Build × Buy × Prompt
Framework by
Osama Romoh
Founder, Inflekt. 20+ years in marketing, builds AI systems. This framework comes from real work with real startups, not a whiteboard session.
Every startup founder right now is staring at the same question: "Should we use AI for this?" And the answer they're getting from the market is useless. Vendors say buy their tool. AI influencers say prompt everything. Developers say build it custom. Nobody's giving founders an honest framework for deciding, because everyone has something to sell.
The result is predictable chaos. Founders are either over-engineering custom AI solutions for tasks that a well-crafted prompt handles in 30 seconds, or they're duct-taping together a Frankenstein stack of 14 SaaS tools when what they actually needed was one custom pipeline. Both paths burn money. Both waste time. And for a startup between seed and Series A, time and money are the only two things you can't afford to waste.
This framework exists because the "should we use AI for this?" question is wrong. The real question is: "What's the right level of AI investment for this specific task?" That's a question with a clear, repeatable answer.
Three Mistakes Founders Make
Building everything custom
The founder has a technical co-founder, so every marketing problem becomes an engineering project. Three months later, the model works but the startup still doesn't have product-market fit.
Buying tools they don't need
Signed up for the annual plan at a conference, and now the team uses it to generate blog posts they could've written with ChatGPT in 10 minutes. We've audited startups paying $3,000/month across seven AI tools when they needed two.
Prompting when they should build
The marketing manager spends four hours a day copy-pasting data into ChatGPT, formatting the output, checking it, and pasting it into another system. Every single day. That's not using AI. That's being a human API connector.
The Framework
X-axis: Strategic Value
How much does this task's quality directly impact revenue, differentiation, or competitive advantage?
Y-axis: Frequency
How often does this task need to happen? Daily? Weekly? Once a quarter?
PROMPT
High Frequency + Low Strategic Value
BUILD
High Frequency + High Strategic Value
SKIP
Low Frequency + Low Strategic Value
BUY
Low Frequency + High Strategic Value
Score any marketing task on frequency and strategic value, land in a quadrant, make the decision.
Quadrant Deep Dives
BUILD
High Frequency + High Strategic Value
Tasks that happen constantly, directly impact your growth, and need to reflect your startup's unique context: your data, your customers, your positioning.
Decision Criteria
- You're doing this task daily or multiple times per week
- The output directly touches customers or drives revenue
- Generic solutions produce noticeably worse results than something tuned to your data
- You have (or can get) proprietary data that makes a custom solution meaningfully better
Risk
If you PROMPT something that belongs in BUILD, you're leaving money on the table. Your competitor who automated this is moving faster than you.
Examples
- Lead scoring and qualification.A custom scoring model trained on YOUR conversion data outperforms any generic tool.
- Customer segmentation for campaigns.Real-time segmentation built on your actual user behavior data beats any off-the-shelf tool.
- Personalized email sequences.Real personalization where content adapts based on user behavior and segment. At volume, this is a BUILD.
PROMPT
High Frequency + Low Strategic Value
Tasks you do all the time, but where "good enough" is genuinely good enough. The output doesn't need to be perfect or uniquely yours. It just needs to exist and not be terrible.
Decision Criteria
- You're doing this task frequently (daily or weekly)
- The quality bar is "competent," not "exceptional"
- A human still reviews the output before it goes anywhere
- Building automation around it would be over-engineering
Risk
If you BUILD something that belongs in PROMPT, you just spent $15K and six weeks automating something ChatGPT handles in a browser tab.
Examples
- First-draft social media posts.Prompt an LLM with your tone guidelines and topic, edit, post. Don't build a "social media AI engine."
- Internal summaries and briefs.Paste a competitor's landing page into Claude. Ask for a competitive analysis brief. Edit for 5 minutes. Done.
- Ad copy variations.Need 10 headline variations for a Google Ads experiment? That's a prompt, not a project.
BUY
Low Frequency + High Strategic Value
Tasks that matter a lot when they happen, but don't happen often enough to justify building custom solutions. When these tasks come up, you want them done well.
Decision Criteria
- This task happens monthly, quarterly, or per-project, not daily
- When it happens, the output quality is critical
- The problem space is well-established with mature solutions
- Building it custom would take longer than the task itself
Risk
If you PROMPT something that belongs in BUY, you get a mediocre result on something that actually matters. For high-stakes tasks, "close enough" isn't.
Examples
- SEO platform.You audit SEO monthly, not hourly. But you need real data: rankings, backlinks, technical issues. Buy Ahrefs or Semrush.
- Analytics and attribution.Setting up proper analytics is periodic. But getting it right determines whether you understand what's working.
- Marketing automation platform.Your quarterly campaign workflows need a real platform. HubSpot, Customer.io, Brevo, whatever fits.
SKIP
Low Frequency + Low Strategic Value
Tasks that barely matter and barely happen. These are the tasks that shouldn't even be on your to-do list, but somehow keep showing up in planning meetings.
Decision Criteria
- This happens rarely (quarterly or less)
- Even if you do it perfectly, the impact on revenue is negligible
- You're only considering it because "everyone else does it"
- A quick prompt handles it, or you can skip it entirely
Risk
If you BUILD or BUY for this quadrant, you're literally setting money on fire. We've seen startups spend $500/month on tools they check once a quarter.
Examples
- Quarterly industry reports.Prompt ChatGPT to draft it, edit for 20 minutes, ship. But honestly, does your seed-stage startup need one?
- Competitive messaging matrices.Nice to have. Not urgent. Spend 30 minutes prompting an LLM if it comes up.
- One-off event recaps.Just prompt it. Or don't do it at all. Your team of 8 doesn't need a newsletter.
Common Marketing Tasks, Plotted
| Task | Frequency | Strategic Value | Decision |
|---|---|---|---|
| Lead scoring / qualification | Daily | High | BUILD |
| Personalized email sequences | Daily | High | BUILD |
| Customer segmentation | Weekly | High | BUILD |
| Content repurposing (blog → social → email) | Weekly | High | BUILD |
| Chatbot / conversational marketing | Daily | High | BUILD |
| Social media post drafts | Daily | Low | PROMPT |
| Ad copy variations | Weekly | Low | PROMPT |
| Internal briefs and summaries | Weekly | Low | PROMPT |
| Blog post first drafts | Weekly | Medium | PROMPT |
| SEO keyword research + audits | Monthly | High | BUY |
| Analytics and attribution setup | Quarterly | High | BUY |
| Marketing automation platform | Monthly | High | BUY |
| Reporting dashboards | Monthly | High | BUY |
| Competitor deep-dive analysis | Quarterly | Medium | BUY / PROMPT |
| Quarterly industry reports | Quarterly | Low | SKIP |
| One-off event recaps | Rarely | Low | SKIP |
A note on "Medium" strategic value: Some tasks sit between quadrants. Blog post drafts, for example. The quality matters, but LLMs are genuinely good at first drafts now. The move is PROMPT it, then invest human editing time.
How to Use This
List your marketing tasks
Write down every marketing activity your team does. Be specific. Not "content marketing" but "write weekly blog post," "create LinkedIn carousel," "update email drip sequence." If you can't list at least 15, you're thinking too broadly.
Score each task on Frequency
How often does this actually happen? Daily, weekly, monthly, quarterly, rarely? Don't score how often it should happen. Score how often it does happen. Reality, not aspiration.
Score each task on Strategic Value
Ask: "If we did this 10× better, would it meaningfully impact revenue in the next 6 months?" If yes, it's high. If no, it's low. Be ruthless. Most tasks are lower strategic value than founders want to admit.
Plot and decide
Put each task in its quadrant. The decision follows. BUILD the top-right. PROMPT the top-left. BUY the bottom-right. SKIP (or quick prompt) the bottom-left.
Audit your current setup
Compare where your tasks landed vs. what you're actually doing today. Are you buying tools for things you should be prompting? Building custom solutions for tasks that belong in PROMPT? This gap analysis is where the money is.
Common Mistakes (And How to Fix Them)
The Over-Builder
Pattern
Everything becomes an engineering project. "Let's build a custom AI content pipeline." "Let's train a model on our brand voice." "Let's create a proprietary analytics engine."
Reality Check
You're a 12-person startup. You don't need proprietary anything yet. Build only when the frequency AND strategic value justify the investment.
Fix
Before any BUILD decision, ask: "Would a well-crafted prompt + 10 minutes of human editing get us 80% of the way there?" If yes, prompt it.
The SaaS Hoarder
Pattern
There's a tool for everything, and you've subscribed to all of them. $4,000/month in subscriptions, half of them logged into once a month.
Reality Check
Most of those tools overlap, and the ones you use rarely don't justify their cost. You're paying for capability you'll "get to eventually."
Fix
Audit every subscription against the framework. If the task it serves is low-frequency AND low-strategic-value, cancel it and prompt when the need comes up.
The Prompt-Everything Crew
Pattern
ChatGPT is the answer to every question. Lead scoring? "Paste leads into ChatGPT." Customer segmentation? "Ask Claude to group our users."
Reality Check
If someone on your team is spending 3+ hours a day on the same prompt-based workflow, you've crossed the BUILD threshold. Manual prompting doesn't scale.
Fix
Track time spent on prompt-based workflows. Anything exceeding 10 hours/week on the same type of task is a BUILD candidate. Automate it.
The "AI Can Do Everything" Delusion
Pattern
Trying to AI-ify tasks that don't need AI at all. "Let's use AI for our weekly team standup notes." Not everything benefits from AI.
Reality Check
AI should be applied where it creates leverage. Where it makes a $1 task produce $10 of value, or turns a 4-hour task into a 15-minute one.
Fix
Before applying AI, ask: "What's the leverage?" If it's "marginal time savings on something we do once a month," skip it. Focus on high-frequency, high-value work.
If you do it daily and it drives revenue → build it. If you do it daily but it's commodity work → prompt it. If it matters but you do it rarely → buy the best tool. If it barely matters and barely happens → skip it.
That's the whole framework. No 47-slide deck needed.