---
title: "How to Become an AI Consultant: RACE and Goal Factoring"
url: "https://goremy.ai/blog/articles/how-to-become-ai-consultant-race-goal-factoring"
author: Luis Chavez-Mattos
editor: Luis Chavez-Mattos
editor_title: Director, Product
published: "2026-10-31T04:20:08.382Z"
updated: "2026-10-31T04:20:08.382Z"
topic: AI Agents & Automation
description: "A step-by-step AI consulting process: find the bottleneck with RACE, goal factor into actions, build, deploy, iterate with evals, and roll out company-wide."
---

# How to Become an AI Consultant: RACE and Goal Factoring

> A step-by-step AI consulting process: find the bottleneck with RACE, goal factor into actions, build, deploy, iterate with evals, and roll out company-wide.

Edited by Luis Chavez-Mattos, Director, Product · Published 2026-10-31

## How do you run an AI consulting engagement from first call to deployed automation?

Run it in six steps. Find the one constraint in the business using Reach, Acquire, Convert, Expand (RACE) numbers. Break that constraint into concrete actions with goal factoring. Build the system with an AI tool such as Claude. Deploy it. Iterate with evals. Then roll it out across the organization. Only one of the six steps involves building, so most of the work is diagnosis and judgment.

The method comes from Nick Sarif, who ran an AI automation consultancy called Left Click that he says reached around $70,000 per month. This article lays out the process on its own terms.

## TL;DR

- An AI consultant's job is to find which work in a business AI can do instead of people, then implement it. **Diagnosis matters more than building.**
- Use **RACE (Reach, Acquire, Convert, Expand)** to locate the bottleneck. Ask for numbers at each stage. The stage with the worst relative conversion is usually the constraint.
- Focus on front-end revenue problems. The creator argues most businesses lack the volume for **fulfillment to be the real bottleneck**.
- **Goal factoring** turns a vague goal into base actions by repeatedly asking "how do I achieve this?" until you reach something you can do today.
- Building is roughly 15 to 16 percent of the process by the creator's estimate. The rest is **picking the right problem and rolling the fix out**.
- Deploy, then iterate with evals, then replicate the working system across the company.

## Why is AI consulting in demand right now?

Most business owners have heard that AI can change how they operate. Few have a solution. In sales terms, they are at the problem stage: they understand the problem but have no answer yet. The consultant supplies the answer.

The creator cites US Census Bureau data suggesting roughly 15 to 20 percent of US businesses used AI between December 2025 and May 2026. His point is that even where companies say they use AI, they often mean occasional chat prompts, not AI doing real work inside operations. He also says the demand exists in enterprise as well as small business, and refers to an Anthropic executive describing enterprise demand for Claude as outpacing other offerings. Treat that as his recollection, not a verified quote.

## What is the six-step AI consulting method?

| Step | What happens |
|---|---|
| 1. Find the bottleneck | Use RACE numbers on the first call |
| 2. Goal factor | Break the bottleneck into base actions |
| 3. Build | Create the system that performs those actions |
| 4. Deploy | Put it into the client's real workflow |
| 5. Iterate | Use evals to measure and improve output |
| 6. Roll out | Sync the working system across the organization |

## What is a bottleneck and why does it matter?

The idea comes from *The Goal* by Eliyahu Goldratt, first published in 1984. A system moves only as fast as its slowest, most constrained step. Widen that step and total throughput rises in proportion. Widen any other step and almost nothing changes, because the constraint still limits the whole.

An example: a business generates plenty of leads but takes two or three days to reply to them. Better lead generation would not help. Reply speed is the constraint, and fixing it is where the gain sits.

## How does the RACE framework find the constraint?

RACE splits a digital business into four stages:

- **Reach:** how many people see the brand or offer.
- **Acquire:** turning reach into interest, such as booked meetings, form opt-ins, or calendar registrations.
- **Convert:** turning interest into revenue. This is the sales process and close rate.
- **Expand:** raising what an existing customer pays over time, meaning lifetime value (LTV).

The creator recommends concentrating on these four because they sit on the front end of the business. In his view, most companies do not struggle to fulfill demand. They struggle to generate enough of it. A solution that touches one of the four stages is also easier to price, since it ties to revenue.

### What does a first call look like?

Walk through RACE one letter at a time and ask for numbers. The creator's worked example is a digital marketing agency making around $50,000 a month:

1. **Reach:** about 1,000 people see the business each day.
2. **Acquire:** about 10 meetings are booked from that, roughly a 1 percent rate, which he reads as a healthy sign of intent.
3. **Convert:** the agency closes 1 of those 10. He cites an industry average of about 25 to 30 percent (his figure), so this is the alarm bell.
4. **Expand:** he asks about client tenure, churn, and LTV. If those numbers are acceptable, they drop out of the conversation.

Conclusion: the constraint is conversion. Every later conversation is about the sales process. That could mean sales training, automated follow-up, written SOPs, or a script overlay on the salesperson's screen. The creator says a similar focus helped take one company from about $3 million to nearly $10 million a month, a 3.5x increase. That is his own claim and I can't verify it.

The failure mode he describes is picking a niche because it sounds impressive (his joke was an AI compliance product chosen "because it has a lot of syllables"). The test is simple: is this the thing that actually hurts the client?

## How does goal factoring turn a bottleneck into something you can build?

Goal factoring has roots in the Center for Applied Rationality and the LessWrong community. The mechanics are simple:

1. State the goal.
2. Ask how you would achieve it. List the answers.
3. For each answer, ask the same question again.
4. Keep going until you reach a base action you could start on right now.

The creator's example uses a reach bottleneck. The goal is more people seeing the client's ads. One way to get there is raising the ad budget, for example from $5,000 to $10,000. That raises the next question of how to decide where the extra spend goes, and the tree continues downward from there. The transcript cuts off before the full tree, so the exact branches are not covered here.

The useful output is a set of leaf-level tasks. Some are tasks an AI system can take over: pulling campaign performance, spotting underperforming ads, drafting variations, or flagging where budget should move. If the work lives in an ad platform such as Meta's, connecting an AI model to that account's data is a natural build. That is an illustration of the pattern, not a claim about a specific product setup.

## What does the build step involve?

You build a system that performs the leaf-level actions. The creator names Claude and Codex as examples of the AI coding harnesses he has in mind. His claim is that people overweight this step. Worrying about how to use coding agents or deploy systems covers roughly 15 to 16 percent of the job. The other 85 percent is common sense: picking the right constraint, asking the right questions, and getting the result adopted.

## How do you deploy and iterate with evals?

Deployment means putting the system into the client's real workflow, not leaving it as a demo. Iteration then uses evals. An eval is a repeatable test of whether the system's outputs meet a standard you defined. For a follow-up system, that might be whether messages hit the right tone, reference the right details, and go out within a set time. Run the evals, find failures, adjust prompts or logic, and run them again.

Tie evals back to the RACE number you started with. If the bottleneck was a 10 percent close rate, the question after deployment is whether that number moved.

## How do you roll a working system out across a company?

Once a system works for one team or person, replicate it. The creator calls this syncing it across everybody in the organization. A good first win in one department gives you a template, and later departments get their own RACE pass and their own bottleneck, so the cycle repeats.

## Is AI consulting worth pursuing?

The case is that businesses want AI to do real work and few know how to start. The risk is choosing problems nobody feels. If you anchor on the client's worst RACE number and measure the result, you are selling something tied to revenue. Treat the earnings figures in this article as one consultant's reported results, not typical outcomes.

## Frequently Asked Questions

### Do I need to be a strong programmer to become an AI consultant?

Not according to this method. The creator puts building at roughly 15 to 16 percent of the work. Knowing how to use AI coding agents helps, but diagnosing the right problem and getting adoption matter more.

### Why focus on Reach, Acquire, Convert, and Expand instead of back-office work?

The creator argues most businesses don't have a fulfillment problem because they lack the volume to strain it. Front-end improvements raise revenue directly, which makes the value obvious and easier to charge for.

### What if the numbers on the first call look fine at every stage?

Compare each stage to a typical benchmark for that business type and look for the weakest relative rate. If everything looks healthy, the constraint may be elsewhere, such as speed of response. Keep asking until you find the step that limits throughput.

### What is goal factoring in plain terms?

It is repeatedly asking "how would I achieve this?" starting from a goal, until you reach an action small enough to do immediately. Those leaf actions are what you automate or build.