Goldman Sachs Led $110 Million Into the AI Replacing Your Underwriter — Taktile's Agentic Decision Platform Hits 95% Automation in B2B Lending

Goldman Sachs just led a $110 million round into the AI running your bank’s loan approvals—already at 95% automation in B2B underwriting and cutting anti-money laundering false positives by 75 percent. This is not a foundation model vendor bet. Goldman backed the workflow layer where the actual business decisions happen, where compliance architecture becomes competitive moat, and where regulatory-grade AI becomes operationally real.

Key Takeaways

• $110 Million Series C, Led by Goldman Sachs Alternatives. Taktile closed its Series C with co-investors Balderton Capital, Index Ventures, Tiger Global, Y Combinator, and Dig Ventures, bringing total capital to $184 million. The Goldman Sachs Alternatives lead is the signal—an investment firm whose own core businesses run on high-stakes regulated decisions is backing the company automating them for others.

• 95% Automation in B2B Underwriting. Taktile’s platform is handling business loan approvals with 95% full automation, meaning human underwriters only step in for the 5% of loans involving unusual complexity. This translates directly to lending capacity scaling without proportional headcount growth.

• 75% Reduction in AML False Positives. Anti-money laundering compliance floors at major financial institutions are staffed with analysts clearing phantom alerts (a John Smith match on a $50 transfer to his nephew costs labor hours). Taktile’s contextual reasoning eliminates 75% of these false flags, fundamentally restructuring bank compliance OPEX.

• Agentic Decision Platforms Are Not Foundation Models. Taktile’s architecture separates AI agents (bounded workers doing specific tasks like data extraction) from deterministic rule engines (hard-coded logic for actual decisions). The AI doesn’t decide; humans and rules decide. The AI reads.

• Architecture, Not Models, Is the Moat. Foundation models commoditize over time. The protective workflow layer—audit trails, human-in-the-loop override, regulatory-grade explainability, compliance engineering—takes years to build and is incredibly difficult to copy.

• Pattern Generalizes Beyond Finance. Any high-volume decision workflow with defined rules, regulatory audit requirements, and heterogeneous inputs (credit bureau data, internal rules, transaction history, model outputs) is a candidate. Insurance underwriting, government benefits, healthcare pre-authorization.

• Operator Watch: Automation Rates in Higher-Stakes Categories and Multi-Jurisdiction Regulatory Acceptance. The test is whether 95% automation holds as institutions push Taktile into higher-complexity loan categories and whether AI-native decision records satisfy regulators across multiple jurisdictions.

Goldman Sachs Led $110 Million Into AI Automating Your Bank’s Loan Approvals

On the surface, the funding announcement feels like a contradiction. A company called Taktile just closed a $110 million Series C from Goldman Sachs Alternatives, a division of an institution where risk management is practically a religion. That same institution has spent decades automating high-stakes financial decisions—credit approvals, trading compliance, wealth management workflows—so the capital allocation itself is the validation. Smart Business Automator noted this week that the Series C brings Taktile’s total raised to $184 million, with co-investors including Balderton Capital, Index Ventures, Tiger Global, Y Combinator, and Dig Ventures. The cap table alone signals that this is not a speculative bet on a generic wrapper; the heavyweights are placing concentrated capital.

What Taktile does is build an agentic decision platform for banks and insurers. That platform automates four core decision workflows: credit approvals for commercial lending, claims reimbursement for insurance carriers, fraud detection, and anti-money laundering (AML) reviews. These are not low-stakes decisions. If a bank gets an AML review wrong, it loses its charter instantly. If a fraud detection system misses a pattern, the losses cascade across the institution. If underwriting automation breaks in a material way, the reputational hit is immediate.

Why Goldman is placing this bet is the insight that matters. Goldman’s own investment banking, private credit, and trading compliance divisions run on exactly these kinds of high-volume, regulated, high-stakes decision workflows. When the firm that executes this work at enterprise scale backs the company automating it for the broader financial services ecosystem, that is pattern validation, not passive capital placement. The signal is: Goldman is betting that the regulatory-grade decision platform architecture works, scales, and will become table-stakes for financial institutions.

The market is listening. According to Smart Business Automator’s tracking of fintech funding flows, Series C rounds in regulated AI have slowed dramatically since 2024 as VCs grew skeptical of AI’s regulatory fitness. Taktile’s $110 million raise breaks that pattern—it is the largest Series C in the regulated-AI-decision-platform category in 18 months.

95% Automation in B2B Underwriting: The Hard Numbers Behind the Valuation

VC firms do not commit $110 million to a company without operational metrics justifying the return. Taktile’s performance figures are production data from financial institutions actively using the platform at scale. Two numbers anchor the valuation: 95% automation in B2B underwriting and 75% fewer AML false positives.

The 95% underwriting metric is the easier one to parse. Business loan underwriting is infamously messy work. A human underwriter doesn’t just pull a credit score; they extract revenue from financial statements, analyze corporate ownership structures, cross-reference tax returns, assess cash flow, and model risk across a dozen dimensions. Banks employ thousands of underwriters globally because the work cannot be easily compressed into a simple rule. Taktile’s platform is handling 95% of this work end-to-end: extracting data from PDFs, verifying it against external sources, running it through risk models, and producing a yes/no decision without human intervention. The remaining 5% involves genuinely unusual edge cases—complex multi-entity structures, unusual industry dynamics, or regulatory questions—that still require human judgment.

The operational impact is massive. A bank can scale lending volume almost infinitely without hiring a proportional army of new underwriters. If a bank was processing 1,000 loans per month with 50 underwriters, and 95% of those loans now run through automation, the institution can either triple lending volume with the same 50 underwriters or return the labor to other functions. The OPEX savings compound across thousands of institutions.

The 75% AML false-positive reduction is the operational earthquake. Most people outside financial services do not understand how broken legacy AML systems are. Banks are required by law to screen every transaction against global watchlists—sanctioned entities, terrorists, organized crime networks. The software systems banks deployed in the 2000s to do this screening operate on keyword matching. They are blunt instruments. If your name is John Smith and a sanctioned warlord shares that name, your $50 Venmo transfer to your nephew gets flagged as potential money laundering. By law, the bank cannot ignore it. A highly paid compliance analyst must physically review the alert and clear it. Multiply that across millions of daily transactions and you have entire floors in financial centers—New York, London, Singapore—filled with analysts staring at screens, clicking through thousands of false alarms every day to find one genuinely suspicious transaction.

Taktile eliminates 75% of these false flags by adding contextual reasoning. When the John Smith alert pops up, the agentic platform doesn’t route it to a human immediately. It analyzes unstructured data around the transaction—address, date of birth, transaction history, behavioral context. It realizes the sender is a 22-year-old college student in Ohio, not a warlord. It packages that context, runs it through the compliance rule engine, and resolves the flag automatically. The human analyst never has to see it. Wiping out 75% of false positives fundamentally restructures bank compliance OPEX and redirects human analysts toward genuinely complex cases where their judgment actually matters.

Agentic Decision Platforms vs. Foundation Models: Why This Architecture Works in Regulated Industries

Here is the fundamental problem with deploying a foundation model into a regulated financial institution: the model is a black box. You ask it a question. It processes billions of parameters through matrices you cannot fully reverse-engineer. It spits out an answer. The logic is hidden. If a regulator audits your institution and asks why you made a specific decision—why you approved this loan or flagged this transaction—you cannot produce mathematical proof. You can only say: “The model said yes.”

That is not acceptable in regulated industries. Banking regulators demand audit trails. They demand explainability. They demand the ability to trace every decision back to verifiable inputs and verifiable logic. That is why most financial institutions have run away from horizontal AI models and built homegrown compliance systems instead. The homegrown systems are slow and expensive, but they are auditable.

Taktile’s architecture inverts the problem. Instead of deploying a black-box model, Taktile owns the decision workflow layer. That layer has three components: AI agents, deterministic rules, and human reviewers.

The AI agents are narrowly scoped. They do specific, bounded tasks where AI actually excels. Data extraction from messy documents. Contextual reasoning about transaction metadata. Pattern recognition in structured datasets. They do not make decisions. They gather intelligence and hand it over.

The deterministic rules are hard-coded business logic. If revenue is above threshold X, proceed. If the transaction amount exceeds threshold Y and the sender has no prior relationship history, escalate. These are predictable. If X happens, Y is the result every single time. No probabilities. No guessing. No hallucination risk.

The human reviewers sit at escalation points. If an AI agent’s confidence score on a data extraction drops below 99%, the workflow routes that specific data point to a human. If a rule-engine decision involves uncertainty or edge-case reasoning, a human makes the call. The human is not overseeing the AI; the human is the final authority on inherently uncertain cases.

This architecture is regulatory-grade because every decision is traceable. When auditors demand proof of how a specific loan was approved or why a transaction was flagged, Taktile can show the exact AI extraction, the exact rule that fired, and the exact human override if one occurred. The compliance infrastructure is not a side effect of the product; the compliance infrastructure is the product.

Why Goldman Sachs Alternatives Leading This Round Matters More Than You Think

Goldman Sachs Alternatives is the investment division of Goldman Sachs that manages returns on the firm’s own balance sheet. It has capital to deploy and the sophistication to evaluate risk differently than a venture firm would. When Goldman leads a Series C in regulated AI, it is not placing a passive bet. It is making a statement about category validation.

Goldman’s own core businesses are the proof point. Investment banking requires high-stakes decisions about client risk and market conditions. Private credit underwriting requires the same decision discipline as bank lending. Trading compliance requires real-time fraud detection and regulatory rule enforcement. Wealth management involves AML screening and regulatory reporting. Every one of these workflows is exactly what Taktile automates.

Goldman deploying $110 million into Taktile is equivalent to the firm saying: “We know this problem intimately because we solve it every day. We know what works. We know what regulators accept. And we are betting that Taktile’s architecture is the right solution for this category.”

The co-investors validate the thesis. Balderton Capital has deep fintech expertise in Europe. Index Ventures has backed multiple regulated-AI companies. Tiger Global has conviction on venture scale-ups in Asia and the West. Y Combinator’s presence signals that Taktile started in the startup ecosystem and has now crossed into institutional acceptance. Smart Business Automator observed that the co-investor group has collectively backed over $15 billion in fintech over the past decade, meaning this cap table is not a diversification play—it is a convergence of expertise.

What Goldman is really validating is that the architecture moat holds. Foundation models will commoditize over 3-5 years. GPT will get cheaper. Claude will get cheaper. Llama will be deployed on every company’s servers. But the compliance engineering, the audit-trail architecture, the human-in-the-loop framework, the regulatory documentation—that takes 3-5 years of tedious, unsexy, unglamorous work to build. Once built, it is incredibly hard to copy. Goldman is betting that this moat is real.

Three Characteristics of Workflows Ready for Agentic Decision Automation

If you run a financial institution, an insurance carrier, or any operation with high-volume human-review workflows, the pattern applies directly to your business. The framework for identifying candidates is straightforward, based on three characteristics.

First: High-volume, repetitive decisions. Taktile’s economics work when you are making thousands or tens of thousands of decisions per month in the same category. A bank approving 5,000 business loans monthly makes sense. An insurance carrier reviewing 10,000 claims monthly makes sense. A single-digit volume makes sense only if each decision is extremely high-value.

Second: Defined decision process. The humans doing the work should not be relying on pure gut instinct. They should be following a specific checklist, a defined set of guidelines, a documented process. This means the decision criteria can be coded into rules. If the workflow is entirely judgment-based and subjective, automation is much harder.

Third: Auditability and compliance requirement. There must be a business or regulatory reason to document and explain every decision. This is what justifies the compliance overhead. If your decision process does not require audit trails, the simpler AI model might be sufficient.

When all three characteristics are present—high volume, defined process, audit requirement—an agentic decision platform is a candidate.

Watching Taktile: The Three-Year Test of Regulated AI

The Series C funding is the validation of the category. The next phase is the test of whether the category actually scales. Three watch-items will determine Taktile’s trajectory over the next 36 months.

First: Does the 95% automation rate hold as financial institutions push the platform into higher-stakes loan categories? Taktile’s current metrics come from the mid-market lending segment—$100K to $5M loans with standard deal structures. What happens when institutions try to deploy Taktile in the $10M-$50M space, where deal structures are more complex and exceptions become more common? Does automation stay at 95% or does it drop to 80%?

Second: Multi-jurisdiction regulatory acceptance. The metrics come primarily from US institutions. Do regulators in the UK, the EU, Singapore, and other jurisdictions accept AI-native decision records as audit-complete? Or do they demand additional human sign-off that breaks the automation rate?

Third: High-profile failure impact. In regulated industries, one significant automation failure—a fraudulent loan approved by the system, a legitimate transaction flagged as money laundering and causing reputational damage to the bank—can pause adoption for years. How does Taktile’s platform respond to its first major failure event?

Frequently Asked Questions

What is an agentic decision platform, and how is it different from a regular AI model?

An agentic decision platform combines narrowly scoped AI agents (workers doing specific tasks like data extraction) with hard-coded deterministic rules (business logic that fires predictably) and human reviewers (at escalation points). A foundation model is a general-purpose system that handles reasoning end-to-end as a black box. The agentic platform is auditable; the foundation model is not. Regulatory institutions require auditability.

Why is Goldman Sachs Alternatives backing this company if the technology is just a safer way to deploy AI?

Goldman is backing Taktile because its own businesses run on exactly these high-stakes regulated decisions. When the firm that does this work at enterprise scale backs the company automating it for others, that is validation that the architecture works and regulatory acceptance is real. It is not just capital; it is credibility.

If Taktile achieves 95% automation, who are the remaining 5% for, and how much does that 5% cost to staff?

The 5% are edge cases: unusual corporate structures, regulatory questions, or complexity that the rules engine cannot handle. These still require human underwriters, but now the bank’s underwriters are focused on genuinely difficult cases instead of routine approvals. This is operationally superior because human expertise is deployed where it actually matters.

What happens if the AI agent makes a mistake during data extraction?

Taktile sets confidence thresholds. If the AI’s confidence in an extraction drops below a defined level (often 99%), the workflow automatically escalates to a human reviewer. The system does not guess; it escalates. This makes the AI more like a quality-control filter than a decision-maker.

Does this pattern apply outside financial services?

Yes. Any high-volume decision workflow with defined rules, audit requirements, and heterogeneous inputs is a candidate: insurance underwriting, government benefits determination, healthcare pre-authorization, supply-chain compliance, permitting workflows. The domain-specific knowledge varies, but the architecture generalizes.

How to Audit Your Decision Workflows for Automation Potential

• Map your highest-cost human-review processes. Which workflows consume the most labor hours? Compliance reviews, loan approvals, claims assessment, customer qualification. List the top 5 by total labor cost.

• Count decision volume per workflow. How many decisions does each workflow process monthly? Automation economics work at scale. 5,000+ monthly decisions per workflow is the minimum threshold for financial justification.

• Document the decision criteria. Sit with the humans doing the work and capture their decision logic. Are they following a checklist? Are they applying rules? If they are relying on pure judgment and intuition, automation is harder.

• Identify the audit requirement. Does your workflow have a regulatory audit trail requirement? A compliance documentation requirement? A financial audit requirement? If yes, that justifies the compliance overhead of an agentic platform.

• Calculate the false-positive or false-negative cost. What is the cost of a single decision error? (An AML false positive costs labor hours. A fraud miss costs loss. A loan approval error costs default risk.) Higher error costs justify higher-precision automation.

• Prototype with 1% of volume. Do not commit to full deployment. Redirect 1% of your monthly workflow volume to an agentic platform pilot. Run it parallel with your existing process for 90 days. Measure automation rate, error rate, and compliance acceptance.

• Build the business case from pilot data. Use the pilot metrics to forecast full deployment ROI. The unit economics of labor hour savings, compliance efficiency, and error reduction should justify the platform cost within 18-24 months.

Bottom Line

Goldman Sachs led $110 million into Taktile not because AI is magic. Goldman backed Taktile because the company solved a specific architectural problem that Goldman itself faces every day: how to deploy AI at scale in a regulated environment without sacrificing auditability or explainability. The 95% automation rate in B2B underwriting and 75% reduction in AML false positives are proof that the solution works. The next step for any operator running high-volume decision workflows is to audit your own processes against the three characteristics—high volume, defined criteria, audit requirement—and ask whether the agentic decision platform pattern applies to your business. Start with your highest-cost human-review workflow and prototype with 1% of volume. If the pattern holds, the unit economics should justify expansion within 18 months. The competitive advantage accrues to institutions that move first on this architecture.

Full transcriptAs always, this content is for educational and informational purposes. Only it is not legal, financial or professional advice. Right. So welcome back to the deep dive, everyone. Today, we are really getting into the deal economy for scaling economies. Yeah. And it is a really fascinating time right now, especially looking at where the capital is flowing here in late June, 2026. It totally is. Because, well, we’re unpacking a massive funding announcement from this week. And it honestly feels like a giant contradiction at first glance. Oh, 100 percent. Because normally, if you take a highly regulated legacy financial institution and try to mix it with AI.

Like a disaster. Yeah, it’s like watching someone try to put a brilliant, but completely unpredictable intern in charge of the company vault. Right. It’s like putting a blindfolded driver in an F1 car. You just don’t do it. Exactly. I mean, large language models are probabilistic. They guess the next most likely word. Sometimes they’re insightful and sometimes they just hallucinate. And if you’re a bank, a hallucination isn’t just some funny quirk on Twitter. It’s a massive compliance violation. Yeah. That’s a fine. It’s a regulatory nightmare. So they naturally run the other way. Which brings us to the headline event today.

We are looking at a company called Tactile, which just closed a $110 million series C funding round. Which is a huge amount of capital. But the real signal here, the thing you have to pay attention to is the lead investor. Right. Because this wasn’t just your standard Silicon Valley venture firm placing a speculative bet on some new AI wrapper. No, not at all. The round was led by growth equity at Goldman Sachs Alternatives. Which is wild. It is. You’re talking about an institution where risk management is practically a religion. And they are making a massive bet on a company building AI for the financial sector.

And that’s what I want to push back on for a second. Because why? I mean, I get that banks need efficiency. But if I’m a compliance officer at Goldman Sachs, I look at generative AI and I just see a black box. Yeah, you see billions of parameters that nobody can fully reverse engineer. Exactly. So how do you deploy that in an environment with zero tolerance for error? Why is Goldman making this bet? Well, they’re making the bet because Tactile is not building a generic horizontal AI model. Okay. They aren’t selling a standard foundation model, tossing it onto a bank’s servers and just, you know, hoping it behaves itself.

Right. So what are they doing? They’re building what is called an agentic decision platform. Okay, let’s define the scope of what that actually means for the listener. Because what Tactile is automating are core high stakes decisions. Very high stakes. We’re talking about credit approvals, claims reimbursement for insurers, fraud detection and anti-money laundering reviews. Yeah, which is a huge deal. Right. And B2B business loan underwriting. Like if a system gets an anti-money laundering review wrong, the bank could literally lose its charter. Instantly. So how does this agentic decision platform differ from like a regular horizontal foundation model in a way that makes it safe for these life or death operations?

It really comes down to who or what is actually making the final call. Tactile does not just deploy the AI, they own the decision workflow layer. The decision workflow layer. What does that look like in practice? So to understand it, you have to separate the AI from the rules engine. In a typical horizontal AI deployment, you ask the model a question. It processes the data and it spits out an answer. Right. And the logic is just hidden inside the model. Exactly. Tactile flips that entirely. In their platform, the AI agents are just workers doing very specific bounded tasks. They operate inside a rigid framework of hard-coded deterministic rules.

Wait, breakdown deterministic rules for me. Are we just talking about basic if then statements? Basically, yeah. A deterministic system is predictable. If X happens, Y is the result every single time. Whereas an AI model is probabilistic, it deals in likelihoods. Right. Exactly. So what Tactile does is combine them. Let’s say a bank gets a hundred page financial document for a business loan. An AI agent is tasked with extracting the revenue figures. Which is something AI is actually really good at. Right. It’s great at that. But the AI does not then decide if the loan is approved. It just hands that extracted data over to the deterministic rule engine.

Oh, I see. And then the rule engine says, OK, if revenue is above this specific threshold, proceed to the next step. If not, reject. So the AI is basically just the reading comprehension clerk. But the hard-coded rule engine is the manager who actually signs off on the decision. Exactly. And if the AI isn’t confident in its extraction, let’s say the document is blurry or the phrasing is weird, the system has a built-in threshold. Oh, yes. So what happens then? Well, if the AI’s confidence score drops below, say, 99%, the workflow instantly routes that specific data point to a human reviewer. Oh.

So the AI agents, the rigid rules, the business context, and the human reviewers all operate together in a strictly defined process. I mean, if a horizontal AI model is a powerful but unpredictable brainstormer in a closed room, tactile’s platform is more like a heavily monitored assembly line. That’s a perfect way to look at it. You’ve got security cameras, quality assurance inspectors, and emergency stop buttons at every single station. The AI is just operating one piece of machinery on the line. Right. It’s not running the factory. And that assembly line structure, that is the entire value proposition here.

That is what makes the system regulatory grade. Because when banking regulators come in for an audit, they don’t just ask what decisions you made. No, they demand the mathematical and logical proof of how and why every single decision was made. Which you cannot easily explain with a foundation model’s output. It’s a black box. Exactly. But with this compliance architecture, specifically the audit trails, the documented human in the loop override, the explainability, you can show them exactly what happened. So the architecture itself is the moat. 100%. The architecture is the moat. Foundation models are going to commoditize.

The intelligence will get cheaper. Right. But the protective workflow layer that safely channels that intelligence into a highly regulated enterprise, that takes years of incredibly tedious compliance engineering. It is incredibly hard to copy. Okay. Let’s look at the hard numbers here, though. Because an auditable assembly line sounds great in theory. But these are ruthless VC firms pouring money in. Oh yeah. The cap table on the Ceresi is completely stacked. Yeah. You have Goldman Sachs alternatives leading. And then they’re joined by Baldur’s in capital index ventures, Tiger Global, Y Combinator and Digventures.

Yep. The heavyweights. And this Ceresi brings tactile’s total capital raise to an absolute staggering 184 million dollars. It’s a massive war chest. But that’s my point, right? To justify a 184 million dollar war chest in this deal economy, the operational expense savings, the OPEC savings, they have to be astronomical. They do. They can’t just be funding a science project. So what are the actual performance figures driving this? What is this software doing to a bank’s balance sheet? Well, the sources give us two specific metric anchors that justify this valuation. First, in B2B underwriting, tactile is achieving 95% automation.

Wow. 95%? Yeah. And second, they are driving 75% fewer AML false positives. Okay. I really want to dissect that AML metric. Because if you’ve never been inside a bank’s compliance department, a 75% reduction in AML false positives might just sound like a nice IT upgrade. Right. Like a need feature. But it is actually a fundamental restructuring of their labor force. Because anti-money laundering regulations require financial institutions to screen every single transaction against global watchlists. Right. Terrorists, sanctioned entities, organized crime. Exactly. And the legacy software systems banks rely on for this are blunt instruments.

They operate on basic keyword matching. So they flag everything. Exactly. If your name is John Smith, and there happens to be a sanctioned warlord somewhere on the planet, also named John Smith, the legacy system flags your $50 Venmo transfer to your nephew as potential money laundering. And by law, the bank cannot just ignore that flag. Right. A highly paid compliance officer has to physically review the alert and clear it. Yeah. So you have physical floors in office buildings across New York and London, filled with analysts just staring at screens chasing phantom alerts all day long. Just clicking through thousands of false alarms to find one genuinely suspicious transaction.

It is soul crushing work. And it costs these banks millions upon millions of dollars in wasted alpaks. So how does tactile solve it without just, you know, irresponsibly lowering the security threshold? Well, by using the AI agent to do the contextual reasoning that the legacy keyword system just can’t do. When that John Smith flag pops up, the egenic platform doesn’t just route it to a human right away. It looks at the context. Right. The AI agent analyzes the unstructured data around the transaction. It cross references, addresses date of birth, transaction history, behavioral context. So it realizes the sender is a 22 year old college student in Ohio, not a warlord.

Exactly. And then it packages that context, feeds it through the deterministic rule engine to ensure it meets the regulatory threshold for clearance and it resolves the flag automatically. The human never even has to see it. Never sees it. And wiping out 75% of those false positives fundamentally alters a bank’s OPEX profile. You are removing millions of dollars of wasted labor hours. While actually improving your compliance posture because your human analysts are now only focusing on the 25% of alerts that are genuinely complex or high risk. Exactly. Which perfectly explains that other metric you mentioned.

95% automation in B2B underwriting. Right. Because business loans are famously complicated. You aren’t just looking at a simple credit score. No. You’re analyzing variable financial statements, really messy corporate structures, tax returns. But by deploying an egenic platform, the AI extracts the data from those messy PDFs, verifies it and runs the risk models through the rule engine. So the human underwriters only step in for the 5% of loans that involve really unusual edge cases. Exactly. And that means a bank can scale its lending capacity almost infinitely without having to hire a massive army of new underwriters.

Okay. So if you are an operator listening to this right now, you know, maybe you run logistics for a supply chain company or you manage a massive HR department, you might be thinking, this is just a specialized banking story. But it isn’t. It isn’t at all. The reason we are exploring this is because the underlying pattern applies directly to your operations, no matter what industry you’re in. Right. Goldman Sachs alternatives is just the perfect validator for this category, because their own core businesses run on exactly these kinds of high stakes regulated decisions. So if you’re listening, where does this apply in your operational world?

Well, the actionable framework really boils down to three characteristics. First, look for a high volume human review workflow. Second, it needs to have a defined decision process. Meaning the humans doing the work shouldn’t just be relying on gut instinct, right? Exactly. They should already be following a specific checklist or set of guidelines. And third, it needs an auditability requirement. You need to be able to explain why decision was made to an internal auditor, a client, or a regulator. So we’re talking about healthcare claims processing, right? Like a human reviewing medical codes to see if a treatment is covered by insurance.

Yes. Or benefits administration. Complex enterprise IT provisioning approvals. Quality control reviews in manufacturing. Basically anywhere with high throughput, strict rules, and a heavy cost of human labor. Exactly. And if you want a checklist, an entry point for your own business, just mentally map out your department. Identify your highest cost human review workflow. The bottleneck. Right. The bottleneck that consumes the most headcount. And ask yourself if you can describe that workflow using four components. Rules plus data plus a trained model plus a human escalation path. Rules data trained model human escalation path.

Yep. If you can break your expensive workflow down into those four pieces, you don’t need to go out and build a bespoke AI. You just need to build the workflow assembly line around the AI. That playbook is so clear. And it’s what justifies this massive enterprise software category. But I really have to push back on the inevitability of this $184 million valuation. Okay. Why? Because selling software to legacy enterprises, especially banks, is absolutely brutal. You can have the most brilliant regulatory grade platform in the world, but financial institutions are notoriously slow adopters. That is very true.

An enterprise sales cycle into a legacy bank can easily take two to three years. Oh. So what are the structural risks threatening that massive debt? Well, the bull case for tactile is that this regulated AI workflow layer is proving to be the ultimate enterprise AI business model. Because it cuts through the friction. Exactly. Selling a generic horizontal AI into a Fortune 500 company is nearly impossible right now, because the chief information security officer will just block it. Right. Due to data privacy and hallucination risks. But selling a heavily guarded specific workflow engine that maps perfectly to their existing compliance requirements, that works.

And Goldman Sachs backing the category essentially rubber stands it for procurement departments everywhere. I understand the rubber stamp. I do. Yeah. But you still have to get through legal information security and probably a six month pilot program before they even think about full deployment. Oh, sure. The sales cycle is still long. When furthermore, the failure risk here is existential. In this environment, high stakes failure risk creates massive regulatory blowback. It does. Like if an AI writes a bad marketing email, you apologize. If this agentic platform, even with a 95% automation rate, somehow has a logic flaw that green lights a massive fraudulent loan or misses a sanctioned entity.

The regulatory blowback hits the bank immediately. The stock tanks, the bank gets fine millions, and they instantly rip the software out of their systems. You have slow moving customer base with absolute zero tolerance for failure. So how do you generate margin justified returns on $184 million raised? That is exactly the bear case. Because of slow adoption and those regulatory watch items, tactile faces a very long path to generate those margin justified returns. Right. And you can’t just sign three or four banks a year to justify that VC money. You need to be onboarding dozens. And expanding your footprint inside those banks constantly, plus every time you try to expand into a new geography, you hit a wall.

Oh, because of different regulators. Exactly. A compliance architecture that satisfies a U.S. federal regulator does not instantly satisfy the European Central Bank or the Financial Conduct Authority in the UK. So every new jurisdiction requires a nuanced, potentially massive adjustment to the deterministic rules engine. Yes. And regulators are going to be watching these automation rates in higher stakes categories like Hawks. Achieving 95% automation in standard B2B loans is one thing. But what happens when tactile tries to push that automation rate into complex commercial real estate? Will regulators even allow high automation rates there?

Exactly. Or will they demand a 50% human in the loop rate just to be safe, which creates a ceiling on the OPEC savings? Right. If the regulator demands more humans in the loop, the value proposition shrinks. So the moat protecting this technology is incredibly deep. But swimming across it to actually capture the market takes years and navigating a minefield of multi-jurisdiction regulations. Which validates your skepticism, but also explains why the capital was necessary. That $184 million isn’t just for software engineering. It’s to fund a massive, prolonged enterprise sales and compliance effort.

It’s a trench war. It is a trench war. Okay, let’s pull all these threads together. We started with the sheer magnitude of tactile’s $110 million series C, led by growth equity at Goldman Sachs Alternatives. Bringing their total raise to $184 million. Right. And we unpack the critical difference between generic foundation models and a regulatory grade, a genetic decision platform. The value isn’t just the AI. It is that decision workflow layer. The deterministic rules, the audit trails, the human overrides. Exactly. And we examined the hard operational metrics driving this. 95% automation in B2B underwriting and 75% fewer AML false positives.

It’s fundamentally restructuring the operational expenses of compliance. And the operator’s playbook for your own business is clear. Identify your highest cost human review workflow. If you can describe it as rules, plus data, plus a trained model, plus a human escalation path, that is your entry point to apply this task. You find that entry point and you can scale output without scaling headcount. But I want to leave you with a lingering question that really builds on this deep dive. Throughout this whole breakdown, we keep leaning on the concept of the human in the loop, right? Yeah, the human escalation path.

The senior experts who step in to handle that 5% of highly complex edge case decisions, the AI can’t confidently resolve. But if an agentic platform successfully automates 95% of routine underwriting and compliance decisions, what happens to the career path of junior analysts? Oh, that’s a fascinating point. Right. Historically, human beings learn their craft by grinding through the boring, repetitive standard work. That is how a junior underwriter develops the pattern recognition to eventually become a senior underwriter. But if the AI is handling all the basic wraps, all that standard work, you lose the training ground.

Exactly. So if the AI handles all the standard work, where do we train the human experts who are supposed to be in the loop for the remaining 5%? We might be solving a massive operational expense problem today, but engineering a severe talent development crisis for tomorrow. It is the paradox of the automation economy. Efficiency at the bottom creates a vacuum of expertise at the top. It is going to be incredibly interesting to watch how highly regulated industries try to solve that tension. It really is. Well, thank you for joining us on this exploration of the deal economy, the architecture of enterprise AI, and the systems scaling our operations.

Keep diving deep, and we will catch you next time.