Tuesday, September 15$199late booking, seats still openBook a seat
AI Ninja AcademyLive AI classes
← All guides

Should I learn this myself or hire someone to build it?

Learn it yourself if the workflow is yours and it changes weekly. Hire if it crosses systems you do not control, like payroll, a client's database, or your booking platform's API. The deciding variable is not difficulty or budget. It is who maintains it in month six. If the honest answer is you, learn it now and skip the handover.

By Uvin Vindula · Checked 2026-09-03 · 8 min read

The deciding variable is maintenance, not difficulty

Most people ask this question as "is this too hard for me?" That is the wrong question. Almost every small-business AI workflow is easy to build once.

The hard part is the second month, and the sixth. The prompt drifts. A supplier changes their invoice layout. A model version is retired. Someone renames a folder and the automation stops.

Software has always cost more to keep than to make. Published estimates put maintenance at roughly 60% to over 90% of total software lifecycle cost, depending on the study (pmc.ncbi.nlm.nih.gov). The range usually cited from IEEE sources is 60% to 80%. AI workflows are worse than average, because the model underneath you changes without asking.

So ask the maintenance question first. Who fixes it when it breaks at 8pm on a Thursday? If that person is you either way, you are going to learn this regardless. Better to learn it before you have paid someone.

What "the workflow is yours" actually means

A workflow is yours when three things are true. You already do the task by hand. You are the person who decides whether the output is good. And the rules change often enough that you cannot write them down once.

Quoting is a good example. So is drafting proposals, triaging inquiries, writing job ads, summarizing site visits, and turning a call into a follow-up email. In every case the judgement lives in your head, not in a specification.

Handing that to a contractor means you first have to write the judgement down. That is usually harder than learning the tool. You will spend two weeks in briefing calls describing what "a good quote" looks like, and the contractor will still get it wrong twice.

A workflow is not yours when it touches systems with credentials, permissions, or other people's data. Connecting your website to your CRM is not a prompt. It is software with an owner, a failure mode, and a security surface.

What US help actually costs in 2026

Rates for AI work are quoted wildly, because "AI consultant" is not a protected term. Anchor on wages first. The US Bureau of Labor Statistics publishes the median for the occupations these people actually come from.

Role you are really buying US median wage, May 2025 Loaded cost per hour Source
Management analyst (process, not code) $101,860/yr, $48.97/hr about $70 bls.gov
Data scientist $120,230/yr, $57.80/hr about $83 bls.gov
Software developer $135,980/yr, $64.44/hr about $92 bls.gov
IT/systems manager $175,140/yr, $84.20/hr about $120 bls.gov

The loaded column is the wage divided by 0.699. That is not a guess. BLS Employer Costs for Employee Compensation for March 2026 puts private-industry wages at 69.9% of total compensation and benefits at 30.1%, averaging $46.60 per hour worked (bls.gov).

Now the market rates. Clutch's September 2026 pricing guide reports US software development companies charging $50 to $99 per hour on average (clutch.co). Most firms listed globally sit in the $24 to $49 band, with Canadian and Australian firms at $100 to $149. For independent consultants, Consulting Success treats $150 per hour as a normal North American floor (consultingsuccess.com).

Read those two paragraphs together. An independent US "AI consultant" quoting $150 to $300 an hour is charging two to four times the loaded cost of the salaried person doing the same work. Sometimes that is fair, because you are buying speed and no hiring risk. Sometimes it is a markup on a Zapier login.

The useful question at a quote is not "is this expensive?" It is "which of the four rows above are you, and why is the multiple what it is?"

The decision table

Situation Learn it Hire it Why
Drafting, summarizing, rewriting your own documents Yes No Judgement is yours; a contractor cannot hold it
Rules change more often than monthly Yes No Handover cost repeats every change
Needs a login to payroll, banking, or a client system No Yes Credentials and liability you should not improvise
Touches customer personal data at scale No Yes Consent, retention, and access controls are real work
One-off migration or cleanup with a fixed end No Yes No month six to worry about
Connects two systems through an API No Yes Breakage is silent and needs monitoring
You would be the only person who could fix it Yes No You will be maintaining it regardless
Regulated output: hiring, lending, insurance, health No Yes Documentation and audit trails are the deliverable
Nobody in the business can say what "good" looks like Neither yet No Scope the problem before you buy anything

The month-six test

Run this before you sign anything or start any course. It takes about twenty minutes.

  1. Write the task out by hand. Describe one real instance, start to finish, in plain sentences. If you cannot, the problem is not solved by either route yet.
  2. Count how often the rules changed in the last six months. Prices, wording, who approves what. Two or more changes means you own it.
  3. List every system it touches. Anything needing a password that is not yours personally goes in the hire column.
  4. Name the month-six owner. A person, not a company. Write the name down.
  5. Ask that person if they know. If it is you and you have not agreed, stop. That is the real decision.
  6. Price both routes over twelve months, not one. Build cost plus twelve months of changes, against your hours plus training.
  7. Decide, then write down what would change your mind. Revisit in ninety days.

Most owners who do this honestly find the answer flips. The build was never the expensive part.

What learning actually costs you

Learning is not free, and the sales pitch for it is usually dishonest about hours. Be concrete.

For a workflow you already do by hand, expect five to ten hours to get a first version that saves time, then an hour or two a month keeping it current. That is a real number for prompt-and-document work: drafting, summarizing, extraction, structured rewriting.

For anything with a connection to another system, multiply by five and add a maintenance habit you do not currently have. That is the honest line between the two columns.

There is evidence that time spent learning pays back sharply. Slack's Workforce Index found that 61% of desk workers had spent under five hours learning to use AI, and 30% had no training at all (slack.com). The bar for being better than average is remarkably low.

Why so many paid builds produce nothing

Before you hire, know the base rate. MIT's Project NANDA report The GenAI Divide: State of AI in Business 2025 found that around 95% of enterprise generative AI pilots produced no measurable profit-and-loss impact, against $30 to $40 billion of enterprise spend (mit.edu, via reporting). The report attributes the gap to integration and learning, not model quality.

Boston Consulting Group frames the same finding as a budget rule: about 10% of the effort is algorithms, 20% is technology and data, and 70% is people and process (bcg.com).

A contractor sells you the 10% and the 20%. Nobody can sell you the 70%. That is the strongest argument for learning enough to own the process yourself, even when you also hire.

The hybrid that usually wins

The best outcome for most small firms is not one column or the other. It is a split along the credential line.

You learn and own the prompts, the checking step, the document templates, and the decision about what "good" means. You hire for the plumbing: the connection, the database, the permissions, the thing that must not leak.

Write the split into the contract. Ask for three things specifically: the prompts in a document you can edit, a written list of what breaks it, and a named person's rate for fixes after handover. If a contractor will not give you the prompts in plain text, they are selling you a dependency, not a workflow.

Also check the base rate before you spend. Census Bureau data for December 2025 to May 2026 shows under 20% of firms with four or fewer employees using AI at all (census.gov). The figure for firms with 250 or more employees is 37%. The gap is not talent. It is that small firms have nobody whose job is month six.

Where these figures come from

People also ask

A page of standing instructions that writes your weekly document for you

Rather do this with someone than read about it?

Automate reports and documents with AI is a live two-hour class built on exactly this problem. 25 seats, your own files, and you leave with a saved set of instructions that writes that document for you, the way you want it.

Read next

Can lawyers use ChatGPT and keep client confidentiality?

Yes. ABA Formal Opinion 512, issued 29 July 2024, permits generative AI use under the existing Model Rules.

Can my ChatGPT conversations be subpoenaed?

Yes, and they already have been. In New York Times v. OpenAI, a court ordered OpenAI to produce 20 million de-identified ChatGPT conversations.

How do I stop AI making things up in my documents?

You cannot turn it off, so you build a checking step instead. Feed the AI your own source files, make it quote and cite, then verify every name, number…