Three Functions Deep

Andreessen Horowitz published a long deck on the state of the markets this week, and buried in the middle was the most useful AI chart I've seen all year.

Forget models and valuations. This one was about returns. Specifically, how many companies report getting a real return from AI, sorted by how many parts of the business they'd actually built it into.

With AI embedded in no specific business function, about 25 percent reported returns. With AI built into one or two functions, about 55 percent. With three or more functions, 93 percent.

Twenty five, fifty five, ninety three.

The same deck had a few other numbers that frame it. AI is still under about 4 percent of enterprise IT budgets. Sixty nine percent of S&P 500 companies have something live. But only around 30 percent report a quantified result, and roughly 2 percent describe an AI job as mission critical.

Read those together and the story is pretty clear. Almost everybody has AI somewhere. Very few have it anywhere that matters. And the ones getting paid are the ones who went deep across several parts of the business instead of sprinkling it on top of one.

So that's the play this week. Go deeper with AI in a few more places, and wire those places together.

Here's what I think explains the jump from 55 to 93.

When AI lives in one function, say marketing writes faster with it, the gains are real but they're trapped. Marketing produces more content, and then that content hits a sales team working the same way it always has, handing off to a delivery team working the same way it always has. The bottleneck just moves downstream. You saved hours in one department and the business as a whole barely notices.

When AI lives in three or more functions, the gains start to connect. Marketing captures better information about leads. Sales uses it to qualify and follow up faster. Delivery gets a clean handoff with everything the client said on the sales calls. Finance sees what was sold and bills it without anyone retyping it. Each function's output becomes better input for the next one.

That's where the compounding happens. Cleaner handoffs between functions do more for you than speeding up any single one. Most of the waste in a small business lives in the handoffs: the information that gets lost, retyped, asked for twice, or forgotten between one person's desk and the next.

Most small businesses I talk to are in the 25 percent bucket and don't know it.

They use AI, absolutely. The owner drafts emails with it. Someone on the team uses it to clean up social posts. Maybe there's a chatbot on the website. But none of it is built into how a function actually runs. It's a tool individuals reach for when they remember, not a part of the process.

That's the dabbler trap. You've got the subscriptions. You've got some time savings. But if you stopped using AI tomorrow, the business would run exactly the same way, just a little slower. Nothing depends on it, so nothing compounds.

More tools won't get you out of it. What does is picking a few functions and rebuilding how work actually moves through them.

Get a piece of paper. Write down the five or six core functions of your business. For most small companies it looks something like this.

Getting attention. Marketing, content, ads, referrals.

Converting interest. Sales calls, proposals, follow up.

Delivering the work. The actual service or product.

Keeping clients. Support, account management, renewals.

Running the money. Invoicing, collections, bookkeeping.

Running the team. Hiring, onboarding, internal communication.

Next to each one, write an honest score from zero to two. Zero means AI isn't part of how that function runs, even if someone occasionally uses a chatbot. One means at least one recurring task in that function has AI built into the standard process, so it happens the same way every time regardless of who's doing it. Two means AI is woven through most of the function's recurring work and the function would genuinely struggle without it.

Most people I walk through this score themselves two or three total out of twelve. That's normal. It's also the gap.

Here's the call you have to make. You're going to pick three functions to take to at least a one, and you're going to pick three that hand work to each other.

Don't pick marketing, finance and hiring. They barely touch. You'll get three isolated wins and no compounding.

Pick a chain. For most service businesses, the best chain is converting interest, delivering the work, and keeping clients. That's the sale, the delivery, and the relationship, and the handoffs between them are where clients either feel looked after or feel like they're starting over with every new person.

For a business with a sales problem, the chain might be getting attention, converting interest, and running the money, so that every lead is tracked from first touch to paid invoice.

The point is that each function's output feeds the next one. That's what makes three connected functions worth more than three separate ones.

FROM THE AI NEWSROOM

The exact systems behind everything in this issue. The audit sheets, the routing logic, the templates and the review cadences, built out step by step so you can copy them straight into your own stack. One time, forty seven dollars.

Get the Blueprint for $47

.....

This is the part almost everyone gets backwards.

The natural instinct is to make each function faster on its own. Faster proposals. Faster delivery. Faster support replies. Those are fine. But the bigger win is the moment information passes from one function to the next.

Take the chain of sales, delivery and client care. Here are the three handoffs worth building.

Sales to delivery. Every sales call gets recorded and transcribed. Fathom handles that. A model pulls out what the client said they want, what they're worried about, any deadlines, budget constraints, and the exact words they used to describe their problem. That becomes a kickoff brief, automatically created in your project system the moment a deal closes. Your delivery person reads it before the first meeting. The client never has to explain themselves twice. If you've ever had a client say "I already told your sales guy this," you know what that's worth.

Delivery to client care. At the end of every project phase, a model drafts a plain language update from the delivery notes: what got done, what's next, anything the client needs to do. A person reviews and sends it. The client care side starts every renewal conversation with a full record of what was delivered, in the client's own terms.

Client care back to sales. Every support conversation and check in gets tagged for expansion signals. The client mentioned a new location. They asked about a service you offer but they don't buy. They complained about a competitor they're also using. Those tags route back to whoever owns growth as warm opportunities instead of dying in a ticket history nobody reads.

If you run your business in one platform like Go High Level, all three handoffs can live there as workflows. If your tools are spread around, Make.com is the glue that carries information from one to the next.

That chart measured whether companies reported a return. Most small businesses can't answer that question about their own AI use because they never decided what return they were looking for.

Before you build anything, pick one number per function in your chain and write down today's value.

For sales, maybe it's days from first call to signed proposal. For delivery, hours spent on kickoff and status reporting per client. For client care, renewal rate or the number of expansion conversations per month.

Then measure again in sixty days.

You don't need a dashboard. A row in a spreadsheet with the before number, the after number and the date is enough. If you want to know where your own hours actually go before and after, Rize will show you without you having to keep a timesheet.

That's how you land in the 30 percent with an actual number instead of the 70 percent going on gut feel. And gut feel won't tell you what to fix next.

Here's the test for whether a function is really at a one: if the person who usually does the work is out sick, does the AI step still happen?

If the answer is "only if the substitute remembers to use it," it's a habit, not a process. And habits are the first thing to go when things get busy.

So write it down. For each function in your chain, the standard operating procedure should include the AI step explicitly. "After every sales call, the transcript is processed into a kickoff brief. The delivery lead reads the brief before scheduling the kickoff meeting." Not "feel free to use AI to help with this."

Here's how I know a business has stopped dabbling: the AI steps show up when new hires get onboarded. Nobody has to discover them. They're just how the work is done.

The a16z deck noted that the top 1 percent of AI spenders outspend the median by around 600 times, and that agents burn about five times the tokens humans do.

Don't let that scare you. The returns in that chart came from depth, not spend. A small business can get a function to a one with tools it already pays for. The cost is mostly time: an afternoon to map a chain, a couple of days to build the handoffs, a few weeks of adjusting until the team trusts it.

What it does cost is focus. You can't take three functions deep while chasing every new tool that launches. Pick the chain. Build it. Measure it. Then pick the next one.

If you'd rather work through this with other owners doing the same thing, that's what communities like Scaling.com are for.

You've got about twelve weeks left in the year. Here's how I'd spend them.

Weeks one and two. Map your functions and score them. Pick your chain of three. Write down one baseline number for each.

Weeks three through six. Build the first handoff, the one with the most obvious pain. Get it working, write it into the process, and make sure it runs when the usual person is out.

Weeks seven through ten. Build the second and third handoffs. By now, each function in the chain should be at a one.

Weeks eleven and twelve. Measure against your baseline. Decide which function goes to a two next year and which chain you build after this one.

Come January, AI will be part of how work moves through your business, from first contact to paid invoice. Based on that chart, that's the difference between a one in four chance of seeing a return and better than nine in ten.

FROM THE AI NEWSROOM

For operators who would rather build it with someone than build it alone. The full stack, installed alongside you, with the decisions made in the room instead of guessed at. Ninety seven dollars.

Join the Accelerator for $97

.....

Take fifteen minutes tonight with the six functions. Score each one zero, one or two, honestly.

Circle the three that hand work to each other and matter most to your revenue.

Then pick the single handoff between two of them where information gets lost most often. The thing clients repeat, the detail that gets retyped, the question your team asks each other every week.

That handoff is your first build. Everything else in the quarter starts there.

Jordan

The AI Newsroom | Practical AI for people with a business to run.