AI Tools for SaaS Founders: Fix AP Blind Spots Before They Cost You

The $40,000 Problem Nobody Can See

Here is a number that should stop any CFO cold: $40,000 per month in late fees—paid, coded incorrectly as operating expenses, and invisible to leadership until someone builds the reporting infrastructure to surface them.

That’s not a hypothetical. That’s a real pattern David, CEO of Predict AP, identified repeatedly after two decades as an accountant at Colony Capital, one of the world’s largest real estate investment firms. David’s credibility isn’t theoretical—he managed accounting operations across 12,000 distinct legal entities, watched a 30-year AP director retire and bring a $60 billion company to a billing standstill, and then built a SaaS platform specifically to solve the problem he lived firsthand.

This episode isn’t about AI hype. It’s a forensic dissection of where enterprise finance operations break down, why AI tools for SaaS founders and finance leaders solve a fundamentally different problem than most people think, and what it actually takes to scale a vertical SaaS from warm relationships to a repeatable growth machine.

Key Takeaways

• 80% of AP staff time is spent researching how to code invoices—not processing them. That’s the primary automation opportunity, not data entry speed.
• Late fees are systematically miscoded as operating expenses because overworked AP staff round up. The variance is invisible to CFOs. The aggregate cost is not.
• AI cannot replace human judgment on new vendors, edge cases, or negotiated situations. The correct frame is augmentation—AI handles patterns, humans handle exceptions.
• Tribal knowledge is a business continuity risk. A single retirement can freeze accounts payable at a multi-billion dollar organization.
• Multi-entity structures create millions of valid coding permutations—a complexity level that makes manual consistency statistically impossible at scale.
• Cloud architecture enables unit economics that let you serve a mid-market CFO doing hundreds of invoices per month and an enterprise running millions—profitably at both ends.
• Founder-led sales hits a ceiling at roughly 150 relationships (Dunbar’s number). Scaling past that requires a deliberate transition to a systematic top-of-funnel growth machine.

Deep Dive: Where AP Automation Actually Creates Value

Why 60% of Your Accounting Ledger Starts With a Bill

Most finance leaders think of accounts payable as a back-office function. David’s framing resets that assumption immediately.

“Almost, you know, 60 plus percent of every entry that hits the accounting world starts from paying a bill, paying an invoice. And so, the majority of what’s hitting your ledger for an accountant, for, you know, CFO is coming from this process.”

That statistic reframes AP coding from administrative overhead to foundational data quality infrastructure. Every miscode at the AP level propagates upstream into financial statements, budget variance reports, and strategic decisions. AP automation software isn’t a convenience tool—it’s a data integrity layer for the majority of your accounting output.

For SaaS founders building financial operations at scale, and for B2B finance leaders evaluating accounts payable AI, this is the entry point for the business case. The downstream cost of poor AP coding isn’t a rounding error.

The Late Fee Problem: How Invisible Losses Compound

David’s most actionable diagnostic for spotting AP dysfunction is the late fee analysis. The mechanism is simple and lethal.

“Say a bill comes in for electric, $5,000 electric bill. Let me tell you, if you’re a day late, the utility will charge a late fee. So $5,500 bill comes in, it’s 5,000 of electric, 500 late fee. The person who’s overworked just puts it in as $5,500 electric. You pass it through. The CFO sees, oh, we had a little bit of a variance in electricity… As opposed to, oh my gosh, there’s $40,000 a month of late fees we’re paying.”

The operational implication is significant. $40,000/month in late fees isn’t a utility variance problem—it’s a staffing signal. It means AP is understaffed relative to invoice volume, bills are arriving faster than they can be processed, and someone should be hired or the workflow should be restructured. But that decision never gets made because the data never surfaces correctly.

This is what David means when he says the cost is invisible. The money is being spent. The operational failure is real. But without proper invoice coding automation that separates base costs from penalties at the line item level, leadership sees a utility variance rather than a process breakdown.

The Tribal Knowledge Catastrophe

The founding story of Predict AP is worth sitting with, because it illustrates a risk most organizations don’t price until it materializes.

“A 60 billion dollar company couldn’t pay a bill because Chrissa retired and I was like, ‘Oh my gosh, this is a real tribal knowledge problem.’”

Chrissa had 30 years of institutional knowledge—which vendors got paid on which schedules, how costs were allocated across which entities, what the exception rules were for specific properties. None of it was documented in a system. All of it lived in her head.

When she retired, a $60 billion organization lost the ability to process basic invoices correctly. The knowledge walked out the door with her because AP coding in complex multi-entity structures is not intuitive—it requires understanding the specific legal, contractual, and operational logic of each property, tenant relationship, and fund structure.

For SaaS founders thinking about AI and product innovation: this is the core value proposition of AI-powered invoice processing in a vertical like real estate. You’re not just automating keystrokes. You’re encoding institutional knowledge into a system that doesn’t retire.

Multi-Entity Complexity: Why This Problem Is Hard

The scale of the coding problem at enterprise real estate firms is difficult to overstate. Colony Capital’s 12,000 distinct legal entities is an extreme case, but even mid-market operators with $1B–$2B in real estate holdings face a combinatorial explosion of valid coding rules.

David explains the permutation problem:

“We have 12,000 entities right, 12,000 kind of distinct legal structures that it could be sent to, then multiply that by the thousands of what-ifs at the property level and you end up with kind of millions of permutations.”

Every property has specific pass-through rules. Tenant recovery tracking, construction hold items, fund allocation logic, lease terms that dictate which costs get recovered from which tenants—each layer adds exponential complexity to what looks like a simple question: which GL code does this invoice get?

Multi-entity cost allocation at this scale isn’t a training problem for humans. It’s a pattern-matching problem at a scale humans can’t sustain consistently. This is exactly where AI as an augmentation tool—not a replacement—earns its place in the workflow.

The AI Augmentation Framework: Where the Machine Stops and the Human Starts

One of the most commercially important distinctions in this episode is David’s clear articulation of where AI-powered invoice processing works, and where it fails. If you’re evaluating AI tools for SaaS founders or building AI-assisted financial workflows, this framework is the operating boundary you need to design around.

“There are definitely use cases where AI cannot make that right choice, right? Brand new vendors, brand new situations and vendors doing something tricky. You know, when there’s a nuance or there’s not a pattern, right, and there’s something to figure out or negotiate or settle, that’s where people have an advantage that AI will never have.”

The four-step augmentation framework David runs at Predict AP:

1. Identify pattern-based vs. judgment-required decisions. Historical vendor invoices with consistent coding = AI territory. New vendors, unusual line items, disputed charges = human territory.
2. Let AI handle pattern-based coding automatically. High-confidence matches process without human intervention, which is where the 80% time-research problem gets solved.
3. Route exceptions to human reviewers. New vendors, anomalous amounts, coding conflicts surface in a review queue rather than auto-processing.
4. Redeploy human time to strategic work. AP staff who were spending 80% of their time researching coding decisions now own vendor relationships, exception resolution, and escalations.

This is the antidote to the “will AI replace SaaS finance teams?” question that circulates every time a new model drops. David’s answer is unambiguous: “AI is not a replacement for a person. Period. It’s not. People have judgment and AI does not.”

The more useful frame isn’t replacement—it’s redeployment. David describes a mid-market customer where AI automation enabled a departing AP staff member not to be backfilled. That sounds like elimination. The outcome was different.

“And that woman who just did AP coding, she’s been promoted to an accountant. So, this is now a subtask of her overall bigger thing. And she’s on this path to a better career, which is like for me that’s like the chef’s kiss of like everyone’s happy, the business is happy, it’s more efficient, and this person’s on a better trajectory for their career.”

This is the accounts payable platform outcome worth selling: not headcount reduction, but career expansion. The operational savings fund the promotion.

Cloud Architecture That Serves Both Ends of the Market

One of the structural decisions that separates durable vertical SaaS from niche tools is whether the architecture can serve both mid-market and enterprise profitably. David built Predict AP explicitly to avoid the fixed-cost trap.

“We don’t need to buy a server sitting there doing nothing—it makes it very economical. And so yes, we have very large customers and we’re built to scale to millions of invoices for customers. We have some customers that have a few hundred invoice choices a month and it’s the CFO that’s doing that job and we’ve architected this way that it’s very cost-effective for a fraction of what it would cost them to hire one person.”

The design principle: event-driven, on-demand compute that scales unit economics from a mid-market CFO processing a few hundred invoices monthly to an enterprise running millions. Both are profitable customers. Neither subsidizes the other.

For SaaS founders evaluating AP workflow automation vendors, or building their own AI-assisted financial tools, this architectural decision has direct pricing implications. Usage-based pricing aligned to invoice volume works precisely because infrastructure costs move in the same direction as usage.

From Founder Sales to Systematic Growth: The Dunbar Ceiling

David’s growth story follows a pattern RPG sees repeatedly with $2M–$5M ARR SaaS companies. The first year of customers came entirely from relationships cultivated over 20 years in real estate finance. It worked. Then it stopped scaling.

“At the start, you know, one thing for folks listening—always cultivate relationships and by that I mean actually be there to help people when you need it… But that stops at a certain point, right? And so then you need a real machine to start to create this top.”

The theoretical ceiling on relationship-based selling is around 150 close relationships—Dunbar’s number. For a founder with deep domain credibility and a long career, that’s enough to land initial traction. Bridge Investment Group became a landmark early customer. Tens of thousands of invoices per month followed.

But systematic growth requires different infrastructure: clear ICP definition, repeatable messaging, a head of sales who can build the top-of-funnel engine the founder can’t run alone. The transition from founder-led to system-driven sales is where most vertical SaaS companies stall—not because the product isn’t working, but because the growth machine hasn’t been built.

About David

David is the CEO of Predict AP, an AI-powered accounts payable automation platform built specifically for the real estate industry. He spent 20 years as an accountant at Colony Capital, managing financial operations across 12,000 legal entities before founding Predict AP to solve the AP coding and tribal knowledge problems he encountered firsthand. His customer base spans mid-market real estate operators ($1B–$2B in holdings) through enterprise firms with multi-billion dollar portfolios and millions of invoices monthly.

Ready to Stop Leaving $40K/Month on the Table?

David’s story exposes a pattern that shows up in every B2B operation that relies on manual processes and institutional knowledge: the costs are real, the data is invisible, and no one makes the right decision because they can’t see the right numbers. Whether you’re evaluating AI invoice coding for your own AP function or you’re a SaaS founder trying to turn deep domain expertise into a scalable growth machine, the playbook David laid out is directly actionable. At RPG, we work with $2–5M ARR B2B tech companies to build exactly the kind of systematic top-of-funnel machine David described—so your growth stops depending on who picks up the phone.

Talk to a Growth Strategist →

Frequently Asked Questions

How much time do AP staff spend on invoice coding vs. processing?

According to David, CEO of Predict AP, roughly 80% of AP staff time is spent researching how invoices should be coded—not processing them. That ratio means only 20% of AP labor actually moves invoices forward, making it one of the highest-leverage targets for AI automation.

What are the hidden costs of manual accounts payable processing?

The biggest hidden cost is late fees that get miscoded as operating expenses. A $5,000 utility bill arriving with a $500 late fee becomes a $5,500 electric entry. Multiplied across hundreds of invoices, David found companies carrying $40,000 per month in invisible late-fee charges that should have triggered hiring decisions.

Can AI handle complex invoice allocation rules across multiple legal entities?

AI handles pattern-based decisions well—especially across multi-entity structures with thousands of permutations. But David is explicit: AI fails on new vendors, novel situations, and anything requiring negotiation. Effective AP automation routes those edge cases to human reviewers rather than automating blindly through ambiguous decisions.

How do you prevent late fees in accounts payable?

Late fees accumulate when AP teams are understaffed relative to invoice volume and bills aren’t processed on time. Surfacing late fees as a separate line item—not rolled into the base expense code—makes the problem visible. AI invoice coding that correctly separates base costs from penalty charges gives CFOs the data to act.

Will AI replace SaaS finance teams and AP staff?

No—and the evidence from Predict AP’s customers supports the opposite outcome. David describes a mid-market customer where one AP staff member was promoted to full accountant after AI automated her coding workload. AI handles pattern recognition at scale; humans handle judgment, vendor relationships, and edge cases that AI structurally cannot resolve.