your ICP map, decoded. from first touch to full implementation.

In March 2026, Julien Bek at Sequoia published Services: The New Software, what might be the clearest picture of where AI money is heading. Its illustrative map puts over $1T in services markets across two axes: whether the work is outsourced or insourced, and whether it requires judgment or intelligence. The sales playbook below is my interpretation of that map.

That distinction matters more than most people realize. The quadrant a business sits in tells you how to sell to them, what to build, and how fast the window is closing.

Four quadrants. Four completely different playbooks. And a full sales cycle most people haven't thought through yet.

One thing before we get into it. This isn't about selling software subscriptions or licensing a platform. It's about building custom AI systems and infrastructure that the business owns internally. You build it, you implement it, you train their team, and they run it. That's the model.

the map

Autopilot territory sits top right. Outsourced work that runs on intelligence, not judgment.

Insurance brokerage ($140-200B). IT managed services ($100B+). Payroll and compliance ($50-70B). Claims adjusting ($50-80B). Accounting and audit ($50-80B). Healthcare revenue cycle ($50-80B). Mortgage origination ($30-50B). KYC/AML ($30-50B). Paralegal/LPO ($36B). Tax advisory ($30-35B). Legal transactional ($20-25B). Real estate closing ($20-25B). Cost estimation ($16B).

Combined TAM: north of $700B. These businesses process information for other companies. The work is rules-based, well-documented, and running on legacy systems staffed by junior analysts doing repetitive tasks. Their clients have zero emotional loyalty to the provider. When you show up with something 40-60% cheaper that runs 24/7 and can process the 200th file of the day without fatigue, while still needing checks for errors, the conversation moves fast.

Don't sell "AI." Sell operational margin.

Copilot territory sits top left. Outsourced work that runs on judgment.

Management consulting ($300B+). Graphic/UX design ($30B+). Executive search ($20B+). PR and comms ($20B+).

These businesses sell human judgment. You can't sell automation here. You sell augmentation. "AI makes your consultants 3x more productive." The pricing flips from cost savings to revenue enablement. The systems you build let them take on more clients without hiring. Smaller TAM ($370B) but larger deal sizes and stickier contracts.

Next wave sits bottom right. Insourced work that runs on intelligence.

Supply chain and procurement ($200B+). Pharmacy back-office ($30B+). Wealth management ops ($30B+). Medical admin ($20B+). Fund administration ($15-20B).

These companies do the work internally. You're displacing a department, not a vendor. Different objections entirely. Not ready for full rollout today, but 12-18 months out the math will be undeniable. The play right now is pilots and relationship-building. When this wave tips, the people who already have case studies will own it.

The watch list sits bottom left. Insourced work that runs on judgment.

Recruitment ($200B+). Advertising ($100B+). Freight brokerage ($100B+). And a dozen more. Hardest to sell into right now. These are tool plays, not services plays. Monitor for verticals migrating into the other quadrants as models improve.

I've built systems in six of these verticals personally. Health-tech, biotech, finance, commerce, martech, and about twenty agencies. The pattern is the same everywhere. The work that looks complex from the outside is usually a handful of documented processes being executed manually at scale. Once you see it, you can't unsee it.

getting them in the room

Most people open with what they built. "We have an AI platform that..."

Nobody cares.

The first conversation isn't about your technology. It's about their pain. And if you don't know their pain yet, you're not ready for the conversation.

Before you pitch anyone, run ten research calls. Not sales calls. Actual conversations with people who do the work every day. "I'm building AI systems for [vertical]. Can I ask you five questions about where you waste the most time?" No pitch. No demo. Just listen.

What you're listening for is the workflow that makes them sigh. The one they know is broken but can't fix because it's been done this way for fifteen years. The one where they lose people because nobody wants to do it. Every company in these verticals has at least one. Usually more.

After enough of these calls you notice something. Most of them already know. They've done the math, or at least the back-of-napkin version. A $100K/month operation running seven people, and the founder can point to exactly which roles are doing work that follows a process someone wrote down years ago. They've thought about automation. They may have even tried it. But what they found was either too generic for their compliance requirements or required an engineering team they don't have and can't hire fast enough. You're not convincing anyone that AI works. You're solving a build problem they've been stuck on.

Once you've heard the same pain described by five different people in the same vertical, you have your opening. Not "we do AI." Something closer to: "I've talked to a dozen firms in [vertical] and every single one is hemorrhaging money on [specific workflow]. We built something that cuts that by 60%."

That's a meeting.

The other thing that gets you in the room is proof. One real result from one real company. Numbers, not adjectives. Not "we helped a client improve efficiency." Nobody's taking a meeting for that. "We reduced claims processing time by 74% for a mid-size brokerage and cut their error rate to near zero." That's a meeting.

If you don't have a case study yet, get one. Find one company willing to let you run a pilot. Charge nothing if you have to. The first result is worth more than a year of cold outbound.

the scope audit

You got the meeting. Here's where most people blow it.

They walk in with a demo. They show the AI doing something impressive. The prospect nods politely. Then asks "how would this work for us specifically?" And the seller starts improvising.

Don't improvise. Audit.

Before you propose anything, you need to understand what's actually happening inside their operation. Not what they tell you in the first meeting, which is always the cleaned-up version. What's actually happening. Where time goes. Where money leaks. Where the same data gets touched by four different people before someone makes a decision.

The audit maps every workflow that involves human labor against what AI handles reliably in production today. Not in demos. In production. There's a real gap between what a model can do when you prompt it carefully and what it can do inside a business process with compliance requirements, edge cases, and twenty years of accumulated workarounds baked into the operation.

This isn't a surface-level assessment someone runs from a checklist they found online. We've been running these for three years across 30+ businesses and 12 verticals. Every engagement adds to a dataset of operational patterns, automation outcomes, and failure modes that most people in this space don't have.

That backlog changes what the audit can do. We walk into a new company and identify inefficiencies they don't know exist because we've seen the same pattern at fifteen other businesses in different industries. We know which workflows look automatable on the surface but will eat you alive in implementation because we've hit those walls and measured exactly where they break. We know which ones look intimidating but collapse into clean automation once you strip away the accumulated manual workarounds. We're not estimating. We have benchmarks from real implementations. Time saved, error rates before and after, adoption curves, edge case frequency. When we tell a client a workflow will see 70% automation, that number comes from comparable operations we've already built.

And it compounds. Every engagement makes the next one sharper. Three years in, we catch things in hours that would take someone starting fresh weeks to find.

When the audit is done, something shifts in the conversation. It stops being "should we adopt AI?" and becomes "we had no idea how much of this was automatable." You're not selling anymore. You're showing them what they didn't know about their own business.

The audit also protects you. Not everything that can be automated should be. Some processes have so many edge cases and judgment calls embedded in them that the build cost exceeds the savings. The audit separates real opportunities from traps before you commit to either.

The signals that a workflow is worth proposing on: the input is structured, the rules are documented, the volume is high, and the cost of errors is quantifiable. When those things line up, you have a build worth scoping.

the scoped proposal

The audit gives you the map. The proposal gives them the plan.

Most people scope too big. They look at ten automatable workflows and propose a twelve-month transformation program. The prospect's eyes glaze over, procurement gets involved, and the deal stalls in committee for six months.

Scope small. Start with the one workflow where the ROI is so obvious it barely needs explaining. The one where the numbers make the CFO's eyebrows move. That's Phase 1.

Phase 1 should feel like a no-brainer. Tight scope. Measurable outcome. Timeline short enough that they can see the finish line from the starting line. If Phase 1 takes more than 60-90 days, you've scoped too big. Cut it in half.

The proposal should show the ROI for Phase 1 alone, not for the full engagement. People buy what they can measure. They expand based on results, not promises.

Phase 2 and beyond go in the roadmap, but positioned as contingent on Phase 1 results. This does two things. It makes the initial commitment feel low-risk because it is. And it gives you leverage afterward. Once you've delivered results, the expansion conversation is about continuing something that's already working. That's a fundamentally different negotiation than pitching something new.

Prioritize the roadmap on two dimensions: ROI and implementation complexity. High return, low complexity goes first. High return, high complexity gets scheduled after you've built trust. Low return items stay on the roadmap but never lead.

Every phase needs a concrete "done" definition. Not "improve the process." A number. "Reduce processing time from 4 hours to 45 minutes." "Cut error rate from 3.2% to below 0.5%." "Handle 80% of inbound volume without human review." If you can't attach a number to it, the scope isn't tight enough.

positioning

There are a few ways to sell AI services to businesses. Most people default to the wrong one.

The dev shop model: a client comes with a spec, you build it, you hand it over, you invoice. The problem is you're competing on rate with every other builder who can prompt an LLM. That's a race to the bottom with no recurring relationship on the other side.

The consultant model: you advise, assess, recommend, write reports. The problem is the client still needs someone to build it. And consultants who can't build what they recommend have a credibility ceiling that drops every quarter as AI makes the building easier.

Then there's the operator model. You audit the operation, scope the work, build the systems, implement them inside the business, and train the team to run them without you. You own the outcome, not just the deliverable.

I landed on this after watching the alternatives play out. Dev shops hand over code that nobody inside the company knows how to maintain. Consultants produce reports that sit in a drawer. I've seen both happen to my own clients when other vendors were involved. The moment I started owning the full cycle, audit through implementation through training, the results changed. Clients stopped calling with problems and started calling with expansion requests.

This is the one that wins. Not because it's the most scalable model. Because it's the most valuable to the buyer. The client doesn't want a piece of software or a recommendation. They want the problem solved. End to end. With someone who takes responsibility for whether it actually works in their environment with their team.

When you position as an operator, the competitive set shrinks dramatically. You're not competing with freelancers or firms that hand over a repo and disappear. You're competing with the handful of people who can do the whole thing. Right now, that handful is small.

The positioning also changes the pricing conversation. When you price against the outcome instead of your hours, the numbers shift. You can charge more because you're delivering more. And the client stops asking about your rate because they're measuring you against the result.

closing

Two things close deals in this space. Proof and urgency.

Proof is the case study. Not a slide that says "we deliver results." A real story from a real company with real numbers. "We took this brokerage from 4-hour claims processing to 45 minutes. Error rate dropped from 3.2% to 0.4%. The team processes 4x the volume without adding headcount."

That's arithmetic. And arithmetic closes deals because the buyer can run the same math on their own operation.

Urgency is the market. The Sequoia map exists because these verticals are in motion. The companies that automate first will operate at cost structures their competitors can't match. Everyone in these industries senses this. Your job is to make it specific. Show them what the math looks like when a competitor runs the same workflows at 60% lower cost.

Price against value, not effort. If Phase 1 saves the client $400K a year, your price should be a fraction of that savings, not a function of your hours times your rate. Value-based pricing changes the entire dynamic. They stop asking "how long will this take" and start asking "what's the expected return."

The objections you'll hear are predictable. "We need to talk to IT" means bring IT into the audit early so they become allies instead of blockers. "What if it doesn't work" is why the phased approach exists. Phase 1 is the proof point. Low commitment, high signal. "Our data is sensitive" is actually a good sign because it means they're taking the conversation seriously. Walk them through how you handle data governance. Compliance expertise separates real operators from people who just build demos.

The deals that die usually die for two reasons. The champion inside the company doesn't have budget authority. Or the scope was too big and triggered a formal procurement process. Qualify early. Make sure you're talking to someone who can sign. And keep Phase 1 small enough to stay under whatever threshold triggers their procurement machine.

But the most common objection is "we're not ready." I've heard it from companies doing $140K/month with 7 people who were putting in 58 hours a week. They weren't not ready. They were drowning and couldn't imagine finding time to change anything while keeping the lights on. The move that worked was scoping Phase 1 around one workflow their team actively hated. The one with the highest turnover. Don't ask them to adopt change. Ask them to stop doing the thing that's burning people out. That reframe turned "we're not ready" into a signed engagement in under two weeks.

implementation and adoption

You closed the deal. Now comes the part where most AI sellers fall apart.

Building the system is not the hard part. Getting people to use it is.

I've watched roughly thirty businesses attempt this transition. The ones that failed made the same mistakes. They automated in the wrong order, pointing AI at the most visible problem instead of the most impactful one. They cut headcount before the systems were ready to absorb the work. They treated the whole thing like a technology purchase instead of a structural redesign. The ones that succeeded did something different. They kept the team intact through Phase 1, let the system prove itself, and only restructured roles after the data showed what was actually load-bearing and what was drag. That sequencing is the whole game.

Implementation has to happen alongside the team, not behind closed doors. The people who've been doing this work manually for years need to see how the system works, need to trust it, and need to feel like the transition is happening with them, not to them. If they feel replaced, they'll resist. If they feel upgraded, they'll adopt.

The first 30 days matter more than anything that comes after. This is where the team decides whether the new system is something they use or something they quietly work around. And if they work around it, you've built expensive software that sits on a shelf while they go back to the spreadsheet.

Start with the simplest version of the workflow. Not the full system on day one. The piece that's most obviously better than the manual process. The one where the person doing the work says "oh, that actually saves me two hours" the first time they see it run. That moment of recognition is more important than any feature you ship after it.

Training isn't a one-day session with slides and a Q&A. It's hands-on, spread over weeks, built around the actual work the team does every day. The best training doesn't feel like training. It feels like doing the job with a better tool, with someone there to answer questions when the edge cases show up. And the edge cases always show up.

You're done when the team runs the system without you. Not when it's built. Not when it's deployed. When the people who use it every day can operate it, handle the common issues, and know when something needs to be escalated because the AI can't handle it. That's the actual deliverable. Everything before that is setup.

The clients who get the most value are the ones whose team feels ownership of the system by the time you leave. They're not calling you when something breaks. They're calling you when they want to automate the next workflow. And that call is where the real revenue compounds.

Phase 1 was the proof. Phases 2 through 5 are where the engagement grows. Once you've delivered results and the team trusts the process, the next workflow is an easy yes. The expansion sells itself because the results already did the selling.

Over $1T sitting on that map. Most of it still running on spreadsheets and manual review. The window is open because the incumbents haven't built yet. When they do, the economics change for everyone trying to sell into these verticals.

I've done this across 30+ businesses and 12 verticals. If you want to go deeper on any of this, dm me.