How to Choose an AI Marketing Agency: The 12-Point Checklist

Vetting an AI marketing agency in 2026 is harder than it looks. Most have impressive websites, generic case studies and hourly rates that vary by 5×. Here is the exact checklist we'd run if we were the buyer.

Score each one 0-2 (no / partial / yes). Anyone under 18/24 is high risk.

1. Have they shipped a real product, not just a deck?

Look at three live URLs they built in the last 12 months. If they hand you screenshots and pitch decks instead, you're hiring marketing storytelling, not engineering.

Our public work is one click from the homepage. That's a deliberate filter.

2. Can they explain their AI cost model in one sentence?

"Roughly $X per generated lead, $Y per content piece, $Z fixed monthly." If they can't, they don't track unit economics — and you'll be the one absorbing the variance.

3. Who owns the IP after engagement?

Some agencies retain ownership of "their methodology" (read: prompt files, code). Insist on full IP transfer for any code or prompts produced for you. This single contract clause has saved every client we've seen change agencies.

4. What happens if their main model provider doubles prices?

In 2025 OpenAI cut prices 10× in two years. In 2026 some providers raised them. A serious agency has a multi-model architecture — Claude / Gemini / GPT — and can swap providers without breaking your stack. Ask for the abstraction layer.

5. How is their staff exposed to your data?

If they build agents over your CRM, your customer data is going through their infrastructure. Ask about encryption, audit logs and SOC2 status. Smaller agencies often skip this; the right answer isn't "we're SOC2", it's "here is our DPA and here are the model providers we route data through."

6. Can they show a working internal tool they built for themselves?

The fastest filter for "real builders" is whether they eat their own dog food. We run our own AI CRM. An agency that tells clients to adopt AI but runs on Mailchimp is selling, not practising.

7. What is their handover process?

When the engagement ends — and it should, for any healthy build — what do you receive? Repository access? Deployment runbooks? A 1-hour video walkthrough? "Ongoing managed services only" is a red flag.

8. Do they cap deliverables in tokens, hours or outcomes?

Three different worlds. Token caps protect their margin, hour caps protect yours, outcome caps share risk. The most senior agencies will discuss outcome-based engagements. The greenest will only quote in hours.

9. How do they handle hallucination?

Every model still hallucinates. The right answer isn't "our model is too good for that." It's "we have a confidence threshold, a human-in-the-loop layer for X, and a logging system that surfaces low-confidence outputs."

10. How big is their backlog?

A 4-person team booked 3 months out can't start your project until July. If you need pipeline this quarter, you need to know.

11. Do they invoice in your currency / handle GST?

Tedious but real. International agencies often quote USD but bill in odd ways. A local agency saves you 3-7% in FX and tax friction.

12. Is there a kill switch?

Month-to-month after the initial engagement, with a defined notice period (30-60 days). Anyone insisting on a 12-month lock-in for a build engagement is selling lock-in, not value.

---

What good looks like at the end

A great AI marketing agency engagement leaves you with:

• A working system on your own infrastructure
• A clear cost model per lead / content piece / customer
• Code, prompts and IP transferred in full
• A 6-12 month roadmap of independent improvements you can run yourself

If you'd like a frank conversation about which projects on our work page most resemble what you're trying to build, contact us.