Build vs buy an AI marketing agent: which is right for your team?

The decision is about your team, not the demo

"Should we build our own AI marketing agent, or buy one?" is now the default question in most growth-ops conversations. That is fair — a couple of years ago the build path barely existed, and today a capable engineer can wire an AI agent to live ad-platform data in an afternoon. So the question is real, and it is a good sign that a team is asking it before making a commitment.

The mistake is answering it from the demo. A demo shows what is possible. The decision you are actually making is about what your team will own and operate for the next two years: which skills it needs, which parts of the stack it should maintain in-house, and how it will know whether the thing works. Frame it that way and the answer gets much clearer.

The short version of this guide: build the edge, buy the core. Your team's differentiated skills live in a thin layer it should build itself. The layer underneath — connections, scheduling, approvals, measurement — is a commodity, and re-building it is how teams accidentally turn a weekend prototype into a full-time job.

This guide is the decision-framework companion to our build-vs-buy framework, which goes deeper into the economics and the 90/10 rule. If you are weighing a specific build, read the honest cost of building there first — that is where most teams get stuck.

---

The 90/10 rule

Here is the heuristic to take back to your team:

The 90/10 rule: about 90% of an AI marketing agent is commodity — live connections to your channels, a scheduler, an approval gate, and a system that measures and records what happened. That core is the same for every team that builds one. The remaining 10% — how your team words recommendations, which reports it actually reads, which channels it prioritises — is what is genuinely about your business.

Spend your own building on that last 10%. Rent or buy the first 90%.

Why the split matters is compounding. A commodity core does not get worse when you ignore it — someone else maintains it, compatibility keeps up with platform updates, and the integration set grows. A bespoke core gets more expensive every quarter: every API refresh, every schema change, every platform sunset is on your team. That is the difference between a system that compounds and one that slowly becomes a job.

This is not an argument against building. It is an argument about where to build. The productive debate in your team is where the line between the 90% and the 10% sits — not whether "build" or "buy" is the correct religion.

---

When building is the right call

Building an AI marketing agent in-house can be the right call, and it is worth naming the situations where it clearly is:

• The 10% is your product. If your edge over competitors is the analysis or the recommendations themselves — if the reasoning is what you sell — then the reasoning layer is not a commodity, it is your moat, and you should own it.

• You already run the core. If your team already maintains the integrations, the scheduler and the data plumbing for other reasons, the marginal cost of adding an agent on top is low, and staying in-house avoids layering a second system.

• You have a dedicated engineer to own it. The honest requirement is not "someone to build a demo" — it is a named owner who keeps it alive, handles the ever-changing platform APIs, and fixes it when it quietly breaks. If that person exists as a real role, building can genuinely pay off.

• The 90% can't meet your needs. If you need data handling or compliance before anything touches a live account, and you must control every line, then buying a core that hides its internals is not acceptable. In that case, you build — not because it is cheaper, but because the risk profile demands it.

If none of those apply, the build path is usually a trap dressed up as ownership.

---

When buying is the right call

Buying is right when the core layer is identical for every team and running it at home is pure cost:

• You want a working system, not a project. If the goal is marketing results — more measured, shipped changes per week — rather than the satisfaction of owning the plumbing, a working core gets you there in days, not quarters.
• Team size does not include an owner for it. The classic failure is a team of three marketers with one shared contractor: the demo is exciting, the prototype ships, and then nobody is paid to keep it alive. If there is no quiet, continuous owner, buying the core is the honest option.
• The reasoning stays yours anyway. A well-built bought core does not lock you out of the 10%. You keep your model choice and your wording, and you carry the agent's reasoning yourself — the harness just handles the commodity underneath.
• You can check its work. A worthwhile buy measures what it ships and shows you the record, misses included. If a tool only shows you the wins, you cannot audit it, and you should not trust it with real budget.

The trap to avoid is buying a "black box" that hides the reasoning and makes promises you cannot check. Buy the core as a working system you can drive — not a locked cage around the part that is yours.

---

Team-size guidance

Rough guidance, so you can check yourself before the debate runs long:

• A solo founder, or a team of 1–3. There is no world where a small team with no dedicated engineering owns a live multi-channel integration layer and the AI agent on top of it — and also does the marketing. Buy the core, spend your hours on the 10%.
• A growth team of 3–10 in a product company. If you already have an engineer comfortable with API integrations, the build line can move. Keep the buy for your core, and let them build on top of it — a dashboard you want, a report style you want, a channel you want treated differently. The core stays a commodity; the 10% gets bespoke.
• A team with a product or agency model. When the reasoning is what you sell, invert the plan: own the reasoning layer outright, and rent the commodity connections and measurement around it. This is the one shape where "build" is the default for the edge — and still "buy" for the 90%.

The patterns are the same at every size: the smaller the team and the less an eng owner exists, the more you buy; the more the reasoning is your product, the more you build the edge.

---

The one-sitting checklist

If your team is split on this today, run the decision in one sitting instead of letting it drag:

1. Name the core. List the channels your marketing actually depends on — Google Ads, Meta, GA4, Search Console, your warehouse. Add scheduling, an approval gate, and outcome measurement on top. That list is your true "build" number.
2. Apply the 90/10 rule. Ask honestly: is most of that list identical to what any competent builder would produce for any other business? If yes, it is candidate-for-buying, not candidate-for-building.
3. Find the owner. Name the person who would keep a home-built core alive after the demo. If the name comes back blank, that is the decision made for you.
4. Protect the 10%. Write down the few things that are genuinely yours — your recommendation style, your Monday report, your priorities. Keep those in your control, whether you build or buy.
5. Check who measures. Whatever you choose must tell you, for each shipped change, whether it worked — a verdict, misses included. If neither option can show you its misses, keep looking.
6. Decide on a month of data. Commit to a small, time-boxed experiment — a free start, a low-tier plan, or a single hand-built script — and let the measured outcome, not the demo, make the call.

---

How this fits the rest of the series

This piece is about the team and the decision. For the full economic argument and the deeper definition of the 90/10 rule, read the anchor of the build-vs-buy series. If your real question is the opposite one — not build-vs-buy but whether an AI agent can run marketing operations at all, day to day — start with can an AI agent run marketing operations.

---

The bottom line

Build the edge; buy the core. Your team's differentiated 10% — the reasoning, the wording, the priorities — deserves your own building time. The commodity 90% — connections, scheduling, approval, measurement — deserves to be bought or rented, and to stay someone else's maintenance problem.

Cogny exists to be that pre-built core you can drive with your own AI. Your people decide; every change gets a verdict, misses included. See the plans on our pricing page, and ship a change to see it measured.

Start free — no card required.