apply to every executive or every type of decision. Three conditions need to align.

First: a repetitive decision cadence with digital data. Pipeline reviews, forecast updates, budget approvals, portfolio reviews — any recurring meeting where the executive needs an update on the status of multiple items. If the decision happens once a quarter and depends on qualitative factors that aren’t in any system, a personal agent doesn’t help much.

Second: decisions requiring synthesis from multiple sources. If all the information you need is in a single dashboard, you don’t need an AI system — you need to look at the dashboard. The value appears when you need to cross-reference meeting transcripts with CRM data with emails with product metrics. That synthesis is what a human does slowly and a system does fast.

Third: willingness to define what “good output” means. This is the condition most people underestimate. Killeen had to articulate his own prioritization criteria — what makes a deal critical, what signals indicate risk, what health score threshold triggers an alert. Most executives apply those criteria intuitively. Codifying them requires an introspection exercise many haven’t done.

Where doesn’t it work yet? Decisions that depend on personal relationships — knowing that your client’s VP just had a team change and is under pressure, or that there’s internal tension on the account that doesn’t show in any data. Creative strategy decisions — positioning a new product, redefining a market narrative. Those still require human intuition that no system captures yet.

What Changes When Data Arrives Unfiltered

When an executive shifts from receiving filtered information to receiving data processed by their own criteria, the effects go beyond saving themselves a meeting.

Earlier intervention. In a traditional pipeline review, an at-risk deal gets discussed Thursday in the meeting. The CPO learns about it, asks for context, and acts Friday or the following Monday. With a personal system, the risk signal arrives the same day it appears in the data. Killeen can message the account manager on Tuesday about a change detected Monday. The difference between intervening Tuesday and the following Thursday can be the difference between retaining and losing an enterprise account.

Cross-deal pattern detection. An account manager knows their 8 accounts. The CPO, in theory, knows all 45. But in practice, information arrives fragmented — one deal at a time, one summary at a time. Killeen’s system processes all 45 deals every time it runs. It can detect that three accounts in the same segment show the same risk signals simultaneously — a pattern that individual meetings would miss.

Consistent criteria application. Humans are inconsistent. The same deal presented Monday morning gets different attention than the same deal presented Thursday at 4pm. A system applies the same criteria every time. No recency bias, no meeting fatigue, no distraction by whichever deal the most charismatic account manager presents first.

In our experience with B2B clients, many executives discover, when trying to define their criteria for a system, that they didn’t have them as clear as they thought. It ends up being an exercise in clarifying their own decision criteria — and that alone justifies the effort even if the system never gets built.

Building Your First Personal AI System

The code isn’t the barrier. Claude Code is available. Granola has an API. Clary has an API. Any developer can connect the pieces in a couple of days.

The hard part is knowing which process to attack first.

The question that works: “which recurring meeting do I walk into always thinking ‘I should have known this before sitting down’?”

It might be the pipeline review. It might be the forecast meeting. It might be the project portfolio review. It might be the budget approval session. What they all share: the executive needs an update before they can make decisions, and that update arrives during the meeting itself — not before.

That meeting is your starting point. The information you need before walking in tells you what data to capture. The criteria you apply during the meeting tell you what scoring to define. The actions you take afterward tell you what outputs to generate.

Killeen didn’t start by saying “I want an AI system.” He started by saying “I don’t need to be in that meeting if I already have the information.”

Tell us which recurring meeting makes you think “I should have known this before walking in.” We’ll map the data sources, evaluate feasibility, and tell you whether a personal system makes sense for your case. No commitment — just a 20-minute technical conversation.

Schedule a conversation →