BUILD VS BUY · APR 22, 2026 · 8 MIN

AI Automation for Marketing Ops: What to Build In-House vs. Buy

Marketing ops in 2026 is a build-vs-buy question at every layer. This is the practical framework we use with CMOs to decide what to keep on Zapier or n8n, what to build on Claude Code, and what to license from a vendor.

The marketing ops layer (the connective tissue between Shopify, Klaviyo, the affiliate platform, the CRM, the warehouse, and the AI tools) is where most marketing teams spend the most time and where most CMOs have the least visibility. The build-vs-buy decisions at this layer compound. A small wrong call in 2024 becomes an expensive cleanup project in 2026.

This piece is the framework we use with marketing leaders to decide which workflows belong on Zapier (or n8n), which belong as Claude Code workflows, and which should be licensed from a vendor. The decisions are not abstract. The wrong call typically costs $50K to $300K per year in operational drag.

The Three Tiers of Marketing Ops Workflows

Every marketing ops workflow falls into one of three tiers, and the right tooling decision depends on which tier the workflow is in.

Tier 1: Plumbing

Simple data movement between systems. "When a Shopify order is created, add the customer to Klaviyo." "When a TikTok Shop affiliate produces a sale, log it in the warehouse." "When a creator agreement is signed, create a record in the CRM."

These workflows have a clear input, a clear output, and minimal logic in between. They are ubiquitous. A typical DTC brand has 30 to 80 of them.

The right tool for tier 1: Zapier, Make, or n8n. The decision between them is mostly about price and team familiarity. Zapier is the most expensive and has the deepest integration library. n8n is open source, requires self-hosting, and is meaningfully cheaper at high volume. Make sits in the middle.

What does not belong here: anything with branching logic that gets complicated, or anything that requires LLM calls to make a decision. Both fail the cost-and-reliability test on Zapier-class tools.

Tier 2: Logic-Heavy Workflows

Workflows that require real conditional logic, multi-step decision trees, or LLM calls to handle ambiguous inputs. "When a customer service ticket arrives, classify it, draft a reply, route it to the right team, and escalate if sentiment is negative." "When a paid ad's ROAS drops 20 percent week over week, pause the ad, flag the team, and produce a diagnostic summary."

These workflows are where the brand's specific intelligence lives. They cannot be cleanly mapped to a Zapier template. They benefit from being written in code, version-controlled, and tested.

The right tool for tier 2: Claude Code or a similar agentic-development environment, talking to the underlying systems via MCP. The cost is engineering time (1 to 4 weeks per workflow), but the workflows are durable, debuggable, and they compound as the brand's prompt library matures.

What does not belong here: workflows that vendors have already commoditized (e.g. customer service triage on Gorgias or Zendesk).

Tier 3: Specialized SaaS

Workflows that are deep, well-defined, and serviced by a vendor whose entire product is that workflow. Customer service triage (Gorgias, Zendesk, Front). Email send-time optimization (Klaviyo, Iterable). Affiliate management (Impact, Refersion, ShopMy). Ad creative testing (Motion, Pencil).

The vendors at this tier have spent years on the specific problem. They have specialized data, specialized features, and economies of scale. Building these in-house typically does not pay back.

The right approach for tier 3: license the vendor, integrate them into your stack via tier 1 (plumbing) and tier 2 (logic) workflows.

What does not belong here: anything that is more about your brand's specific intelligence than about the platform's commoditized capability. The "brand voice" of your customer service replies should still be your team's, even if the triage layer is Gorgias.

The Decision Tree

For each marketing ops workflow you are evaluating, run it through this tree.

  1. Is this workflow about plumbing data between systems? If yes, use Zapier or n8n. Stop.
  2. Is there a vendor whose entire product is this exact workflow? If yes and the vendor's product is mature, license them. Stop.
  3. Does this workflow require brand-specific intelligence (voice, claim constraints, brand context)? If yes, build in-house with Claude Code.
  4. Does this workflow require LLM calls and multi-step logic? If yes, build in-house with Claude Code.
  5. None of the above? Default to Zapier or n8n for simplicity.

The tree gets you to the right answer in most cases. The hard cases are the ones where the answer is "tier 3 plus tier 2," meaning you license the vendor for the commoditized layer and build your own intelligence layer on top. Most of the highest-leverage marketing ops workflows are this hybrid shape.

Where Zapier MCP Fits

Zapier's MCP server lets Claude Code trigger Zapier workflows and read from them. The practical effect: you can have Claude Code orchestrate complex workflows that include Zapier as one of the steps.

When this is useful: you have a Zapier workflow that is doing tier 1 plumbing reliably (e.g. pushing affiliate data to your warehouse), and you want a tier 2 workflow (e.g. an AI agent that monitors affiliate performance and produces weekly recommendations) to call into the Zapier workflow as a step.

When this is not useful: as a way to avoid building tier 2 workflows. The Zapier MCP server does not let you skip the tier 2 build. It lets the tier 2 build talk to the existing tier 1 plumbing.

Common Build-vs-Buy Mistakes

Five patterns we have seen consistently produce bad outcomes.

Mistake 1: Trying to Build Tier 1 In-House

Some teams build their own ETL and integration layer because "Zapier is expensive." For most brands, the engineering time to build and maintain a custom integration layer costs meaningfully more than the Zapier subscription it would replace. Zapier and n8n are commodity infrastructure. Use them.

Mistake 2: Trying to Buy Tier 2 From a Vendor

Some teams license an "AI marketing automation platform" that promises to handle complex workflows out of the box. The vendor's prompt library is generic, the brand voice is wrong, and the team ends up with workflows that do not produce. Tier 2 belongs in-house.

Mistake 3: Trying to Build Tier 3 In-House

Some teams build their own customer service triage or their own affiliate management system. The vendor at tier 3 has spent years on the specific problem and has features the in-house team will not match. Building tier 3 in-house typically takes 3 to 5x longer than estimated and produces something less mature than the vendor.

Mistake 4: Underestimating the Operate Cost of In-House Workflows

Tier 2 workflows have an operate cost. The prompt library degrades. The integrations break when systems update. The model behavior changes. Budget 0.25 to 0.5 FTE per 10 production workflows for ongoing maintenance.

Mistake 5: Letting Tier 2 Workflows Sprawl

Once one tier 2 workflow ships, the team naturally wants ten more. Without discipline, the marketing ops surface becomes a sprawl of workflows that nobody fully understands. Document them, version-control them, and run a quarterly audit of which ones are still earning their keep.

What This Costs Across a Typical Marketing Ops Stack

For a DTC brand at the $20M to $100M revenue tier:

The brands that get the build-vs-buy decisions right at this layer typically run a marketing ops budget that is 3 to 6 percent of revenue. The brands that get them wrong run higher and produce less.

What This Looks Like When It Works

A DTC apparel brand we work with had a marketing ops setup that was 80 percent Zapier (300+ active Zaps), 0 percent custom workflows, and 20 percent vendor SaaS. Total cost: $260K per year. Operational drag: high, with frequent broken Zaps and team frustration.

The reorganization moved them to: 30 percent Zapier (commodity plumbing only), 30 percent vendor SaaS (Gorgias, Klaviyo, Impact), and 40 percent custom Claude Code workflows for the brand-specific tier 2 work. Total cost after reorganization: $310K per year (slightly higher, mostly due to the engineering FTE addition). Operational drag: dramatically lower. Team time saved: roughly 15 hours per week aggregate.

The framing for the CMO conversation is that marketing ops is a build-vs-buy decision at every layer. Getting the tier-by-tier mix right is what separates a marketing org that operates cleanly from one that spends most of its time fighting its own tools.

If you want help running the marketing ops audit and rebalancing the tier mix, the diagnostic at clankersapp.com is where to start.