A proposal for a trades-focused AI consulting practice inside ISI, and a working argument for why the frontier-lab giants (Anthropic, OpenAI, Google) cannot reach this market.
Between April 22 and May 12 of this year, three of the most powerful AI companies in the world announced consulting ventures within a single thirty-day window. None of the three reach below the Fortune 1000.
Standalone enterprise AI services firm with Blackstone, Hellman & Friedman, Goldman Sachs, and a consortium including Apollo, General Atlantic, GIC, Leonard Green, and Sequoia. Embeds Anthropic engineers inside mid-size clients to deploy Claude into core workflows.
OpenAI Deployment Company capitalized by a 19-firm consortium including TPG, Brookfield, and Capgemini. Acquires applied-AI consulting firm Tomoro, bringing 150 forward-deployed engineers into the new entity on day one.
Committed at Cloud Next '26 to a 120,000-member partner ecosystem to fund value assessments, agent prototyping, and embedded forward-deployed engineering alongside Accenture, Capgemini, Cognizant, Deloitte, HCLTech, PwC, and TCS. Plus a McKinsey JV.
The frontier-lab consulting arms are priced for the Fortune 1000. The trades and middle market, which is about a third of U.S. private-sector GDP, are economically out of their reach. The shop that needs an AI advisor is the fifty-truck HVAC company, the forty-employee body shop, the regional hurricane-shutter contractor.
Every buyer-side platform that publishes data tells the same story. The trades and mid-market are paying for AI help in volume. The data below shows what they are asking for and where the gap is.
Yoav Hornung, Head of Verticals, Fiverr (Spring 2025 Business Trends Index). The same Fiverr report logged a +18,347% growth in searches for AI Agent Development over the eight-month window.
The live scoreboard each client builds with ISI, using a shared pattern and the operating numbers of their own shop. Below is the shape with mock numbers, mid-month. Click any cell or row to see how the drill-down works.
Overhead to clear this month: rent, payroll, fixed costs.
Booked revenue target by month-end.
72% of Goal · pace target 75%
| Job 1042 | $4,200 | Wed | booked |
| Job 1043 | $3,800 | Thu | booked |
| Job 1044 | $5,100 | Fri | booked |
| Job 1045 | $2,650 | Fri | scheduled |
Click any term to read its definition.
The client builds the Game alongside ISI. The typing starts on the consultant’s side and finishes on the client’s. By the time ISI rolls off, the code, credentials, hosting, data, and middleware are the client’s. Every Game ships with a runbook the owner can hand to a successor.
The argument so far has been about the Game and the practice that runs it. The asset that makes both durable is one ISI already owns. Three decades of operating documents from hundreds of trades engagements: Cash Flow Models, MAM minutes, Bloodwall logs, action lists, UDE catalogs, Game scoreboards, and the working files behind every engagement the practice has run. Sanitized and trained into an open-weight model, that archive becomes the Apprentice.
Thirty years of working documents from hundreds of trades engagements. This is the practice in document form, and it does not exist anywhere else. Anthropic's new consulting venture starts from zero on trades operating data. ServiceTitan knows ServiceTitan-shaped data. McKinsey owns boardroom-shaped artifacts, not service-bay close-the-month exhibits.
Identifying detail is stripped and specifics are generalized into patterns. A current-generation frontier model does the heavy lifting on the structured extraction work; human review covers the high-risk samples. The output is a corpus of operational knowledge with no client-traceable content in it. The discipline is the moat.
An open-weight base model (the current Llama, Qwen, or Mistral generation, whichever holds the open frontier) is fine-tuned on the sanitized corpus. A retrieval layer sits in front so the model can pull specific generalized cases when the situation at hand maps to a pattern the practice has already seen.
Deployed inside the client's own infrastructure. Owned by the client. Prepares the weekly MAM pack with patterns drawn from three decades of practice. Speaks CPM natively. Knows the Wall shapes that work in trades shops. The model running in the shop is a living artifact of the practice, kept current by the Fractional AI Chief retainer.
Each offer is anchored in established trades-consulting capability.
A build of the live monthly scoreboard. Tracks weekly progress against the monthly Goal across the four or five weeks of the Game. Built against whatever systems the client already runs. The client takes the typing over the engagement; the code, credentials, hosting, and operation are theirs by the time ISI rolls off.
The Game wired into the weekly Management Accountability Meeting. An agent prepares the agenda, the open UDEs, the prior week's commitments, and proposed action items before the meeting starts.
A productized customer-outreach system the client's marketing lead can run without the consultant in the room. Compose, schedule, send, track, suppress, parse replies. Designed for multi-hundred-recipient scale.
A single drillable HTML deliverable: revenue and gross margin by service line, year-over-year, with click-to-source drill on every aggregate. Plus GBP/SEO scoreboards and AI-crawler audits.
A monthly retainer that puts the Apprentice on the client's hardware and keeps it sharp. Re-baking when a better base model ships. Refreshing the corpus as new patterns surface. Swapping inference providers when the math changes. Plus the financial bodyguard the SaaS vendors don't get to bill past: vendor selection, policy, governance, and a hard look at every automated upsell the platforms aim at the owner. The Apprentice is owned by the client; the retainer is the value, not the lock-in.
Trades-specific middleware, dashboards and drill-downs, audit engines, embedded agents, and the credential and integration disciplines that keep production work safe. The catalog grows with every engagement.
An honest read of where the practice starts. Strengths and opportunities the positioning rests on; weaknesses and threats the launch plan has to answer for.
The numbers above name the problem. The Game answers each piece of it.
The Game is the operational instrument the owner reads weekly. No AI expertise required.
The Game consolidates them into one scoreboard: Wall, Goal, Steps, thermometer, gauges.
Every Game engagement is a teaching engagement. The owner learns CPM and where AI sits inside it.
Client-built tooling, vendor independence at the model layer, and a documented playbook mean the owner sees what was built, who runs it, and what it costs to operate.
ISI's consultants have lived inside HVAC, plumbing, auto, construction, food processing, and adjacent shops for years.
Demand is forming. No incumbent advisor sells the integrated answer to the trades. The launch plan below closes the gap.
ISI consultants are already training clients on AI. The Game is one option inside that work. Four steps make the pattern transferable across the practice and turn the archive into an asset.
Stand up the full weekly CFM-to-UDE-to-Bloodwall-to-Game-to-MAM operating loop end-to-end on a single engagement. The Game is ready to stand up. The rest of the loop gets wired around it. The deliverable is a working reference engagement, not a slide.
Write the Game's build guide as a vertical-neutral document any ISI consultant can hand a client as part of the AI work they're already doing. The test is whether a consultant can pick it up and start building inside their next engagement.
A consultant outside the reference engagement builds a Game on their own client, working from the playbook. If that build lands, the Game graduates from a one-engagement pattern into a documented option ISI consultants can carry into the AI work they're already doing. Whether five take it up, fifty, or it becomes a default is a decision the firm makes from a position of evidence.
While Steps 1 through 3 prove the Game's pattern transfers, the practice begins the technical project that turns ISI's thirty-year archive into a deployable model. Sanitization first, then fine-tune onto an open-weight base, then deployment alongside the reference engagement. The Apprentice runs on the client's hardware and is owned by the client. The Fractional AI Chief retainer is what keeps both threads sewn together at every client going forward.
The token economics that make these tools cheap to build today sit below the unit cost of the AI infrastructure buildout behind them. OpenAI posted between $5 and $13.5 billion of losses against $3.7 to $4.3 billion of revenue in 2025. Alphabet's announced 2026 capex steps from roughly $75 billion to $175 to $185 billion in a single year. The window in which a small consultancy can build, ship, and price tools at current API economics is open now. Step 1 of the launch plan is best executed inside it.
ISI consultants are already training clients on AI in trade shops across the country. The Game is one option inside that work. Document the pattern, prove it transfers, and turn ISI's thirty-year archive into the Apprentice the labs cannot manufacture and the SaaS giants do not have. The category is not won by the firm with the cleverest model. It is won by the firm whose model knows what a trades shop's month actually looks like.
Download the full whitepaper