What KPMG and Anthropic Just Proved About AI Agents in Professional Services
KPMG deployed Claude to 276,000 professionals. Here's what the same AI agents cost and require for a 10-person professional services firm.
A framework for reading the KPMG-Anthropic announcement as a small professional services firm and knowing what to do with it. Specifically: which parts of KPMG's deployment apply directly to a 5-15 person practice, and where the actual constraint lies — it is not tool access.
An AI agent is a software system that carries out a defined business function autonomously — reading documents, generating drafts, routing tasks, tracking follow-ups, and producing structured outputs — without a human operator managing each step. Unlike a chatbot that waits to be queried, an agent runs on a defined workflow: it picks up a task, completes it within guardrails, and hands off the result.
On May 19, 2026, KPMG announced a global alliance with Anthropic. Claude is now embedded in Digital Gateway, KPMG's client delivery platform, accessible to 276,000 professionals in tax, audit, and advisory. A specific claim from the announcement: building an AI agent to help clients adjust to changing tax regulations — a task that previously required weeks of cross-functional effort — now takes minutes (Anthropic).
The question this creates for every professional services firm below the Big Four tier: does this announcement apply to us, or is it enterprise news we can note and move past?
What KPMG Is Actually Doing
The KPMG-Anthropic deal is not about equipping consultants with a better chat window. It is about embedding Claude directly into client workflows — inside KPMG's existing Digital Gateway platform, where professionals already build and run AI tools in client engagements (KPMG press release).
The argument that AI is capable has run for two years. What this announcement exposes is that the build-to-deploy cycle for agent-based automation has collapsed. The infrastructure KPMG built over two years means a workflow that used to require multiple teams to configure now takes one person minutes to stand up.
Building an AI agent to help clients adjust to changing tax regulations used to take weeks and required teams to switch between multiple tools. Embedded in Digital Gateway, it now takes minutes to construct (KPMG press release).
The underlying models — Claude's API, Managed Agents, agentic workflow construction — are not proprietary to KPMG. They are available through Anthropic's API. KPMG's advantage is the infrastructure layer: two years of builds, a dedicated AI & Data Labs team, and a delivery platform that smaller firms do not have.
The Access Problem for Small Firms
KPMG serves multinational enterprises, PE portfolio companies, and the tax departments of large organizations. A 12-person accounting firm, a 6-attorney practice, or an 8-person management consulting firm is not KPMG's client.
This is the gap the announcement exposes without naming it. Smaller firms reach the same underlying models through Anthropic's API; what they lack is a deployment layer configured against their specific workflows.
| Capability | KPMG | 5-15 Person Practice |
|---|---|---|
| Underlying AI model | Claude (Anthropic) | Claude (Anthropic) |
| Platform | Digital Gateway (proprietary) | Configurable via API + existing stack |
| Build time for workflows | Minutes (embedded infrastructure) | Weeks (without established infra) |
| Internal AI expertise | Full AI & Data Labs teams | Typically none |
| Entry cost | Enterprise contract | $7,500-$25,000 build + monthly operations |
Not sure where AI fits in your operations?
Take the Free AI Readiness Assessment →What KPMG's Playbook Looks Like for a 10-Person Practice
The functions KPMG is automating — regulatory monitoring, document review, client workflow management, reporting — are structurally identical to the operational bottlenecks in a 10-person professional services firm. The volume is smaller. The structure is the same.
A 10-person accounting firm handling SR&ED tax credit claims spends 15-20 hours per client project on eligibility assessment, documentation review, and application drafting. The bottleneck is not expertise — the staff has it. The bottleneck is the unstructured transfer between steps: client intake to technical review to CRA format to deadline tracking.
An agent configured to handle intake classification, documentation extraction, and first-draft application completion reduces that 15-20 hour workflow to 4-6 hours of review and approval. The firm processes more claims without adding staff.
The outcome is removing the hours of structured, repetitive work that currently prevent staff from taking more clients, without changing headcount.
The Adoption Gap and Why It Persists
Law firms with 51 or more attorneys use AI at roughly double the rate of smaller firms, according to Embroker's 2024 survey of over 200 American lawyers (Wisconsin Law Journal). This pattern holds across professional services more broadly: the OECD's 2025 analysis of SMB AI adoption found that deployment depth varied sharply by firm size, with larger firms running systematic workflows while smaller firms stayed at individual tool adoption — staff using ChatGPT independently, not integrated agents operating inside firm processes (OECD).
The barrier is process clarity. Smaller firms typically lack two things at once: a documented description of the workflow to automate, and someone who can configure an agent system against that specification. Enterprise firms solve this by hiring. KPMG built Digital Gateway and a dedicated AI team over two years. Small firms do not need to replicate that infrastructure — they need to find an implementation path that fits their scale.
Why Small Firms Haven't Already Done This
The obvious question: if the tools are accessible, why have smaller professional services firms stayed at the individual tool-adoption stage?
Three reasons, in order of frequency. First, most firms have not documented the workflow they want to automate. An agent built on an undocumented process produces inconsistent outputs because the underlying specification is inconsistent. Automating a process that does not yet exist in written form requires the firm to define it first — and that definition work often reveals that the process has informal exceptions that someone has to resolve before the agent can run it cleanly.
Second, smaller firms carry implementation risk differently than large firms. A failed AI pilot at KPMG is a contained experiment. A failed implementation at a 10-person accounting firm consumes a significant share of the partner's attention for weeks. This asymmetry makes small firms appropriately cautious — but it also means the difference between a scoped, warranted implementation and an open-ended experiment matters more.
Third, until recently, the available implementation partners were either enterprise consultants at enterprise prices or tool vendors selling self-serve platforms that required internal expertise to configure. The market for scoped, outcome-specific agent deployment at the SMB scale is still forming.
The Window the KPMG Announcement Opens
The Federal Reserve's April 2026 monitoring study found that small businesses are now adopting AI faster than large firms — a reversal in the trend data that had not appeared before in their monitoring (Federal Reserve).
The reversal is real. But the lag between tool adoption and operational integration — between "our people use AI" and "our workflows run on AI" — is where the competitive position still sits.
The firms that will look different in 18 months are not the ones that started using ChatGPT earlier. They are the ones that moved from tool use to workflow deployment: agents running intake, review, and reporting on a schedule, without a human managing each step. That shift does not require a KPMG infrastructure budget. It requires a documented process and a scoped implementation.
The KPMG-Anthropic deal makes one thing clear: agent-based workflow automation in professional services is no longer a proof-of-concept conversation. It is an operational baseline being built at enterprise scale. The question for a 10-person firm is not whether this applies to them — it is which workflow they automate first.
- The KPMG-Anthropic deal proves AI agents work in professional services at scale. The same underlying models are accessible to any firm.
- The adoption gap between large and small firms is not a technology gap. Small firms are blocked by implementation capacity and workflow definition, not tool access.
- Firms that deploy integrated agent systems in the next 12-18 months will carry a structural throughput advantage over firms still at the individual tool-adoption stage.
- SR&ED, client intake, document review, and regulatory monitoring are the highest-frequency workflows for Canadian professional services firms — and the most straightforward to automate.
[FAQ]
What does AI agent deployment actually cost for a 5-15 person professional services firm?
A scoped build typically ranges from $7,500 to $25,000 depending on the complexity of the workflow being automated and the number of existing systems it needs to connect with. Monthly operating costs for the agent infrastructure, once built, typically run $2,000 to $5,000. Recovery time depends on which workflow is automated first and how frequently it runs. A firm that automates a process running daily or weekly recovers build cost faster than one starting with a monthly workflow.
What is the difference between using a tool like ChatGPT and deploying an AI agent?
A tool requires a human to frame every query and interpret every output. An agent is configured to run a defined workflow without that human intervention — it has context about the firm's processes, connects to the firm's existing data, and produces structured outputs that feed directly into the next workflow step. The distinction is the difference between asking a question and running a process.
Does the KPMG-Anthropic deal apply to Canadian professional services firms specifically?
The workflow structures KPMG is automating — regulatory monitoring, document review, client reporting — are present in every professional services firm regardless of size or geography. The Canadian context adds specification (PIPEDA compliance, bilingual requirements, SR&ED, GST/HST), not additional barriers. The implementation requirements are the same.
Does firm size limit what an AI agent can do?
Size is not the constraint. Workflow clarity is. If a firm can describe a repeatable process — how documents move through review, how clients are onboarded, how follow-ups are tracked — an agent can be built to run it. Smaller firms often have cleaner, more consistent processes than larger firms with informal exceptions and workarounds. That clarity makes deployment faster, not slower.