AI consulting is less about showing a business a clever agent and more about finding a measurable constraint, improving one existing workflow, and making the result repeatable without the consultant in the loop. In his 2027 AI consulting course, Nick Saraev lays out six steps: find the bottleneck, goal-factor it, build, deploy, evaluate, and share what works with the team. The strongest part of the method is the order. A polished automation attached to the wrong problem does not create much value.
Start with a buyer who already has a working process and a painful, measurable delay or loss. Improve that process in a small pilot, compare results with a baseline, then hand the system and its operating rules to the team. The video's $12,373/month is a headline, not a documented average income or a promise. The article below treats Nick's examples as his experience, not independently reproduced client outcomes.
Watch Nick's Full Course
Source note, 29 October 2026: the supplied transcript and video are the source for Nick's method, client anecdotes and opinions. Anthropic's linked documentation was checked for product behavior and evaluation guidance. We did not audit Nick's consultancy revenue, client metrics, guarantees or the economics implied by the title.
1. Find the Bottleneck Before Selling a Build
Nick borrows the constraint idea from Eliyahu Goldratt's The Goal: improving a non-constrained step may make that step faster without increasing the system's final output. His diagnostic shorthand is RACE: Reach (people who see the business), Acquire (people who express interest), Convert (buyers), and Expand (value from existing customers). On a first call, he asks for the actual counts between stages before proposing software.
| Question | What to measure | Possible first pilot |
|---|---|---|
| Reach: do enough relevant people see the offer? | Qualified visits or views by source, not impressions alone. | Review top-performing creative and draft new variants for human approval. |
| Acquire: do interested people take the next step? | Form completions, booked calls and response time. | Draft faster lead replies or simplify a booking handoff. |
| Convert: do conversations become sales? | Qualified meetings, close rate, deal size and reasons lost. | Summarize calls and propose follow-ups for a salesperson to review. |
| Expand: do customers stay and buy again? | Retention, repeat purchases, churn reasons and support load. | Surface renewal risks or draft helpful account check-ins. |
Nick's example agency reaches 1,000 people, books 10 meetings, then closes one. That makes conversion the place he would inspect. It does not prove the close rate is poor without lead quality, pricing, sales cycle and cohort context. He favors growth constraints, but that is his consulting preference, not a universal business rule: fulfillment, low margins, high refunds or compliance can be the binding constraint too. The first deliverable is a baseline and a hypothesis, not a dashboard full of invented benchmarks.
2. Goal-Factor the Outcome into Actions
Goal factoring asks what would achieve the goal, then what would make each subgoal possible, until the leaves of the tree are concrete tasks. Nick's reach example breaks "get more qualified eyes on our ads" into choices such as budget, channel, creative and copy, then narrows creative work into finding existing winners and drafting testable variants. Some tasks can be delegated; other choices need business judgment or permission.
For the agency with a weak meeting-to-sale rate, a more relevant chain is: increase won deals → learn why qualified calls fail → review a sample of consented call notes → label recurring objections → draft a follow-up playbook → test it on a small set of leads with sales approval. Each arrow is a claim to test. If the issue is poor lead fit rather than follow-up, the plan changes. That is the value of doing this before connecting an agent to the CRM.
3. Build One Controlled System, Not a Demo Factory
Nick demonstrates an ad workflow: read an account's recent performance, group active ads, draft fresh hooks and copy, and create paused variants for review. This is useful as a shape of a project, not a universal instruction to connect a client's ad account. Begin with approved exports or read-only access, define what counts as a winning ad, and keep a person responsible for brand, claims, spend and publishing. Competitor ads can inform category research; do not copy protected creative or imply a competitor's results are your own.
A good first pilot has one process owner, a written scope, a representative sample and a clear rollback. Avoid shared logins and vague "give the agent access to everything" requests. Client-owned accounts, named permissions and revocable credentials make the eventual handoff far less fragile. The build is only one of Nick's six steps because a runnable workflow is not yet a business result.
4. Deploy Repeatable Work with the Right Schedule
Nick wants the client to receive the operating system, not depend on him to repeat a prompt. He suggests turning a successful procedure into a skill or scheduled task. Anthropic's current docs draw important distinctions: Claude Code cloud schedules can run while the computer is off but do not inherit local files; Desktop schedules run on the user's machine; session-scoped /loop needs an active session. Cowork scheduled tasks run remotely on eligible paid plans and depend on configured connectors and access. Availability can also depend on organization settings.
For a first client, a scheduled report or draft is safer to operate and easier to evaluate than automatic posting, outreach or account changes. Anthropic's own Cowork safety guidance recommends simple, low-risk recurring tasks and close review of results. Write down where the task runs, who owns it, which data it can read, who approves its output and how to disable it.
5. Evaluate the Business Outcome, Not Just the Agent's Output
Nick argues against judging a workflow after one attractive output. Run multiple cases, include difficult examples, and grade quality before and after a change. Anthropic's agent-evaluation guide similarly defines tasks, trials and graders and notes that repeated trials reveal variation. A useful evaluation combines rule-based checks, human review and a real outcome measure.
For the follow-up pilot, record a baseline such as median response time, percentage of qualified leads contacted, meeting-to-sale conversion, time spent per lead and corrections required. Test on a small, consented sample. Count errors, missed objections and customer complaints, not only drafts produced. Avoid claiming causality from a week of results if lead mix, seasonality or sales staffing changed. A system that saves time but lowers trust is not a win.
6. Transfer the Method to the Team, Then Find the Next Constraint
Nick recommends a weekly team sync where high performers explain how they achieved a result, and the shared procedure is updated. That is more credible than a consultant delivering a prompt and vanishing. A practical handoff includes the owner, instructions, examples, access map, test cases, failure responses, cost limits and a change log. Share only the work examples and chat records that the client is allowed to share; remove customer-sensitive material before turning them into team-wide guidance.
Once the original constraint improves, measure the full flow again. Faster lead replies may expose a shortage of sales capacity; better conversion may expose fulfillment problems. The next engagement should follow the new evidence, not a pre-sold list of automations.
What the Income Headline Leaves Out
The title's $12,373 monthly figure is an attention-grabbing scenario, not a rate card. The transcript does not present a reproducible calculation showing how a new consultant reliably reaches that number. Nick describes his own consultancy and client results, but those figures are self-reported here. Client acquisition time, implementation hours, model and hosting costs, insurance, support, taxes and churn all change take-home income. A first offer is better priced from a bounded scope and real cost estimate than from a monthly-income target.
Three mistakes Nick calls out are especially worth keeping: automating a process that does not matter, building something no one owns, and skipping measurement. I would add a fourth: mistaking a sales metric for the whole business. An automated funnel can raise booked calls while increasing refunds, overload or bad-fit customers. Track the downstream effects as well as the headline win.
A First-Client Pilot Brief
This is an original, copyable brief inspired by Nick's method, not his prompt or a tested consulting offer. It is deliberately narrow enough to discuss with a real buyer before asking for access.
Diagnose before automating
One buyer, one bottleneck, one measurable test.
Video Chapters
Diagnose the business
| Time | Topic |
|---|---|
| 00:00 | Introduction |
| 00:34 | What an AI consultant does |
| 02:25 | The six-step method |
| 03:23 | Find the bottleneck |
| 05:00 | RACE: reach, acquire, convert, expand |
| 07:34 | The first client call |
Build and operate
| Time | Topic |
|---|---|
| 10:04 | Goal factoring |
| 12:46 | Build the system |
| 16:08 | Deploy with routines |
| 19:31 | Iterate with evaluations |
| 21:14 | Sync the team weekly |
Keep improving
Sources and Limits
Checked 29 October 2026. The video and user-supplied transcript are the source for Nick's RACE framework, the six-step sequence and examples. We did not reproduce his client work, confirm the revenue anecdotes or test this pilot. Product details below are from Anthropic's documentation and may change.
| Source | What it supports | Reading note |
|---|---|---|
| Nick's course | The consulting method and his first-person examples. | A creator account, not an independent earnings study. |
| Claude Code scheduling and Cowork scheduling | Where scheduled work runs and what it can access. | Check plan, admin settings and the current interface. |
| Anthropic on agent evals | Repeated trials, graders and outcome checks. | Design the tests around the client's own failure modes. |
| Anthropic on Cowork safety | Review and caution for recurring tasks. | Does not replace the client's permissions or internal policy. |