At one point last year, I sat down and added up every marketing SaaS subscription I was paying for. Reporting dashboards. Creative tools. Landing page builders. SEO platforms. Audience research tools. Campaign management add-ons. The total came to just under $2,000 per month.
That's $24,000 a year. For a solo agency. To run tools that, in most cases, did about 30% of what I actually needed and nothing else.
I don't have a philosophical objection to SaaS. I have a practical one. I was paying premium prices for general-purpose tools, then spending additional hours working around their limitations. The landing page builder couldn't do the exact layout I needed. The reporting dashboard couldn't calculate attribution the way I wanted. The SEO tool gave me data I had to export and reprocess anyway.
So over about six months, I replaced nearly all of it. The monthly cost went from ~$2,000 to roughly $60-100 in API fees. Here's exactly what I replaced, what I built, and what it actually costs.
The Old Stack
Here's what I was paying for before the switch. These are real numbers, rounded slightly for clarity.
| Tool Category | Example Tools | Monthly Cost |
|---|---|---|
| Reporting / Attribution | Triple Whale, Northbeam | $300-500 |
| Ad Creative | AdCreative.ai, Canva Pro | $200-400 |
| Landing Pages | Unbounce, Instapage | $100-200 |
| SEO | Ahrefs, SEMrush | $200-400 |
| Audience Research | SparkToro | $100-200 |
| Campaign Management | Various add-ons, time cost | $100-200 |
| Total | $1,500-2,000 |
Every one of these tools does what it says. I'm not claiming they're bad products. But they're built for the broadest possible audience, which means they're optimized for the average use case. When your workflow is specific, you spend half your time fighting the tool instead of using it.
The breaking point for me was attribution reporting. I was paying $400/month for a dashboard that still couldn't answer the question I actually cared about: "How much revenue did this campaign generate according to Shopify, not according to Meta?" I kept exporting CSVs and running the numbers myself. That's when I realized I was paying for a middleman between me and an API.
The Decision Framework: Build vs. Buy
Before I dive into the replacements, here's the framework I used for each tool. Not everything should be built custom. The question isn't "can I build this?" It's "should I?"
Build when:
- The API you need is free or cheap, and the SaaS is mostly a UI on top of it
- You need the tool to integrate tightly with your other tools
- The SaaS forces you into workflows that don't match yours
- You'll use it frequently enough to justify the build time
- The maintenance burden is low (stable APIs, simple logic)
Buy when:
- The tool does something genuinely hard that would take months to replicate (e.g., Ahrefs' backlink index)
- The cost is low relative to the time you'd spend building
- You only need it occasionally
- The tool has network effects or proprietary data you can't access otherwise
With that in mind, here's what I built.
1. Reporting Dashboards
Replaced: Triple Whale, Northbeam (~$300-500/month) Built: Custom Python reporting tools Current cost: ~$0/month
This was the easiest win. The reporting tools I was paying for are essentially API aggregators with a nice UI. They pull from Meta's Marketing API, Google Ads API, and Shopify's API, then display the data in dashboards.
The thing is, all three of those APIs are free.
I built a unified reporting system that pulls Meta and Google performance data into consolidated weekly reports. One command, one client name, and I get markdown and HTML reports for any date range. The system also generates Shopify reports — sales, customer lifetime value, cohort analysis — and critically, it calculates true attribution by matching Shopify orders by UTM source against actual ad spend.
That last part is what I was paying $400/month for and still not getting right. The custom tool matches UTM-tagged orders in Shopify to the campaigns that drove them, then calculates actual ROAS from first-party data. Not the inflated numbers that Meta and Google self-report. I wrote about why platform ROAS is misleading in a separate piece.
The reporting system also covers Amazon advertising, inventory tracking, and Klaviyo email performance. All from the same interface. All structured output that I can send directly to clients.
Build time: About two weeks for the core system. Another week for Amazon and Klaviyo integrations.
Maintenance: Minimal. Meta changes their API versioning occasionally, which requires updates. Maybe 2-3 hours per quarter.
2. Ad Creative Tools
Replaced: AdCreative.ai, Canva Pro (~$200-400/month) Built: Image generation CLI + design generator + behavioral science copy engine Current cost: ~$15-30/month in API fees
This one took the most work to build, but it also delivered the biggest upgrade in output quality. The SaaS creative tools I was using were template-based. You pick a layout, drop in your product image, add some text, and export. The output looks like... a template.
The replacement is a multi-tool pipeline:
Image generation: A CLI tool that generates marketing images from text prompts using multiple AI models through fal.ai. Product photography, lifestyle shots, hero backgrounds, ad creative. Different models for different jobs — fast cheap models for iteration ($0.003/image), high-quality models for final assets ($0.05-0.15/image). A full creative set for a campaign costs about $1-3 to generate, compared to $500+ per photoshoot or $50-100 in stock photos.
Design generation: A system that produces layered SVG designs from structured JSON briefs. Brand kits — colors, fonts, logos — are defined once per client and applied consistently across every output. It handles Meta ad sizes (1080x1080, 1080x1350), Stories, and Reels formats.
Copy engine: This is the part that SaaS tools can't touch. I built a conversion copy system that draws from 30+ cognitive biases and frameworks from direct response legends (Schwartz, Ogilvy, Halbert, Sugarman). It doesn't just generate "ad copy." It generates copy structured around specific behavioral science principles, targeted at specific awareness levels, for specific audience personas. Every variant comes with a breakdown explaining the psychology behind it.
UGC-style ads: A dedicated tool for generating user-generated content stills with consistent AI characters and Instagram/TikTok caption overlays. A full 9-ad UGC campaign across three audience segments costs about $1.80 to produce.
The total API cost depends on volume. For a typical client month — maybe 30-50 images, 20-30 copy variants — I spend $15-30 on AI API calls. That's the entire creative generation budget.
Build time: About six weeks across all the creative tools. The copy engine took the longest because getting the behavioral science frameworks right required extensive testing.
Maintenance: Moderate. AI model providers update their offerings regularly. I've switched primary models three times in the past year as better options became available. Maybe 4-5 hours per month.
3. Landing Page Builders
Replaced: Unbounce, Instapage (~$100-200/month) Built: Custom HTML landing page builder + Supabase-backed split testing Current cost: ~$5/month
Landing page builders charge $100-200/month for a drag-and-drop editor and basic A/B testing. The output is bloated HTML behind a subdomain you don't control, with conversion tracking that may or may not match your actual analytics.
The replacement is a component-based system for building conversion-optimized landing pages. Self-contained HTML files. No CMS. No WordPress. They load in under a second, they're deployed via GitHub Pages (free), and they message-match the ad creative exactly because they're built from the same campaign brief.
The split testing infrastructure is a Supabase-backed admin dashboard. I define tests through a browser interface — headline variants, CTA copy, hero images, layout changes — and the changes take effect on live landing pages immediately. No code deployment. No developer tickets. Visitors are assigned to variants automatically via weighted random allocation, assignments are sticky via localStorage, and the dashboard calculates statistical significance in real time using z-tests.
Want to test a new headline? I type it into the dashboard and it's live in 30 seconds. With Unbounce, that same change involves duplicating a variant, editing the page in their builder, setting traffic allocation, and publishing. Five minutes minimum, plus the cognitive overhead of navigating their interface.
Supabase's free tier handles the traffic volumes I work with. GitHub Pages is free. Total hosting and infrastructure cost: about $5/month for the Supabase project once you exceed the free tier limits.
Build time: About three weeks for the landing page system. Another two weeks for the split testing dashboard.
Maintenance: Low. The landing page builder is stable because it outputs static HTML. The split test system uses a simple Supabase schema that hasn't needed changes since launch.
4. SEO Tools
Replaced: Ahrefs, SEMrush (~$200-400/month) Built: Custom SEO platform using DataForSEO API + Claude Current cost: ~$50/month
This is the one where "build vs. buy" gets interesting, because tools like Ahrefs have genuinely proprietary data — their backlink index, specifically — that you can't replicate. I didn't fully replace Ahrefs. I replaced the 80% of it I was using daily and kept an occasional Ahrefs check for backlink analysis.
The custom SEO platform handles:
- Keyword discovery — via DataForSEO's keyword API, which gives you the same Google Keyword Planner data that SEMrush resells
- AI-powered keyword classification — Claude Haiku classifies keywords in batches. Is this relevant to the brand? Commercial or informational intent? Realistic ranking difficulty? This replaces hours of manual spreadsheet work
- Keyword clustering — Groups related keywords into content themes automatically
- Content generation — Full articles generated with the client's specific brand voice, positioning guidelines, and competitor framing rules injected into the prompt. The output reads like the brand, not like AI
- Link building prospecting — Finds relevant sites for outreach using SERP analysis and directory mining
- Publishing — Pushes content directly to Shopify blogs or GitHub Pages
DataForSEO charges by the API call, and their pricing is reasonable. For the volume I need — keyword research for a few clients, periodic SERP checks, competitive analysis — the monthly cost runs about $50. Claude API costs for keyword classification and content generation add another $5-10/month.
The big tradeoff: I don't have Ahrefs' backlink index. For link building research, I either use DataForSEO's backlink endpoint (which has a smaller index) or I buy a one-day Ahrefs pass when I need deep backlink analysis. This happens maybe once a month.
Build time: About four weeks for the core platform. Ongoing refinement as I add modules.
Maintenance: Moderate. DataForSEO's API is stable, but the AI classification prompts need tuning as I onboard new clients with different brand contexts. About 3-4 hours per month.
5. Campaign Management
Replaced: Time cost of manual campaign building in Ads Manager and Google Ads UI Built: Meta Ads CLI + Google Campaign Builder Current cost: ~$5-10/campaign in Claude API costs
This one doesn't have a clean dollar-to-dollar SaaS comparison because there isn't a SaaS tool that builds campaigns for you. The cost I was paying was time — clicking through Ads Manager, manually entering targeting, uploading creatives one by one, setting up A/B tests through the UI.
The Meta Ads CLI lets me define entire campaigns in structured JSON briefs and push them from the terminal. Campaign structure, ad sets, targeting, creatives, budgets — all defined in a file, reviewed in a dry-run preview, and pushed with a single command. Ads start paused by default. There's a confirmation prompt before anything goes live.
The Google Campaign Builder does the same for Search campaigns. Keyword research, RSA headline and description generation (with character limit enforcement), campaign structure, and deployment — all from JSON briefs.
The Claude API costs come from the copy generation step. When I'm building ad creatives, the system uses Claude to generate headlines, descriptions, and primary text based on the campaign brief and the behavioral science framework. For a typical campaign build — maybe 10-15 ad variations — the API cost is $5-10.
The time savings are harder to quantify but significant. Building a full campaign structure in Ads Manager with 5 ad sets and 3-4 creatives per ad set takes me about 2-3 hours of clicking. The same campaign, defined in a brief and pushed via CLI, takes about 30 minutes. Most of that time is strategic — deciding on structure and messaging. The mechanical work is nearly zero.
Build time: About three weeks for the Meta CLI. Two weeks for the Google builder.
Maintenance: This one requires the most attention. Meta updates their API quarterly, and breaking changes are common. Google's API is more stable but still requires periodic updates. I budget about 4-6 hours per month for API maintenance across both tools.
6. Audience and Interest Research
Replaced: SparkToro, audience research tools (~$100-200/month) Built: Custom Meta interest finder + market research engine Current cost: ~$0/month (Meta API is free)
SparkToro is a good product for what it does — show you where your audience hangs out online. But for Meta advertising specifically, what I need is targetable interests — the specific interests, behaviors, and demographics available in Meta's targeting system.
The custom interest finder takes a product brief (brand, positioning, competitors, customer demographics) and queries Meta's Marketing API to find targetable interest categories. It scores and ranks them by relevance, outputs them in tiers, and suggests ad set structures for testing.
The market research engine goes deeper. It pulls customer voice data from reviews, Reddit, forums, and social channels. Analyzes competitor positioning, offers, and creative patterns. Maps market sophistication. The output is a structured document that feeds into the persona builder, copy engine, and ad prompt generator.
Meta's Marketing API is free. The market research engine uses web scraping and search (also free). The AI analysis costs for a full market research pass on a new client are about $2-3 in Claude API calls.
Build time: About one week for the interest finder. Two weeks for the market research engine.
Maintenance: Very low. Meta occasionally changes available interest categories, but the tool queries the API live, so it always returns current options.
The Honest Tradeoffs
I'd be lying if I said custom tools are strictly better than SaaS. They're better for me, in my specific situation, with my specific skills. Here's what you give up.
You become the support team. When Unbounce breaks, you open a ticket and someone else fixes it. When my landing page builder breaks, I fix it. At 2 AM if necessary. For a solo operator who can code, this is manageable. For a team of non-technical marketers, it's a non-starter.
There's no UI for your team. My tools run from the terminal or through minimal browser dashboards. There's no polished interface with onboarding flows and tooltips. If I bring on a team member, they need to be comfortable with command-line tools and JSON files. That limits the hiring pool.
Build time is real. The tools described in this article represent roughly 4-5 months of part-time development. That's time I wasn't billing clients. The investment pays off over 12+ months, but the upfront cost is significant.
API dependencies are your problem. When Meta changes their API (which they do constantly), I have to update my code. When DataForSEO adjusts their pricing, I have to evaluate alternatives. SaaS tools absorb these changes for you. Custom tools put you on the hook.
Some things genuinely can't be replicated. Ahrefs' backlink index is built from years of web crawling. I can't replicate that. Triple Whale's integration with Shopify checkout extensibility goes deeper than what the standard API exposes. Some SaaS products have earned their price through proprietary data or infrastructure that would cost millions to build.
Debugging is harder. When a report looks wrong, I can't blame the software vendor. I have to trace through my own code, check API responses, verify calculations. This is actually a benefit for accuracy — I understand every number I report to clients — but it takes more time than clicking "refresh" in a dashboard.
The Math
Here's the actual before-and-after breakdown.
Before (monthly SaaS costs):
| Category | Cost |
|---|---|
| Reporting / Attribution | $400 |
| Ad Creative Tools | $300 |
| Landing Page Builder | $150 |
| SEO Platform | $300 |
| Audience Research | $150 |
| Misc. Tools / Add-ons | $200 |
| Total | ~$1,500 |
After (monthly API/infrastructure costs):
| Category | Cost |
|---|---|
| DataForSEO (SEO data) | $50 |
| fal.ai (image generation) | $15-30 |
| Claude API (copy, classification, research) | $10-20 |
| Supabase (split testing) | $0-5 |
| Hosting (GitHub Pages) | $0 |
| Meta / Google / Shopify APIs | $0 |
| Total | ~$75-105 |
Net savings: ~$1,400-1,425/month. ~$17,000/year.
That doesn't include the time savings, which are harder to quantify but equally significant. Campaign builds that took 3 hours now take 30 minutes. Weekly reporting that took 2 hours now takes 5 minutes. Creative generation that required external designers now happens in-house in an afternoon.
Where the Money Goes Now
The savings didn't just disappear into profit. Most of it got redirected into two places:
More ad spend for clients. When your overhead drops by $1,400/month, you can afford to charge lower management fees or allocate more budget to actual advertising. Several clients got budget increases funded entirely by my reduced operating costs.
Better AI tools. The Claude API costs for a full client campaign — market research, persona development, copy generation, image prompts, keyword classification — come to about $5-10. That's not a cost I'm trying to minimize. It's an investment that directly improves output quality. I'd rather spend $10 on AI-generated copy grounded in behavioral science than $0 on copy I wrote in 15 minutes because I was rushing.
The shift is from paying for platforms to paying for intelligence. SaaS subscriptions buy you access to someone else's interface. API costs buy you raw capability that you shape to your exact needs.
The Bigger Point
I'm not arguing that every agency should throw out their SaaS stack and start coding. Most agencies shouldn't. The build-vs-buy math only works if you can actually build the tools, maintain them, and your specific workflow benefits from the customization.
But I do think the default assumption — that you need a stack of $200-400/month subscriptions to run a modern marketing operation — deserves questioning. Most of these tools are interfaces sitting on top of APIs that you can access directly. The SaaS premium buys you convenience, support, and a UI. Whether that's worth $1,500/month depends entirely on your situation.
For me, it wasn't. The tools I built are faster, more integrated, more precise, and cost 95% less. The tradeoff is that I maintain them myself. For a solo operator with technical skills, that's a trade worth making.
If you're running a brand and wondering whether your agency's tool costs are showing up in your fees, they probably are. We keep our overhead low so your budget goes to advertising, not software.