How a small AI-first marketing agency runs in 2026

Running a small marketing agency in 2026 looks nothing like it did three years ago. At Choco Media, we operate as a compact, AI-first team out of Rovaniemi — and our small marketing agency operations model is built around a simple premise: do the thinking that only humans can do, and let AI handle the rest. This post is a full, honest breakdown of how we actually run — the tools, the rhythms, the decisions, and the parts we’re still figuring out.

If you’re a solo founder, a small agency owner, or someone considering building a leaner version of the classic agency model, this is for you. We’re not going to tell you AI is magic. We’re going to show you exactly what our week looks like, where AI earns its place, and where we still rely on judgment calls that no model gets right consistently.

This is also a living document for us — a checkpoint on what’s working in mid-2026, written for the people who run businesses like ours.

The Core Structure: Two People, Three Layers

We don’t have a large team. We have a core of two people doing strategic and creative work, supported by a layer of AI tools that handle volume, and a thin outer layer of trusted specialists we bring in for specific deliverables — video production, technical development, media buying at scale.

This three-layer model is the thing we get asked about most. It’s not outsourcing everything to AI. It’s not hiring a big team. It’s designing the agency so that human attention stays on the work that compounds over time: client relationships, strategic decisions, creative direction, and quality control.

• Layer 1 — Core: Strategy, client relationships, QA, and anything that requires context about the client’s business
• Layer 2 — AI: First drafts, research, briefs, reporting, ideation, content expansion, scheduling, and workflow automation
• Layer 3 — Specialists: Deliverables that require physical production, technical depth, or platform-level expertise we don’t maintain in-house

The boundary between layers is the most important design decision we’ve made. When it blurs — when we let AI touch things it shouldn’t, or when we over-invest human time in things AI handles fine — quality drops and hours spike. Keeping the layers distinct takes deliberate discipline.

How We Structure the Week

We run on a loose weekly rhythm that gives structure without rigidity. The key insight from the past two years: predictable internal rhythms matter more than any individual productivity tool.

Monday: Intake and Orientation

Monday mornings start with a shared status read — what’s live, what’s in review, what’s scheduled to go out this week. We use Notion as our operating system. Every active client has a Notion workspace with a weekly status page, and we review them together before doing anything else.

• Review client status pages (30 min)
• Triage any weekend messages or platform alerts
• Set the week’s three priorities per client (not more than three)
• Confirm what’s going to AI for drafting vs. what needs human-first work

Tuesday–Thursday: Deep Work and Delivery

These three days are protected for execution. We try not to schedule external calls before noon. AI-generated first drafts come back during this window, we review and edit them, and deliverables move to client review. For content-heavy clients, Tuesday is typically when we run AI production — briefing, generating, QA’ing batches of content in one focused session rather than drip-feeding it through the week.

Friday: Review and Forward

Friday is for closing loops. We send our weekly client digest — a short update on what went out, what’s in pipeline, any notable data from the week. We also run a brief internal retrospective: what worked, what didn’t, anything to change next week. This ritual has saved us more times than any individual process change.

“The agencies we watch fail don’t fail because of bad tools or bad strategy. They fail because they lose the rhythm — reactive work takes over and proactive thinking disappears. Protecting Friday for review is the single operational habit we’d fight hardest to keep.”

The AI Tools We Actually Use Day-to-Day

We’ve tested a lot of tools and settled on a short list. This isn’t the definitive AI toolkit — it’s ours, chosen for this specific way of working. We’ve covered the full tool stack in more depth in the exact AI tools we use in production, so here we’ll focus on the operational role each plays.

For Writing and Content

• Claude (Anthropic): First-draft content, briefs, internal documents, strategy memos. We prefer Claude for long-form work because it holds tone better across a full document.
• ChatGPT: Ideation, variation generation, quick research synthesis. Better for generating multiple options fast when we’re not sure of direction yet.
• Notion AI: In-context editing and summarization inside client workspaces. Saves context-switching when the work is already in Notion.

For Research and SEO

• Perplexity: Real-time research with citations. We use it instead of a browser search when we want sourced answers fast.
• Ahrefs: Keyword and competitor research. Still the most reliable for understanding search landscape before writing anything.

For Workflow Automation

• Make (formerly Integromat): Connects platforms, triggers workflows, handles the “if this then that” logic that keeps our production pipeline moving without manual hand-offs.
• Zapier: Simpler automations where Make would be overkill. Usually client-facing — form fills, notification triggers, CRM updates.

Client Management: What We Changed in 2025

We made two significant changes to how we manage clients in the past year, and both came from pain rather than planning.

The first was introducing a proper onboarding sprint. We used to start producing too fast. Clients wanted to see output immediately, and we wanted to deliver it. The result was content that sounded like us more than it sounded like them. Now we do a structured 14-day onboarding process before any production begins — brand voice calibration, audience documentation, competitive positioning review. The content that comes after is consistently better.

The second change was shifting from deliverable-based reporting to signal-based reporting. Instead of “we published 12 posts this month,” the Friday digest now leads with: what’s ranking, what’s converting, what we’re adjusting and why. Clients care about business outcomes, not activity metrics. The shift changed how they perceive our value — and how we think about our own work.

• Deliverable counts → Signal and outcome metrics
• Ad-hoc communication → Weekly digest, async by default
• Open-scope retainers → Defined scope with a change-order process
• Vague briefs → Structured brief template that AI can actually execute on

If you’re running a retainer model and want to see how we structure quarterly reviews with clients, that’s something we’ll cover in detail in an upcoming Studio Notes post. Our bespoke retainer service is built around exactly these principles — defined scope, signal-based reporting, and outcomes over activity.

How We Price and Filter Work

We say no to a significant portion of the leads that come in. Not because we’re precious about it — because the wrong client at the wrong scope costs more than the revenue it brings. We’ve learned this the hard way.

Our filter questions at the discovery stage:

1. Is there an internal champion — someone who will actually use and advocate for the work?
2. Does the client have a clear enough sense of their target customer that we can build content and messaging with real specificity?
3. Is the timeline realistic? (Clients who want results in 30 days from SEO work are never satisfied, regardless of what we do.)
4. Is there budget alignment — not just for our fee, but for the media spend or production costs required to make the strategy work?

If any of these four are a clear “no,” we either adjust the scope until they’re all yes, or we pass. This sounds simple. In practice, it requires discipline every single time.

On pricing: we don’t bill by the hour. We never have. Hourly billing creates the wrong incentives for an AI-assisted agency — the more efficient your tools make you, the less you earn. We price by scope and outcome. The fee reflects the value of the output, not the time it took to produce.

The Operational Infrastructure

Here’s the actual stack that holds the operation together:

• Notion: Central operating system — client workspaces, content calendars, brief templates, internal documentation
• Slack: Internal communication only. Clients communicate via email or their preferred channel, not our Slack.
• Google Workspace: Email, Docs for client-facing deliverables, Sheets for reporting data
• WordPress: The blog and service site you’re reading now — we run it ourselves rather than relying on a third-party CMS we don’t control
• Linear: Task and project tracking for active deliverables. Simple, fast, low overhead.
• Make: Automation backbone — connects intake forms, Notion, email, and publishing workflows

Total SaaS cost for this stack: under €600/month. We’re deliberate about not adding tools unless they solve a specific problem that’s costing us more than the tool costs.

What We’ve Stopped Doing

In some ways, the best operational decision we’ve made is the things we’ve removed. A partial list:

• Time tracking: Stopped when we moved off hourly billing. Was producing anxiety, not insight.
• Social media presence for ourselves: We have a profile, but we don’t post consistently. Our clients come through search, referrals, and this blog — not our LinkedIn activity.
• Pitching for work: We haven’t done a formal pitch deck in over a year. Discovery calls lead directly to a proposal, or we pass.
• Vanity metrics in reporting: Impressions, reach, follower counts. These are out of our client reports unless the client specifically asks.
• Weekly calls as default: We moved to bi-weekly client calls for most retainers. Most of what was covered in weekly calls can be read in the Friday digest.

What We’re Still Working On

We want to be honest about this. There are things about our operational model that aren’t solved.

Capacity planning is the persistent hard problem. Because our output per hour is high (AI handles volume), it’s easy to take on more clients than we can actually give thoughtful attention to. We’ve hit that ceiling twice. The fix isn’t clear yet — we’re experimenting with a hard cap on active retainer clients and being more disciplined about what goes in the “Layer 3 specialist” bucket.

Knowledge management is also unsolved. We produce a lot of strategy documents, briefs, and research for clients. Very little of it is systematically captured in a way that makes future work faster. We know what good knowledge management looks like — we just haven’t built it yet.

And handoffs. When a specialist comes in for a deliverable, the brief-to-delivery-to-QA loop is still more manual than it should be. We’re building automation for this but it’s not done.

Running Lean on Purpose

The model we’ve described isn’t for everyone. There are clients and projects that require a larger team, more specialization, or a physical presence. We’re not that agency, and we don’t try to be.

What we’ve found is that staying small allows us to stay honest — with clients about what we can deliver, with ourselves about what’s actually working, and with the work itself. When there are only two people in the room, there’s no place for projects to hide.

Our AI content creation service and the broader AI automation work we do for clients are direct exports of the operational model described in this post. We’re not selling a system we don’t use. We’re using what we sell.

If this model resonates with where you want to take your business — or if you’re trying to figure out where AI actually fits in your agency operations — we’d be glad to think through it with you. Drop us a line, and let’s have a real conversation about it.