What a San Francisco Law Firm Tells You About Every Services Business on Earth

October 19, 2026
Read Time: 10 minutes

Last week I wrote about services-as-software. The six-to-one ratio of services spending to software spending. The $4.6 trillion opportunity. The gap between what delivery now costs and what the market still pays.

This week I want to show you what it looks like when someone builds into that gap properly.

There's a startup in San Francisco called Global Citizens. They're an immigration law firm. Not a legal tech company selling software to immigration lawyers. An actual firm, taking actual clients, filing actual petitions with the US government.

They built their own internal system that coordinates cases, compiles petitions, and supports the legal team across the firm. They say it's made them four times more efficient.

Four times. In one of the most procedurally rigid, compliance-heavy, risk-averse industries that exists.

I'm not writing about them because immigration law is interesting. I'm writing about them because almost every decision they made is one you'll face too, whether you're selling marketing, advisory, bookkeeping, or anything else that gets delivered rather than downloaded.

They Sold the Service, Not the Software

This is the first fork in the road and almost everyone takes the wrong turn.

The instinct, when you spot an inefficiency in an industry, is to build a tool and sell it to that industry. You've seen the problem, you know the fix, so you package the fix and sell it to the thousand businesses who share the problem. Recurring revenue, scales through distribution, feels like the modern way to do it.

Global Citizens went the other way. They built the capability and kept it. Then went and took the clients directly.

Their CEO is openly bearish on most of the big legal tech companies right now. His argument is that the market is shifting from co-pilot to autopilot, and anyone selling a co-pilot is selling an incremental improvement into a market that's being rewritten underneath them.

The economics explain why.

Sell a tool and you capture tool-sized revenue. A few hundred a month, competing against other vendors, margin compressed by competition, and a long sales cycle because you're asking someone to change how they work.

Deliver the outcome and you capture outcome-sized revenue. An employment-based visa petition runs close to five figures. That's what the market already pays. That's what their competitors already charge. They just happen to deliver it with a fraction of the labour.

That's not a software business with better margins. It's a services business with software economics, and the revenue per client isn't in the same postcode.

The same logic applies whether you're selling AI marketing, AI bookkeeping, or AI advisory. Don't sell the business an AI tool to improve their marketing. Just do their marketing, with AI underneath. One is a hard sell at a low price into a market that doesn't exist yet. The other is a market that's existed for decades, at prices nobody argues with, where the client doesn't have to learn anything or change anything.

The AI should be invisible. The outcome is the product.

The Opportunity Hides in the Work Nobody Does

Here's the most useful diagnostic I can give you.

The immigration process as it currently exists is almost comically broken. You find an employer to sponsor you, make an international move, and pick a big traditional firm to handle the filing because it feels safe.

The first call goes brilliantly. You wire a fee approaching five figures. Then the momentum stops dead.

Because roughly 90% of the work that should be handled by the firm, work that could genuinely be automated, gets pushed straight back onto the client. Gathering documents. Chasing information. Filling in forms. Tracking deadlines. The thing you paid a firm to handle, you end up handling yourself.

That pattern isn't unique to immigration law. It's the shape of almost every traditional services engagement on earth.

High fee. Impressive start. Then an enormous amount of operational drag quietly transferred to the client, because the firm's margin doesn't support absorbing it. They physically cannot afford to do that work at the price they charge, so they offload it and hope you don't notice.

Which means the entry point into almost any services market is the same question. Ask any business who they currently pay for this, and then ask what they still end up doing themselves.

The gap between those two answers is the opportunity. Every time.

A marketing client pays an agency and still writes half the briefs, sources the images, and chases the reporting. A client pays for a strategy document and then does all the implementation thinking themselves. Someone pays a professional firm and then spends forty hours assembling paperwork.

You're not competing on price. You're competing on the fact that you can actually afford to do the work they're currently doing for free. That's a far stronger position, because it's a better experience at the same price rather than a worse experience at a lower one.

Know Exactly What You're Actually Selling

This is the credibility test for the whole argument, and it's where people building AI-first services go badly wrong.

One of the Global Citizens co-founders is an immigration attorney. Asked about her approach, she said that if she doesn't hear from a client within two hours of a flight landing, she assumes they've been pulled into secondary screening, and she will drive to the airport and wait.

That is not a workflow. It cannot be systematised. And it's almost certainly why clients choose them.

Their internal framing is sharp: we're no longer in co-pilot mode, we're moving from co-pilots into autopilots. But look at what they actually put on autopilot. Coordination. Compilation. Cross-referencing. Correspondence. The work that eats hours and creates errors.

What stays human is judgment, strategy, and care.

There's a moment in one of their case strategy meetings that captures it. They're working on a client applying under extraordinary ability criteria, and someone raises that hackathon judging is coming under increased scrutiny from immigration authorities. The conversation that follows is about how to strengthen the surrounding evidence, how to highlight the standing of the other judges, how to pre-empt the objection before it's raised.

No system produces that conversation. That's experienced people reading a regulatory environment and adjusting strategy.

And commercially, that conversation is what the client is paying for. The 400 pages of exhibit compilation was never the value. It was the cost of delivering the value.

So split every deliverable you sell into two piles. One is coordination, compilation, drafting, reporting, research, correspondence, data entry. The other is judgment, strategy, relationship, taste, accountability.

Automate the first aggressively. Protect the second fiercely.

Automate the second and you've built a commodity that'll be competing on price within a year. Refuse to automate the first and you've kept a cost base someone else is about to undercut.

Remove the cost. Keep the value.

Capacity Is the Prize, Not Headcount

The numbers underneath their four times claim are worth sitting with.

A case manager at a traditional firm handles 15 to 20 active cases at once on high-volume applications. That's the realistic ceiling for a human managing that much documentation and deadline tracking manually. With proper systems, their view is that number doubles or triples comfortably. On the attorney side, they believe they can hold 80 to 100 active cases per month.

And the underlying work hasn't gotten easier. One petition they showed was 1,034 pages. Another compiled over 70 exhibits across 394 pages. As one of their engineers put it, no human is realistically reviewing 400 pages in an afternoon and catching every inconsistency. That's not a limitation of effort, it's a limitation of biology.

So the system reads every file, compiles the exhibits, cross-references for contradictions, flags anything conflicting with the case strategy, and hands the attorney something they can actually assess.

What I found most interesting is what they said about hiring. Their stated goal isn't to shrink the team. It's to expand what the business can take on while continuing to hire. Automating the grunt work lets them dedicate case managers' time to the things that actually matter, and develop those people into practitioners who understand the nuance of the law rather than being another gear in a machine.

One of their team described it from her own side: she can simply take more cases now, because the system lets her work at a faster rhythm, while a human stays on the other side of every case.

This is the reframe that matters most for anyone already running a firm. Recruitment is a ceiling. Good people are scarce, slow to find, expensive to train, and risky if they don't work out. Automating the workflow layer inside existing roles breaks that ceiling without you having to solve the hiring problem at all.

Same team, substantially more output, higher margin, no recruitment cycle. And the people doing the work enjoy it more, because the tedious part of their week disappears. That's a much better conversation to have internally than anything involving the word reduction.

Side Note: Two Ways In

Everything here is the ambitious version. Build the system, rebuild the delivery, take the clients directly.

If you want a smaller entry point, there are two that run on exactly the same logic without you building anything.

Local service businesses miss roughly 27% of their inbound calls. The owner is on a roof or under a sink, the phone rings out, and the caller dials the next name on Google. You sell them an AI receptionist that answers every call, $297 a month plus $497 setup. Our platform handles delivery rather than you. We build the whole agency for you in 24 hours, brand, website, offer, demo script, proposal and contract in your name. $27 one time.

Launch a local AI agency

Or if you've got genuine industry expertise, the higher-ticket version is advisory. You go into businesses, map their workflows, diagnose where the money is leaking, and prescribe what to fix and in what order. Audits run $5,000 to $50,000 depending on size, leading naturally into implementation and retainers. Same diagnostic lens as Global Citizens, applied to someone else's business rather than your own.

Launch an AI consultancy

The Incumbents Structurally Cannot Follow

This should give you genuine confidence.

Software is crowded, well-understood, margin-compressed, and full of established competitors. Services is six times larger, priced on outcomes rather than seats, and almost entirely owned by incumbents who are structurally incapable of moving quickly.

And I mean structurally. This isn't about them being slow or stupid.

A traditional firm cannot rebuild itself as AI-native, because their cost base is their people. Their pricing is anchored to billable hours. Their partners are compensated on a model that automation directly undermines. Every incentive inside the business resists the change, and the people who'd have to approve it are the people it costs the most.

A new entrant has none of those constraints. You design for it from day one.

Which is exactly why this play is running simultaneously across legal, accounting, healthcare administration, insurance, logistics, and recruitment. Different industries, identical structure. High fees, heavy procedural workload, entrenched incumbents, and a buyer who only cares about the outcome.

Meet the World Where It Actually Is

There's a detail in how Global Citizens operates that works as a useful corrective to the fantasy version of all this.

Despite being an AI-native firm in San Francisco, they print everything. Thousands of pages a week. Attorneys review hard copies by hand. The packets get boxed and a FedEx courier collects them around two every afternoon.

Because almost no immigration applications can be filed electronically. The US government requires paper. And if someone in the mailroom decides a packet is defective, or a barcode is missing, the entire petition gets rejected, someone's status goes into jeopardy, and start dates collapse.

So the most technically sophisticated part of the process ends with a cardboard box and a courier.

Your systems have to meet the world where it actually is, not where it should be. Regulators, legacy infrastructure, clients who want a phone call before they'll buy, procedural requirements that make no sense. None of that disappears because your stack is modern.

The firms that win aren't the ones that ignore the constraints. They're the ones that automate everything up to the constraint and then handle the constraint properly.

Where to Actually Start

If you're starting from nothing, don't invent a new model. Go and find one that already works, from any era. Social media marketing circa 2010. Lead generation for local trades. Appointment setting. Bookkeeping. Then rebuild the delivery so AI handles the ads, the creative, the follow-up, the reporting, the coordination.

The model doesn't need to be new. Your delivery does. A business that ran at 55% gross margin in 2019 runs at 95% now with the same number on the invoice.

If you already have domain expertise, that's the most valuable asset in this equation. Global Citizens worked because one of the founders had been an immigration attorney and knew precisely which parts of the process were broken. Your version of that is worth more than technical ability, because technical ability is now purchasable and domain knowledge isn't.

And if you already run a services firm, start with your deliverables rather than your tools. Break each one into component workflows, separate coordination from judgment, rebuild the first and protect the second. Then decide whether to hold your price and bank the margin, or compress your price and open a market segment that was never addressable before. Both are legitimate. It depends on whether your deliverable is visible to the client or whether they're buying trust they can't audit.

The failure mode is the same across all of it, and worth naming.

Most people buy an AI tool, use it for one task, see a marginal improvement, and conclude the whole thing is overhyped. That's co-pilot thinking. Incremental assistance inside an unchanged process.

The firms getting four times improvement asked a harder question: if we rebuilt how this work gets done, with these capabilities available from the start, what would it look like?

Start there instead.

The Window

I said last week that the gap between delivery cost and market price is the widest it will ever be right now.

Global Citizens is proof of what that gap looks like when someone builds into it deliberately, in a regulated industry, against entrenched incumbents, and wins on both economics and client experience simultaneously.

One of their founders described the goal as getting to a world where anyone from any country can live and work in any other country within a week. Whether that's achievable or not, it's the right kind of ambition, because it's only conceivable if the operational cost of the process collapses to near zero.

That's the pattern everywhere. Once delivery costs fall far enough, things that were previously uneconomic become obvious.

Your market has a version of this. Something clients want that nobody offers, because it was never profitable to deliver. That's where the opening is, and it's sitting there right now in almost every services market on earth.

The incumbents can't move. The window is open.

Watch the full behind-the-scenes look at Global Citizens here

See you next week,

– Andrew

💡 Two ways to start

Launch a local AI agency. We build it for you in 24 hours. Brand, website, offer, demo assets, demo script, proposal and contract, all in your name. You sell the 24/7 AI receptionist & 5-star Google Reviews, and our platform handles delivery.

Launch a local AI agency

Launch an AI consultancy. Use your existing industry expertise to run paid AI assessments for businesses. Audits from $5,000 to $50,000, leading into implementation and ongoing retainers.

Launch an AI consultancy