Mid-Market Wedge

A VP Tax is three quarters into her year. Series B Israeli company, US revenue approaching $80M annually, growing fast. She has a CPA, an Israeli accounting firm on retainer, a working relationship with a US tax boutique, and a folder that has grown month by month: intercompany agreements, board resolutions on equity grants, transfer pricing memos, US distributor contracts, a Section 102 trust plan, last year’s tax filings, and the ITA correspondence that arrived in March.

Every cross-border question that arrives lands on top of this history. A new intercompany licensing arrangement has to be consistent with the transfer pricing study from Q1. A revised equity grant has to fit the Section 102 plan the board approved in February. A Pillar Two readiness review has to reconcile with the entity structure documented in the prior year’s tax provision. The questions are not independent. The answers cannot be.

She runs through her options for the question on her desk this week. The board meeting is Thursday. The Big Four can write a clean opinion on the new arrangement, partner-led, defensible at audit, for somewhere between $80K and $200K. What they cannot do at any price she can afford is build continuity. Each engagement starts with a briefing on the company. Each partner who reviews the file reads it for the first time. Each opinion stands on its own and lives in its own file. The accumulated history of the company is not the engagement’s working asset. It is her burden to maintain and re-explain.

Her US boutique can write on US Section 174 capitalization but will not opine on Israeli Article 9(c)(1) Preferred Enterprise treatment. The Israeli firm can write on the Israeli side but will not opine on the US side. Each firm has its own partial view of the company’s history.

She could synthesize the answer herself. She has the qualifications. What she does not have is time. She is also closing the quarterly compliance file, reviewing the transfer pricing documentation, and preparing the Pillar Two readiness assessment her auditors flagged last quarter. Each of these tasks requires reaching back into the same accumulated history, in slightly different ways, with no shared infrastructure that holds it for her.

She drafts a summary herself, flags two open questions for the board to acknowledge as unresolved, and moves on. The deal closes. The opinion is “good enough.” Three weeks later, the next question arrives, and the process repeats from a slightly different angle.

This pattern repeats, in some form, every week. The corridor here is Israel-US because that is where the call came from. The same dead ends appear between any two jurisdictions where a mid-market multinational needs cross-border synthesis that builds on accumulated context, at the price and timeline the company can absorb. In other companies the VP Tax left a year ago and was never replaced. A CFO calls a lawyer who makes three calls to similar dead ends. The pattern is the same.

It is not a tooling problem. It is a structural one.

---

The economic anatomy

The Big Four reported combined global revenue of approximately $220 billion in FY2025, employing roughly 1.5 million professionals across 150+ countries. Deloitte alone reached $70.5 billion. Partner compensation in audit and tax at these firms typically runs from $500K to several million annually, with significant capital buy-in obligations. Hourly rates for cross-border tax work cluster between $400 and $800.

The math of an opinion engagement follows from this. A partner-led cross-border opinion requires partner time for review and sign-off. Partner time at $700 per hour, even with leveraged staff below, accumulates quickly. The all-in cost of a defensible cross-border opinion at the Big Four typically lands between $80K and $200K, with $120K as a working median for moderate complexity.

This is the cost structure. It is not arbitrary. It reflects the actual cost of carrying partner-quality work product, professional indemnity insurance, peer review processes, and the institutional infrastructure that gives the opinion its weight in the first place.

Now the demand side. KPMG’s 2023 Global Tax Function Benchmarking Survey reported an average annual tax-function budget of approximately $6.5M for companies surveyed, which skewed toward $1B+ revenue multinationals, with roughly half allocated to internal talent and 46 percent to outside consultants. For companies in the $40M to $200M revenue band, internal tax headcount is typically one to five professionals. Outside advisory spend is most often in the $50K to $300K annual range, allocated mostly to compliance, with opinion work treated as an exception event.

The collision is straightforward. A single Big Four cross-border opinion can consume 30 to 60 percent of a mid-market company’s annual advisory budget. When the company faces multiple cross-border questions per year, the math fails.

The market response has been to push these companies toward mid-tier and boutique firms. The pricing works better. The depth, in cross-border specifically, does not.

---

Pillar Two as the 2026 inflection

Until recently, this gap was manageable. Mid-market multinationals could absorb cross-border complexity with internal effort and selective Big Four use. The complexity itself, while real, was relatively stable.

Pillar Two changed the equation.

The OECD’s Global Anti-Base Erosion (GloBE) rules, originally agreed in 2021, became operationally live across most OECD jurisdictions in 2024 and 2025. Nearly 60 jurisdictions had enacted Pillar Two legislation by late 2025, with 34 having global minimum tax laws effective for fiscal year 2024. Israel enacted its Qualified Domestic Minimum Top-Up Tax on December 31, 2025, effective January 1, 2026. First GloBE Information Return filings for calendar-year 2024 are due June 30, 2026 for early-adopter jurisdictions. The interpretive infrastructure has grown to match: the OECD’s original Commentary ran over 200 pages in March 2022, the Consolidated Commentary reached 334 pages by April 2024 and 375 pages by May 2025, with Administrative Guidance issued in successive tranches of 111, 140+, and 88 pages. Add approximately 60 sets of national implementing legislation, each with permitted local deviations.

The €750M consolidated revenue threshold technically scopes most mid-market multinationals out of direct Pillar Two liability. This is misleading. Companies below the threshold routinely face Pillar Two readiness work because investors require it before funding rounds, auditors require it as part of going-concern review, and foreign subsidiaries in implementing jurisdictions trigger reporting obligations independent of the parent’s threshold status.

A BDO survey cited in late 2025 found 87 percent of tax leaders view Pillar Two as a significant challenge. Big Four practices have grown to absorb the work. EY reported steady 4 percent fee growth in October 2025, attributed in part to Pillar Two and AI-driven demand. Specialist tax recruiters describe GloBE-experienced talent as in acute short supply. The US has lost approximately 340,000 accountants over the past five years, with 75 percent of current CPAs expected to retire within the next decade.

The result: Pillar Two has roughly doubled the technical surface area of cross-border tax practice while the supply of partner-grade talent has shrunk. The Big Four respond by raising prices and tightening engagement minimums. The mid-market, which could barely access the level of service it needed before 2024, now cannot reach it at all.

This is what structural inflection means. Not a new product opportunity. A change in the underlying market shape.

---

The current AI-in-tax landscape

The AI tax-tooling market has grown rapidly during the same window, and a fair map of who is doing what is the starting point for any discussion of where the gap remains.

Blue J is the most developed vertical AI tax research platform. Founded in 2015 by University of Toronto law professors, the company pivoted to generative AI in 2023 and reached more than 4,000 customer organizations by late 2025, processing 3+ million queries annually. Blue J raised a $122M Series D in August 2025 at a valuation above $300M. The platform combines GPT-4.1 with retrieval-augmented generation over a curated library including Tax Notes and IBFD content covering 220+ jurisdictions. KPMG UK is a publicly named customer. Blue J’s positioning, articulated by CEO Benjamin Alarie, is explicit: the platform delivers research-grade analysis with citations, to accelerate the work of human tax professionals. It is not positioned as opinion-generating infrastructure.

Thomson Reuters CoCounsel crossed one million users across 107 countries in February 2026. CoCounsel uses a multi-model architecture (Anthropic Claude, OpenAI GPT, Google Gemini) with over 4,500 Thomson Reuters subject-matter experts contributing to validation logic. CoCounsel Tax integrates with Checkpoint and ONESOURCE. Following the 2024 acquisition of Materia, Thomson Reuters has been building agentic capability on top of the core research platform. The product is sold as embedded infrastructure within the workflows of professional firms already using Thomson Reuters tax research tools.

Harvey raised $200M at an $11B valuation in March 2026 (co-led by GIC and Sequoia), serving 100,000+ lawyers across 1,300 organizations. Built on OpenAI and Anthropic frontier models, Harvey launched practice-area assistants for M&A, litigation, tax, corporate, and IP. PwC is a strategic partner. Harvey is positioned as legal AI infrastructure for large law firms, with tax as an extension of the core legal product rather than as the primary use case.

The Big Four have all built proprietary AI platforms. EY launched the EY.ai Agentic Platform in March 2025 with 150+ tax agents and announced a target of 100,000 agents by 2028 backed by more than $1B in annual AI investment. PwC’s Agent OS, also launched March 2025, deployed 25,000 agents across client operations by year-end. KPMG Workbench launched June 2025 with 50 agents and 1,000 in development, alongside Clara (audit, 95,000+ auditors), Digital Gateway (tax), and a $2B five-year AI investment commitment. Deloitte’s Zora AI launched March 2025 focused on finance, procurement, and audit automation.

Each of these tools is well-designed for the customer it serves. Each has substantial commercial validation. None is structured to deliver cross-border tax opinions to mid-market companies that cannot economically engage the Big Four.

---

The structural gap

The gap is not a failure of imagination among existing players. It is the predictable output of how each is positioned.

Blue J is research infrastructure. Its value increases with the sophistication of the human user. A senior tax professional at KPMG UK can take Blue J’s output and convert it to opinion-grade work. A mid-market CFO cannot. The gap between research and opinion is exactly the gap that requires partner judgment, and partner judgment is exactly what mid-market companies cannot afford to staff.

Harvey is legal AI extended into tax. Its customer is the large law firm. Its pricing assumes per-seat licensing at firms with hundreds of lawyers. Tax is a practice area within the broader legal-AI product, not a vertical built ground-up for cross-border tax economics. A mid-market multinational does not purchase Harvey seats.

CoCounsel embeds in Thomson Reuters infrastructure already used by tax professionals. It accelerates their existing workflow. It does not replace partner judgment, and it is not sold to companies as a direct service. The customer is the firm, not the underlying client.

Big Four AI is the most consequential category, and the most strictly constrained. Big Four AI is deployed inside Big Four engagements. The customer who benefits from EY.ai or PwC Agent OS is the customer who has already engaged EY or PwC at partner rates. The AI improves Big Four margin and throughput. It does not lower the entry price. A mid-market company that cannot economically engage Deloitte cannot access Zora AI, regardless of how capable Zora AI becomes.

The pattern is consistent. Each player has positioned itself where its economics work: AI for elite professionals, sold through elite professional firms. The mid-market multinational that needs an opinion sits outside the customer set of every major existing tool.

This is not a market failure. It is a market structure. The Big Four cannot serve the mid-market at mid-market prices without dismantling the cost structure that makes them the Big Four. The vertical AI players serve the Big Four because that is where the spend concentrates. No existing actor has the incentive to build infrastructure that lets mid-market companies bypass the Big Four for cross-border opinion work.

The gap is therefore stable. It will not close on its own.

---

The shape of an answer

If the gap is real and structural, the question is what kind of system could fill it. The constraints define the answer.

The system must deliver opinion-grade output, not research-grade. The mid-market CFO is not staffing a tax research function and converting its output into defensible positions. The system itself must produce work that can stand up to audit committee review and tax authority scrutiny.

The system must be defensible at the architectural level, not the disclaimer level. Citation grounding cannot be an afterthought. Confidence calibration cannot be a UI element layered on top of a generic LLM. The Deloitte Australia incident of October 2025, in which a Big Four firm partially refunded a government client after errors were found in an AI-assisted report, is the canonical illustration of what happens when AI tooling is bolted onto delivery rather than designed into how the work is reasoned, retrieved, and produced. The defensibility has to be structural, with audit trails that survive review by professionals who know what to look for.

The system must be cross-border native. The hardest cross-border tax questions are not “what does Israeli law say” or “what does US law say.” They are “what happens when these two systems interact under this fact pattern, in this treaty context, given these specific entity structures.” A system optimized for single-jurisdiction research and extended into cross-border by chaining single-jurisdiction queries will reproduce the failure mode of two siloed boutiques.

The system must be economically structured for the mid-market. The unit economics must work at a price point that companies with $50K to $300K annual advisory budgets can absorb. This is not a software pricing question alone. It is a question of how cost is distributed across the architecture: which model handles which class of query, where caching applies, how human review is scoped, where the system declines to opine.

The system must match partner-level depth in the domain it covers. A mid-market opinion that is technically half-right is worse than no opinion at all. The system either reaches partner-grade depth on the specific cross-border corridor it covers, or it should not be in the opinion business.

The system must accumulate context across the engagement, not answer questions one at a time. A mid-market multinational does not have a single cross-border question. It has a sequence of related questions over years, against a body of facts that grows with each board resolution, agreement, and filing. A system that produces a defensible opinion for one question but cannot connect it to the next question reproduces the discontinuity that Big Four partner economics impose by default. The system that fills the gap has to hold the user’s accumulated facts as a working asset of the engagement.

These are functional requirements. They do not specify a knowledge graph, or a multi-agent architecture, or any particular technical choice. They specify what the answer must do. The architectural questions, which choices are forced by which requirements, are the subject of a separate essay.

---

The bet

This is the bet ktAi is making: that the architecture required to meet these constraints can be built specifically for the mid-market wedge that today gets no qualified service. Whether that bet plays out depends on execution. The structural gap, however, is not in question.

This is not a claim that ktAi has solved cross-border tax. It has not. There are open questions: how to expand jurisdiction coverage beyond the Israeli-US corridor where we began, how to deepen client-specific knowledge without sacrificing tenant isolation guarantees, how to price for a market segment that has rarely been priced. We audit the architecture regularly. We close what we find. We track what we cannot yet close. Engineering posture matters more than a finished product.

What the bet does claim is that the structural gap deserves to be filled by infrastructure designed for it, not by Big Four economics adapted downward or by elite legal AI extended sideways. The mid-market multinational with cross-border exposure is not an underserved segment of an existing market. It is a market segment that has never been served, because the economics never worked. The economics may now be workable, because the architecture to make them workable can finally be built. Whether they actually work is a question this company will answer one report at a time.

The next essay describes one architecture built to meet these constraints: how documents become structured memory, how the user’s accumulated context connects to the body of law, and what changes when both are available to the same query.