Conversation
Highlights

When Tribal Knowledge Walks Out the Door: How PredictAP Rebuilt GTM After Network Depletion

Chrisa worked in accounts payable at Colony Capital for 30 years. When she retired, a $60 billion private equity real estate firm discovered they’d just lost their ability to process invoices efficiently.

The problem wasn’t system failure. Colony had recently completed a major technology modernization—automated payments, updated workflows, modern infrastructure. The gap was knowledge. Chrisa understood how invoice coding actually worked across thousands of entities, tax structures, and tenant relationships. That expertise wasn’t documented anywhere.

In a recent episode of BUILDERS, David Stifter, Founder & CEO of PredictAP, explained the operational reality his former employer faced. An invoice addressed to Colony Capital rarely meant Colony paid it directly. The coding split 37 ways across Luxembourg tax blockers, SPVs holding entities, and property-level allocations. “If you are a tenant in an office building and someone’s cleaning the lobby, well, you’re actually paying for that as a tenant. It’s called common area maintenance,” David explained. Office buildings allocated differently than industrial properties, which used triple net leases passing costs through entirely. Every asset class carried nuanced rules that existed only in Chrisa’s experience.

David, then Colony’s head of technology, spent a year evaluating solutions. RPA and robotic process automation couldn’t handle the contextual logic. Template-based approaches failed when exceptions outnumbered standards. Vendors touting early AI—mostly OCR extracting text—missed the fundamental challenge: the rules weren’t written on invoices.

Then David recognized the training data sitting in Colony’s own systems. “Every time you’ve paid a bill, you have the image, you have the coding. That’s nice structure, that’s good annotation. So we have structured annotated data.” Six years ago, before LLMs existed, annotation represented the bottleneck in machine learning applications. Colony’s historical invoice-coding pairs solved it.

David describes the pattern recognition challenge: “You walk into the coffee shop that you go to every Monday and they know you. There’s like, Brett, here’s your usual, right? You don’t have to explain. They know it’s Monday, it’s cold, I want my hot chocolate, right? Friday, I’m going out, I want my Americana.” Context drives decisions through implicit signals—exactly like invoice coding.

The third requirement: human review as feedback mechanism. “As we’re finding out now with AI is incredibly important, right? It’s very powerful, but it will lie to you. It wants to make you happy,” David noted. Maintaining segregation of duties—basic accounting controls requiring payment review—created the feedback loop that improved model accuracy over time.

At 40, with young children, David left his executive role to build PredictAP.

Network Reality: Who Actually Picks Up When You Call

Early customers came from David’s professional network, but not how most founders expect. Katie Yelzneb had audited Colony’s controls as an Ernst & Young partner. “She would give me the business every year, going through my controls and my process,” David recalled. They maintained contact after Katie moved to Bridge Investment Group as CFO—not for business reasons, just consistent helpfulness on technical problems.

Five years later, when David needed beta customers: “I said, Katie, I have this idea. I want to try to do this. And she saw it before back at the Colony days. And so she’s like, yeah, I’ll give you a few hundred invoices. Let’s see how it goes.” That test evolved into one of PredictAP’s largest relationships, now processing tens of thousands of invoices monthly.

David also learned which networks were transactional illusions. Two early investors provided the contrast. One, a former acquisitions executive controlling billions in deal flow, promised extensive introductions across major firms. “Not a single callback,” David said. The person later experienced what he describes as “an existential crisis” realizing “my network was based upon my ability to have a massive checkbook behind me.”

An asset manager—mid-hierarchy, less external validation—”has been our biggest source of leads. Hundreds.” The difference: relationships built through years of solving problems together rather than wielding purchasing power.

“You find out who your friends are,” David noted. The transition from CTO to vendor founder exposes network quality instantly.

The Two-Year Wall and Complete GTM Rebuild

PredictAP ran founder-led sales for two years. Then: “We hit that wall.” No more warm introductions. No more network connections. Sales stopped entirely.

The contrast between warm and cold outreach became stark. Calls with mutual connections were “very to the point”—trust established, conversations efficient, deals moving quickly. Cold outreach “starts hostile or it starts skeptical”—requiring fundamentally different trust-building approaches and much longer cycles.

David brought in Steve Farrell as head of sales, but his hiring criteria ignored conventional wisdom. “I was less concerned about industry experience. I knew I could bring that credibility,” he explained. Real estate domain knowledge mattered less than one specific capability: “Someone that’s gone through it from that, you know, 3, 4, 5 million to 20 million sales and startup.”

The rebuild was total. “It was a stop, it was a rebuild. It’s okay. Here’s how you start that real funnel now, which is a completely different process.” They added BDRs and two AEs. Implemented systematic content generation—blogs, conference presence, consistent travel schedule. Started testing channels: Reddit ads, LinkedIn campaigns, Google ads, searching for repeatable lead generation.

“You have to invest in it, both from just straight dollars and spending, but also an infrastructure and tooling to allow you to do a lot of things,” David emphasized. The infrastructure investment scared him as a founder watching every dollar. “You’re not going to get a dollar right back. This is not, we’re not a B2C customer, we’re B2B. And the sales cycle is very long.”

PredictAP recently closed their first million-dollar quarter. But David learned the psychological challenge of repeatable systems: “Q4 comes. I’m like, oh geez, like who’s signing, who’s pushing off, Thanksgiving’s here, Christmas is here. What’s this quarter gonna do?” The celebration resets to zero every 90 days.

Customer Sophistication Sequencing

David’s initial customers came from his technology network—CTOs and tech-forward operators understanding AI potential. “The initial folks I went out to were folks that were kind of the leading edge folks,” he explained. These innovators bought on technology promise and willingness to experiment.

As PredictAP matured, they encountered different buyers. “Our sales cycle becomes much longer. The worry about it becomes much longer, the resistance becomes much longer.” Conservative firms required higher proof thresholds before making changes.

“You experience the more conservative firms who have a much higher bar to make a change,” David noted. “That’s why these initial sales have to be very successful and you have to have customers that advocate for you.” Early adopter deployments become proof points for pragmatist buyers who need demonstrated ROI and reference customers before moving.

The ICP shift required different positioning entirely. What closed innovators—AI capabilities, technology potential—didn’t resonate with mainstream buyers evaluating operational risk and integration complexity.

Domain Expertise as Competitive Moat

Real estate’s relationship with technology adoption follows predictable patterns. “At its core, real estate people are deal people,” David observed. “They love this deal, they want to be deal makers and that’s what they worry about. What’s the next deal, the next market, the next. What’s the rate?” Operations become afterthoughts until scale forces attention.

David’s operational background created asymmetric advantage. “I know what they’re going through, I’ve experienced their pain and can relate to them at a level,” he explained. Pure technology founders struggle: “Some smart Silicon Valley people may have really great tech that really solves a problem, but they may not” understand the contextual complexity that makes buyers skeptical.

The skepticism comes from experience. “People are always reluctant to, they think real estate is different for a lot of reasons. It is. And if you don’t have that expertise or background, it can be a lot of skepticism and you’re probably not going to get that sale.”

Workflow Integration Over Technical Superiority

David’s strongest critique targets founders optimizing isolated technical problems. “They have a perfect solution from the technical problem. But it’s not going to work for this firm because it’s not going to fit into their workflow,” he warns.

He maps the complete requirements: integration with existing systems, training processes for teams, user experience matching current workflows, consistency across use cases, speed meeting operational deadlines. “Your problem is not in isolation. Right. It needs to work with all these other things. And I think that’s an area where a lot of people skimp and don’t really think through and it leads to problems with adoption and retention.”

The failure mode is clear: “This is what happens when you get the smart Stanford person who’s exceptionally smart, maybe their AI is that much better and all that. They have a perfect solution from the technical problem” but miss the operational context that determines actual adoption.

David’s philosophy reflects 20 years solving operational problems before pursuing technology solutions. “I’m not the smartest tech person or business person, but I have enough to kind of put those pieces together.” The synthesis matters more than pure technical capability.

Five years after leaving Colony Capital, David’s approach has proven out. PredictAP processes massive volumes for major real estate firms including Bridge Investment Group. The company built their first million-dollar quarter using systematic sales infrastructure rather than network dependence. And the lessons about authentic networks, customer sophistication sequencing, and complete workflow integration continue shaping how vertical SaaS companies scale beyond founder-led growth into repeatable revenue engines.