Free Assessment

AI Approach Decision Matrix

Not every task needs the same AI approach. Some need a copilot. Some need an automation. Some need a custom tool. And some just need a person. Answer four questions to find out.

Built from patterns across real implementation engagements. Includes three common mismatches that waste time and budget.

No spam. Just tools and the occasional useful thing.

Find the Right Approach

Think of one specific task your team does repeatedly. Answer four questions about it.

Question 01

Who does this task?

Is it one person doing it their own way, or a team following a shared process?

Question 02

Does the output need to look the same every time?

Consistent formatting, structure, and quality — or is variation acceptable?

Question 03

Does this need a UI that non-technical people use?

Will the people doing this task interact with a form, dashboard, or interface — or is it behind the scenes?

Question 04

Will this still be relevant in 6 months?

Is this a durable workflow or a short-term need?

3 Mismatches That Waste Budget

The matrix gets you to the right approach. These keep you from picking the wrong one.

Mismatch 01

Building a custom tool when a copilot prompt would do

If one person does this task a few times a week and the output doesn’t need to be standardized, a well-crafted prompt is the right solution — not a $10K+ application. Save the build budget for tasks where multiple people need a shared interface and consistent outputs.

Signal: the “tool” would have exactly one user and no workflow beyond copy-paste.

Mismatch 02

Using a copilot for a task that should be automated

If someone is pasting the same prompt into ChatGPT 20 times a day with slightly different inputs, that’s not “using AI” — that’s manual labor with extra steps. When the task is routine, high-volume, and the logic is consistent, build the automation and get the human out of the loop.

Signal: someone has a “prompt template” doc they copy from every time.

Mismatch 03

Automating before the process is documented

You can’t automate a process that lives in someone’s head. If you ask “what are the rules?” and the answer is “it depends” with no follow-up, the process isn’t ready. Document it, run it manually with the documentation for two weeks, then automate the documented version.

Signal: the project kickoff meeting includes the phrase “well, the way I usually do it is...”

Download the Decision Matrix

PDF with the full matrix, the four questions, and all three mismatches. Share it with your team when someone asks “should we use AI for this?”

Download the PDF

PDF — free, no strings

FAQ

Questions about AI approach decisions

The things teams ask before they commit budget to a copilot, an automation, or a custom build.

When should you use a copilot vs. an automation vs. a custom AI tool?
+
Use a copilot (LLM prompt) when one person does the task ad-hoc and output variation is acceptable. Use an automation when the task is team-wide, needs consistent outputs, and will persist for 6+ months. Build a custom tool when non-technical people need a purpose-built interface. Leave it manual when a team makes judgment calls with variable outputs.
What are common AI approach mismatches that waste budget?
+
Three common mismatches: (1) Building a custom tool when a copilot prompt would do — if it has one user and no workflow beyond copy-paste, a prompt is enough. (2) Using a copilot for a task that should be automated — if someone pastes the same prompt 20 times a day, that's manual labor with extra steps. (3) Automating before the process is documented — you can't automate a process that lives in someone's head.
What counts as an AI copilot vs. an AI automation in practice?
+
A copilot is a human typing prompts into an LLM interface (ChatGPT, Claude, Gemini) to get a one-off output they review and use. An automation is a scheduled or triggered workflow that runs without someone sitting at a keyboard — a Make scenario, a Zapier flow, an n8n pipeline, or a cron job calling an API. The practical test: if you stopped showing up for a week, would the thing still run? If yes, it's an automation. If no, it's a copilot.
When does a task need a custom UI instead of a workflow automation?
+
Build a custom tool when non-technical people need to interact with the AI workflow directly — entering inputs, reviewing outputs, making decisions — and a Make or Zapier flow would leave them stranded. Signs you need a UI: multiple users touch the workflow at different stages, inputs need validation or guardrails, outputs need to be browsed or searched, or the people running the task can't reasonably be trained on a low-code builder. If one power user runs it, skip the UI.
When should you leave a task manual instead of using AI at all?
+
Leave it manual when judgment, context, or relationships dominate the task — pricing negotiations, escalated customer calls, hiring decisions, executive messaging — and the cost of a bad output is higher than the cost of human time. Also leave it manual when the task runs so infrequently that build cost will never amortize. Manual is a legitimate answer, not a failure state. AI is a tool, not a mandate.

Related Resources

Is This Task Worth Automating?
Got "automation build"? Score the specific task to confirm it's worth it.
AI Eval Starter Kit
Whichever approach you pick, measure if the output is actually good.

Know your approach? We'll build it.

Automation builds start at $5,000. Vibe-coded tools start at $7,500. Scoped, priced, and delivered in days.

Book a Discovery Call