Where DIY generative AI works (and where it really doesn’t): A practical guide for startups

Where DIY generative AI works (and where it really doesn’t): A practical guide for startups

Where DIY generative AI works (and where it really doesn’t): A practical guide for startups

Mar 12, 2025

Comic of two people at computers talking to their coworkers. One person says, "AI turns this single bullet point into a long email I can pretend I wrote." The other says, "AI makes a single bullet point out of this long email I can pretend I read."

If you're a founder at a growing startup, you've probably heard “AI can do that” a dozen times in the last year. But what can generative AI actually do well? And where does it still fall short?

The short answer: it depends on the type of work you’re trying to automate.

Here’s a practical framework we use when thinking about where to start with AI in your business — especially if you’re trying to DIY some of your backend operations or sales workflows.

The AI spectrum: from creative to business-critical

Think of AI use cases as a spectrum:

On the left side:
Creative work — writing, brainstorming, marketing, sales outreach, even coding.
There’s more room for subjectivity, you can treat the creation process as a "black box", and the steps in the process can vary every time. AI gets you 80–90% of the way there, and that’s usually enough to move fast, test ideas, and iterate.

On the right side:
Mission-critical work — accounting, legal, compliance, customs filings.
You need perfect accuracy. One mistake can lead to a mess of downstream problems. That’s where generative AI struggles. It guesses. It’s probabilistic. And if you’re not validating every result, it’s risky.

Rule of thumb: start where "pretty good" is still useful

If you're trying to get started with DIY'ing generative AI (e.g. chatting with or integrating ChatGPT, Claude, Gemini, et al.), focus first on high-tolerance areas:

Good AI use cases:

  • Writing product descriptions

  • Generating blog or social content

  • Drafting outreach emails

  • Brainstorming campaign ideas

  • Cleaning up CRM notes

Ideally, all of these would be double-checked and polished by a human.

Risky AI use cases (at least without oversight):

  • Revenue reconciliation

  • Accounting entries

  • Legal contract generation

  • Customs documentation

Use AI where speed matters more than perfection, and where increasing volume creates learning. Not where one error creates hours of rework.

Generative AI is a force multiplier — if you already know what you're doing

One of the most powerful (and under-appreciated) things about generative AI is that it amplifies your existing skill set.

  • If you're already a coder, it makes you faster.

  • If you're doing data analysis, it helps you write queries without the errors.

  • If you think in systems and flowcharts, AI will be your best tool yet.

But if systems thinking isn’t second nature? AI won’t magically give you that clarity. In fact, it might make things harder — especially if you’re trying to automate something like your finance workflows without knowing what all the edge cases are.

What about the middle?

Some roles fall in between — like RevOps, CX, and certain operational workflows. These can be partially automated using AI, but success often depends on how well-defined your SOPs are.

If you're creating business transactions from AI output (e.g. generating invoices or updating inventory), you want those systems clearly mapped before you bring automation into the mix.

The bottom line

If you’re a founder or leader looking to save time with AI (and DIY'ing it for now), start with the parts of your business where “good enough” is still genuinely helpful.

Let generative AI be your junior assistant.
Not your accountant.


Need help figuring out which workflows are safe to automate?
At Crafty Crow, we advise on AI integration strategy and build done-for-you automation systems for growing startups — integrating with your existing tools and SOPs, no AI hype required.

👉 Let’s talk.

If you're a founder at a growing startup, you've probably heard “AI can do that” a dozen times in the last year. But what can generative AI actually do well? And where does it still fall short?

The short answer: it depends on the type of work you’re trying to automate.

Here’s a practical framework we use when thinking about where to start with AI in your business — especially if you’re trying to DIY some of your backend operations or sales workflows.

The AI spectrum: from creative to business-critical

Think of AI use cases as a spectrum:

On the left side:
Creative work — writing, brainstorming, marketing, sales outreach, even coding.
There’s more room for subjectivity, you can treat the creation process as a "black box", and the steps in the process can vary every time. AI gets you 80–90% of the way there, and that’s usually enough to move fast, test ideas, and iterate.

On the right side:
Mission-critical work — accounting, legal, compliance, customs filings.
You need perfect accuracy. One mistake can lead to a mess of downstream problems. That’s where generative AI struggles. It guesses. It’s probabilistic. And if you’re not validating every result, it’s risky.

Rule of thumb: start where "pretty good" is still useful

If you're trying to get started with DIY'ing generative AI (e.g. chatting with or integrating ChatGPT, Claude, Gemini, et al.), focus first on high-tolerance areas:

Good AI use cases:

  • Writing product descriptions

  • Generating blog or social content

  • Drafting outreach emails

  • Brainstorming campaign ideas

  • Cleaning up CRM notes

Ideally, all of these would be double-checked and polished by a human.

Risky AI use cases (at least without oversight):

  • Revenue reconciliation

  • Accounting entries

  • Legal contract generation

  • Customs documentation

Use AI where speed matters more than perfection, and where increasing volume creates learning. Not where one error creates hours of rework.

Generative AI is a force multiplier — if you already know what you're doing

One of the most powerful (and under-appreciated) things about generative AI is that it amplifies your existing skill set.

  • If you're already a coder, it makes you faster.

  • If you're doing data analysis, it helps you write queries without the errors.

  • If you think in systems and flowcharts, AI will be your best tool yet.

But if systems thinking isn’t second nature? AI won’t magically give you that clarity. In fact, it might make things harder — especially if you’re trying to automate something like your finance workflows without knowing what all the edge cases are.

What about the middle?

Some roles fall in between — like RevOps, CX, and certain operational workflows. These can be partially automated using AI, but success often depends on how well-defined your SOPs are.

If you're creating business transactions from AI output (e.g. generating invoices or updating inventory), you want those systems clearly mapped before you bring automation into the mix.

The bottom line

If you’re a founder or leader looking to save time with AI (and DIY'ing it for now), start with the parts of your business where “good enough” is still genuinely helpful.

Let generative AI be your junior assistant.
Not your accountant.


Need help figuring out which workflows are safe to automate?
At Crafty Crow, we advise on AI integration strategy and build done-for-you automation systems for growing startups — integrating with your existing tools and SOPs, no AI hype required.

👉 Let’s talk.

If you're a founder at a growing startup, you've probably heard “AI can do that” a dozen times in the last year. But what can generative AI actually do well? And where does it still fall short?

The short answer: it depends on the type of work you’re trying to automate.

Here’s a practical framework we use when thinking about where to start with AI in your business — especially if you’re trying to DIY some of your backend operations or sales workflows.

The AI spectrum: from creative to business-critical

Think of AI use cases as a spectrum:

On the left side:
Creative work — writing, brainstorming, marketing, sales outreach, even coding.
There’s more room for subjectivity, you can treat the creation process as a "black box", and the steps in the process can vary every time. AI gets you 80–90% of the way there, and that’s usually enough to move fast, test ideas, and iterate.

On the right side:
Mission-critical work — accounting, legal, compliance, customs filings.
You need perfect accuracy. One mistake can lead to a mess of downstream problems. That’s where generative AI struggles. It guesses. It’s probabilistic. And if you’re not validating every result, it’s risky.

Rule of thumb: start where "pretty good" is still useful

If you're trying to get started with DIY'ing generative AI (e.g. chatting with or integrating ChatGPT, Claude, Gemini, et al.), focus first on high-tolerance areas:

Good AI use cases:

  • Writing product descriptions

  • Generating blog or social content

  • Drafting outreach emails

  • Brainstorming campaign ideas

  • Cleaning up CRM notes

Ideally, all of these would be double-checked and polished by a human.

Risky AI use cases (at least without oversight):

  • Revenue reconciliation

  • Accounting entries

  • Legal contract generation

  • Customs documentation

Use AI where speed matters more than perfection, and where increasing volume creates learning. Not where one error creates hours of rework.

Generative AI is a force multiplier — if you already know what you're doing

One of the most powerful (and under-appreciated) things about generative AI is that it amplifies your existing skill set.

  • If you're already a coder, it makes you faster.

  • If you're doing data analysis, it helps you write queries without the errors.

  • If you think in systems and flowcharts, AI will be your best tool yet.

But if systems thinking isn’t second nature? AI won’t magically give you that clarity. In fact, it might make things harder — especially if you’re trying to automate something like your finance workflows without knowing what all the edge cases are.

What about the middle?

Some roles fall in between — like RevOps, CX, and certain operational workflows. These can be partially automated using AI, but success often depends on how well-defined your SOPs are.

If you're creating business transactions from AI output (e.g. generating invoices or updating inventory), you want those systems clearly mapped before you bring automation into the mix.

The bottom line

If you’re a founder or leader looking to save time with AI (and DIY'ing it for now), start with the parts of your business where “good enough” is still genuinely helpful.

Let generative AI be your junior assistant.
Not your accountant.


Need help figuring out which workflows are safe to automate?
At Crafty Crow, we advise on AI integration strategy and build done-for-you automation systems for growing startups — integrating with your existing tools and SOPs, no AI hype required.

👉 Let’s talk.

Sign up for our free newsletter


Get ideas for incorporating the latest in automation and AI in operations, directly to your inbox

Sign up for our free newsletter


Get ideas for incorporating the latest in automation and AI in operations, directly to your inbox

Sign up for our free newsletter


Get ideas for incorporating the latest in automation and AI in operations, directly to your inbox

Crafty Crow

AI agents and workflow automation SaaS for Operations teams

Crafty Crow

AI agents and workflow automation SaaS for Operations teams

Crafty Crow

AI agents and workflow automation SaaS for Operations teams