AFFiNCO Agency Stack 2026: Every Layer, Tool and Real Cost

Everybody publishes a tool stack. Almost nobody publishes the bill.

So here is ours. AFFiNCO runs 385 websites, 46 people across six countries and more than 60 enterprise clients.

This is every layer that keeps it up, what sits in each one, and where the money actually goes. More about the operation sits on the about page.

The thing that surprises people is which line is biggest. It is not hosting. It is not the SEO tools everybody argues about. I will get to it in the second section and it is not close.

✅ Bottom Line Up Front

How to read this

Three things before the tables.

• This stack is shaped by scale, not by taste. Most of these choices would be wrong for a five-person agency. We run 385 sites, and a lot of what looks like over-engineering is just what happens when a manual process meets three hundred repetitions.
• Every layer has a job, and some layers have two tools on purpose. Redundancy at the infrastructure level is not waste. When a provider has a bad week, having somewhere to move to is worth the duplicated cost.
• I have marked what we pay where I am willing to publish it. Where a figure is client-specific or commercially sensitive, it is not here. I would rather leave a gap than print something vague.

Infrastructure, hosting and CDN

The layer everybody underestimates until it is 3am.

Tool	What it does for us	Why it stays
Cloudflare	DNS, WAF, caching, security across the whole estate	Single control plane for 385 domains. Nothing else comes close
Cloudflare CDN	Edge delivery, default for most sites	Included, fast, one less vendor
Bunny CDN	CDN for media-heavy properties	Cheap per GB and genuinely fast. Where Cloudflare is not the right shape
Cloudways	Managed cloud for client and revenue sites	Managed layer without managed-host pricing. Pays a lifetime revshare too
DigitalOcean	Raw droplets for internal tools and app workloads	Predictable pricing, good API, easy to script against
Vultr	Secondary cloud, geographic spread	Locations DigitalOcean does not have
UpCloud	High-IO workloads	Genuinely faster storage. We notice it on database-heavy jobs
Hostinger	Shared hosting for small niche sites	The economics of 200 small sites only work at this price point

The thing worth copying here is the split. Small sites live on cheap shared hosting because their revenue does not justify anything else. Revenue sites and client sites live on managed cloud. Internal tools live on raw droplets.

Putting everything on one tier is the most common infrastructure mistake I see. You either overpay for the long tail or underserve the properties that matter.

Cloudflare in front of everything is not optional at this size. One place to change DNS, one place to block a bad actor, one place to see what is happening. The alternative is 385 separate conversations.

AI and inference — the biggest line

This is the section that changed our cost structure more than anything else in a decade.

Tool	What it does for us	Notes
OpenRouter	Model routing for AI running inside WordPress across the portfolio	~$1,200/month. Our single largest software line
Claude (Max)	Strategy, long-context analysis, anything client-facing, most of our internal tooling	The one I use personally, all day
ChatGPT (Max)	Second opinion, different failure modes, specific workflows	Different model, different blind spots
Grok	Fast queries, real-time-flavoured work	Cheap per task
Grok Bot	Agent-style automation on a cloud machine	Useful. See the security caveat below
Genspark (Max)	Research passes and deck building in one place	Go and find the material, then build from it
Abacus AI	Team access to multiple models under one seat	Cost-effective way to give a team model access

Why OpenRouter is the biggest line

Because we run AI inside the product, not just alongside it.

A large part of the portfolio has AI features built into WordPress. Content assistance, internal linking suggestions, on-page work, summarisation. Every one of those calls a model, and at 385 sites the calls add up fast.

OpenRouter sits in front of all of it. One API, one bill, and the ability to change the model behind a workflow without touching the workflow. When a cheaper model gets good enough for a given job, we move it and the saving is immediate.

That routing ability is the actual product. If we had wired every site directly to one provider, we would be locked to that provider’s pricing and that provider’s outages.

The honest number: roughly $1,200 a month, and it moves with usage. It is more than our hosting. It is more than our SEO tools. Nobody predicted that three years ago, least of all me.

One caveat on agent tools

Grok Bot gives each account a single shared cloud machine. Every bot and every plugin runs on it, and xAI’s own documentation says plainly not to treat separate bots as a security boundary.

So we use it, and we do not put payment credentials on it. That is the whole policy and it is not negotiable.

Data, scraping and enrichment

Tool	What it does for us	Notes
Firecrawl	Crawls and structures any site we take on	The tool I would fight hardest to keep
Apify	Actors for specific recurring extraction jobs	Where Firecrawl is the wrong shape
DataForSEO	Bulk SERP, keyword and backlink data via API	At our volume, manual tools are not physically possible
Custom API scheduling	Internal jobs that pull, transform and store on a schedule	Built in-house, runs on our own droplets

Why Firecrawl is the lifeline

It is the first thing we point at every new website.

Client onboarding, competitor analysis, a portfolio site nobody has opened in a year. The first move is always the same — crawl it properly and get structured output we can reason over.

Before this, that job was a person with a spreadsheet and a weekend. Now it is a job that runs while we do something else, and the output is clean enough to feed straight into an analysis.

If it disappeared tomorrow, several workflows stop that day. There are alternatives. None of them are drop-in, and the switching cost is measured in weeks.

That is a dependency and I am aware of it. Some dependencies are worth accepting.

On DataForSEO

This one changed how we handle an algorithm update.

When a core update lands, the first question is always the same: what actually moved, across how many properties, in which direction. With a UI tool you answer that for ten sites. With an API you answer it for 385.

Pulling that data through Claude is what made it usable. The API gives you a firehose. Something has to turn it into a sentence a human can act on, and that is the part that used to take a week.

Automation and orchestration

Tool	What it does for us	Notes
n8n	Long-running workflows, most of them years old	Self-hosted for anything touching client data
Boost.space	Data sync and centralisation between systems	Where n8n is not the right tool
Custom scheduling	Internal cron jobs and API pulls	Cheaper and more predictable than paying per execution

We still run n8n and I want to be straight about why. Not because it is the newest option. Because our old workflows work, they have been debugged over years, and rewriting working automation is a hobby, not a business decision.

It bills on full workflow executions rather than per step, which suits agent-shaped work where one job fires forty tool calls. The community edition is free to self-host, and that is what we use for anything touching a client.

The rule across this layer: if it touches client data, it runs on infrastructure we control. No exceptions, and that rule has cost us convenience more than once.

SEO and content

Tool	What it does for us	Notes
Semrush	Position tracking, backlink audit, competitive research	The backlink audit earns its place on its own
Surfer SEO	Content gap analysis against what is already winning	We use it to find the gap, not to hit a density target
DataForSEO	Bulk everything, via API	Covered above
Rank Math	On-page SEO across the WordPress estate	Runs on the majority of our sites

The honest note on this layer: it is smaller than people expect for an SEO agency.

We had more tools. We cut them, because most SEO software overlaps heavily and paying three vendors for the same rank data is a habit, not a strategy.

What survived did so because it does one thing nothing else does. Semrush’s backlink audit. Surfer’s gap analysis. DataForSEO’s volume. Rank Math’s presence on every install.

Social media and publishing

Tool	What it does for us	Notes
ContentStudio	Scheduling and publishing across client and owned accounts	The workhorse
Zernio	Publishing through official APIs, handles posting quotas	Quota handling matters more than it sounds
Various small tools	Niche jobs, specific platforms	Dozens of them, none load-bearing

The quota point deserves a sentence. Instagram caps API publishing at 100 posts per rolling 24 hours, carousels counting as one. At our volume that ceiling is real, and a scheduler that silently truncates a batch at 100 is a problem you find out about later.

And the long tail of small tools is a genuine mess. We have dozens. Most were bought for one campaign and never cancelled. I will come back to that in the weaknesses section because it is the least defensible part of this stack.

Design and creative

Tool	What it does for us	Notes
Higgsfield	Character-consistent stills and short video	Paid subscription. Genuinely good for avatars
OpenArt	Image generation and character work	Check the licence tier before you generate at volume
Magic Studio	Editing, cleanup, headshots	Formerly InstaHeadshots
Canva	Everything visual, for six years	Still there. Used far less than it was
InVideo	Video when a specific job calls for it	Occasional rather than daily

Canva is the interesting entry here. Six years as the default tool for anything visual, and it is now one of the least-used things in this stack.

Not because it got worse. Generative tools took over the part we used it for most.

Canva is now where we go when somebody needs to edit a thing rather than make one.

Paid media

Platform	What we run	Notes
TikTok for Business	Client and owned campaigns	Official partner
Meta Business / Meta Ads	The volume platform	Still where most of the spend sits
Google Ads	Search and performance campaigns	Obvious, unavoidable
Reddit Ads	Niche targeting, specific verticals	Underrated, cheap in the right niche
ChatGPT Ads	New surface, still learning it	Too early for me to have a strong opinion
Point2Web	Google Ads agency we work with	We do not do everything in-house, and this is one of them

On Point2Web specifically: we run a full-service agency and we still use another agency for some Google Ads work. That is not a contradiction.

Specialism beats breadth on paid search, and pretending otherwise costs clients money. If somebody is better at a thing than my team, using them is the correct answer.

Antidetect and proxies

Tool	What it does for us	Notes
Dolphin Anty	Profile management for multi-account work	Primary
GoLogin	Secondary profile manager	Different team, different preference
AdsPower	Third option, automation-friendly	Good API
NodeMaven	Residential proxies	Pays crypto only — worth knowing before you sign up
IPRoyal	Proxies, our main volume provider	We hold a dedicated coupon with them

Three antidetect browsers is more than anybody needs. I know. Different teams standardised on different tools at different times and consolidating them has never been urgent enough to actually do.

That is an honest answer rather than a good one.

Affiliate tracking

Tool	What it does for us	Notes
RedTrack	All affiliate and campaign tracking	The fastest we have used

One tool, no redundancy, and that is deliberate.

Tracking is the layer where switching is most painful. Every campaign URL you have ever published points at your tracking domain. Moving it means either breaking history or running two systems in parallel for months.

So we picked one, we picked carefully, and we stay.

RedTrack’s Builder tier gives two million events at $69 a month. For comparison, the best-known competitor charges $119 for one million. That gap is why we are here.

The thing to understand about tracker pricing is that an event is not a click. Most cloud trackers count impressions as billable events too, so a million events is not a million visits. Model your own funnel shape before you compare plan prices, because the plans are not comparable.

Newsletter and email

Tool	What it does for us	Notes
beehiiv	The AFFMaven newsletter, 69,572 subscribers	Where the audience lives
Emailit	Transactional sending	One of the best tools we run. Quietly excellent
MailerLite	Campaign sending on specific properties	Situational now rather than core

Emailit is the one nobody has heard of and it deserves the mention. Transactional email is invisible when it works and catastrophic when it does not. Ours works.

beehiiv holds the thing that matters most in this whole stack. 69,572 people who chose to hear from us, across 104 countries. Traffic is rented. That list is owned.

If I had to keep one asset from this entire article and delete the rest, it would be that list. It is not close.

CRM and internal ops

Tool	Status	Notes
Our own CRM	Current	Built in-house, fits how we actually work
Google Workspace	Current	Mail, docs, the boring backbone
Airtable	Dropped	Was the backbone for years
Google Sheets	Dropped for ops	Still fine for a quick calculation. Not for running a business

We built our own CRM and I would not recommend it to most people.

The reason it made sense for us is that our workflow is unusual. A media network and an agency share the same contact graph, with affiliate partners, sponsors, clients and publishers all in one place.

No off-the-shelf CRM models that well, and we tried several.

What killed the spreadsheets was not the spreadsheets. It was that we could finally build something better without hiring a development team for a year. That is a genuine change in what a small operation can do, and it happened in the last eighteen months.

Google Sheets in particular was a crutch. It was where anything without a home went, and things without a home multiply. I do not miss it.

Booking and scheduling

Tool	What it does for us
Cal.com	Client and partner calls
TidyCal	Secondary booking, bought through AppSumo

Small layer, nothing clever. TidyCal came from an AppSumo deal years ago and still works, which is the best thing you can say about a lifetime deal.

Analytics

Tool	What it does for us	Notes
Cloudflare Analytics	Server-side traffic across the estate	No client-side tag, no consent problem, no ad-block gap
Google Analytics	Where clients expect the numbers to be	Reporting, mostly
Heatmaps	Specific pages during CRO work	Situational, not always-on

Cloudflare Analytics being first is deliberate. It sits at the edge, so it sees requests that never reach a JavaScript tag. On a portfolio where a meaningful share of visitors block trackers, the edge number and the tag number disagree, and the edge number is closer to the truth.

We still run Google Analytics because clients want to see it and because the integrations matter. But when the two disagree, I trust the edge.

Landing pages

Tool	Status
Swipe Pages	Still in the stack, used far less
Everything else	Dropped

This is the layer I would now argue you may not need at all.

We used more page builders than I want to list. Today, most of our landing pages come out of AI-assisted builds — described, generated, refined, shipped — and the result is faster and better-fitting than a template.

The builder became the slow step. When the constraint was design ability, a builder solved a real problem. When the constraint is speed, dragging blocks around is friction.

Swipe Pages stays for specific jobs. I am not sure it survives another year, and that is not a criticism of the product.

Invoicing and accounting

Tool	What it does for us
Zoho Books	Invoicing, primary
QuickBooks	Accounting, where the accountants want it

Two tools because the people who send invoices and the people who file accounts want different things. Consolidating would make one of those groups unhappy and save almost nothing.

Payments and payouts

Tool	What it does for us	Notes
Wise	Multi-currency, international	Our favourite. Best rates, least friction
Payoneer	Where networks pay by default	Many affiliate networks only pay here
PayPal	Legacy and small payments	Still unavoidable
Skydo	Receiving international payments into India	Built for exactly this problem
Stripe Atlas	Company structure	Infrastructure, not a payment rail

Wise is the one I would tell anybody to start with. Running a business across six countries means currency conversion is a real cost line, not a rounding error, and the difference between a good provider and a bank is measured in percentage points.

The reason there are five entries here is that you do not get to choose how people pay you. Affiliate networks pay where they pay. Clients pay how their finance department allows. You hold accounts because the other side holds accounts.

Skydo is worth naming for anyone operating from India. Receiving international payments into an Indian account has specific friction, and a tool built for that specific problem beats a general one.

Virtual cards

Tool	What it does for us
RedotPay	Cards funded from crypto, issued on demand
E.PN	Virtual cards for ad accounts
PST.net	Virtual cards, media buying
KAST	Business account with low fees and one-click crypto conversion

Why an agency needs four virtual card providers is a question with a boring answer: ad platforms decline cards, and they decline them unpredictably.

Running paid media at any scale means having more than one way to fund an account. This is not sophistication, it is redundancy against a problem nobody has solved.

The crypto-funded cards are about speed, not ideology. Moving money into an ad account in minutes rather than days is worth a fee when a campaign is live.

Domains

Registrar	Use	Why
Porkbun	First choice for new domains	Best pricing, clean interface, no upsell theatre
Namecheap	Second choice, bulk	Good at volume, reliable
GoDaddy	Legacy holdings	Where old domains live

Recommended in that order, and the order matters. Porkbun for anything new. Namecheap when buying in bulk. GoDaddy because some domains have been there for a decade and moving them is a job nobody wants.

Renewal pricing is where registrars get you. The first year is a promotion. Check year two before you buy 40 domains, because at portfolio scale that difference is a real number.

What we dropped and why

More useful than the list of what we kept.

Dropped	Why	What replaced it
Airtable	Outgrew it. Wanted our own data model	Our own CRM
Google Sheets for ops	Became the place things went to be forgotten	Our own CRM
Most landing page builders	The builder became the slow step	AI-assisted builds
Several SEO tools	Three vendors, one rank number	Semrush and DataForSEO
Canva as the default	Generative tools took the job it did most	Higgsfield, OpenArt, Magic Studio
A long list of small social tools	Bought for a campaign, never cancelled	Still cancelling, honestly

The pattern across every one of those is the same. We did not drop a tool because it got worse. We dropped it because the job it did stopped being a job, or because something upstream made it unnecessary.

Airtable is the clearest case. It was genuinely good and it was genuinely the right answer for years. What changed was that building the exact thing we wanted stopped requiring a development team, and at that point a flexible database became a compromise rather than a solution.

The small social tools are the least defensible line in this article. Dozens of subscriptions, bought for specific campaigns, each one too small to notice on a statement. Collectively not small at all.

If you take one operational lesson from this whole piece, make it that one. The subscriptions that hurt are never the big ones. You review the big ones. Nobody reviews an $11 tool.

Who should not copy this

Questions people actually ask

Stop collecting tools. Start naming the job each one does.

Every layer here exists because something broke under repetition. Nothing here was bought in anticipation.

Evidence beats hype. Every time.