How to Choose a Marketing Agency in the AI Era

Seven questions to ask marketing agency partners before you sign. Spot real AI capability, protect your CAC and ROAS, and avoid expensive pilots.

A CMO I know ran a bake-off last quarter. Three agencies, same brief, same $40k monthly budget. The first pitched a "proprietary AI engine" that turned out to be a GPT wrapper with a login page. The second had real tooling but no media buyers who could read a cohort retention curve. The third showed up with a working attribution model, a named team, and a plan to cut wasted spend inside 30 days. Guess who won. That pattern is everywhere right now. Every agency claims to be an AI marketing agency. Very few can prove it. The gap between AI as a slide and AI as a system is where most bad agency decisions get made, and it is widening fast. Analysts tracking [2026 marketing trends](/dgtg/blog/2026-digital-marketing-trends-what-agencies-must-know-now) keep landing on the same point: the winners will be the teams that operationalize AI across the whole funnel, not the ones with the loudest demo. So here is the checklist we use when founders and marketing managers ask [how to choose a marketing agency](/dgtg/blog/how-to-choose-a-digital-marketing-agency-tips-for-founders-cmos) that actually ships. Seven questions, in the order that matters. ## 1. Which specific workflows does AI touch, and what changed because of it? Vague answers here are a red flag. "We use AI throughout" means nothing. You want named workflows and measurable deltas. Ask them to walk through three: creative production, audience targeting, and reporting. Then push for before-and-after numbers. - Creative: How many ad variants ship per week now versus 12 months ago? What is the CTR spread between top and bottom performers? - Targeting: How are lookalike and propensity models built? What data feeds them? - Reporting: How fast does a spend anomaly get flagged? Hours or days? If they cannot name the tools and the humans who run them, you are buying a narrative. ## 2. Who owns the model, and what happens when it is wrong? This is the question most buyers skip. AI systems fail quietly. A bidding model drifts. A generative creative tool starts producing off-brand copy. A lead-scoring model overweights a channel that used to work. You want to hear a governance story, not a success story. Who reviews outputs weekly? What is the escalation path when ROAS drops 20% in a channel the model loves? What is the rollback plan? Agencies that treat AI as a set-and-forget layer will burn your budget before anyone notices. Agencies that treat it as a supervised system will catch the drift early. Ask for a recent example of a model that underperformed and what they did about it. The answer tells you more than any case study. ## 3. How do they connect AI output to CAC, ROAS, and LTV? Speed is not the goal. Margin is. Plenty of agencies now produce 10x the creative volume with AI, then fail to move CAC because volume was never the constraint. Bring your own unit economics to the meeting. Give them your current CAC, payback window, and LTV assumptions. Then ask how their AI stack changes each number. | Metric | What a real answer sounds like | What a weak answer sounds like | |---|---|---| | CAC | "We expect a 15 to 25% reduction in paid social CAC by month three via creative iteration and budget reallocation." | "AI helps us optimize spend." | | ROAS | "Blended ROAS target of 3.5, with channel-level floors so we do not starve upper funnel." | "We aim to maximize ROAS." | | LTV | "We model cohort LTV by acquisition source and shift budget toward high-LTV segments." | "LTV is more of a product team metric." | If they cannot speak your numbers, they will optimize for theirs. ## 4. What is the human-to-machine ratio on our account? AI does not replace judgment. It amplifies it. The best setups we see run lean teams with senior operators who know when to override the model. The worst run one junior account manager babysitting twelve dashboards and calling it automation. Ask directly: how many hours per week will a named human spend on our account? Who is on the escalation path? What is the ratio of strategists to accounts? A good answer sounds like: "One senior strategist at 15 hours a week, one media buyer at 10, plus a data engineer on call. AI handles reporting and variant generation. Humans handle budget decisions and creative direction." A bad answer sounds like: "Our platform does most of the work." ## 5. Can they show a case where AI failed and they fixed it? Every honest agency has one. The ones that do not are either new or lying. Listen for specifics. A model that over-indexed on a dying audience. A generative pipeline that produced 400 variants and zero winners. A forecasting tool that missed a seasonality curve. Then listen for the fix: retraining cadence, human review gates, new data sources. This question separates operators from presenters. ## 6. What do they want you to keep in-house? This is counterintuitive, and it is the fastest filter we know. Agencies that genuinely leverage AI know where their edge stops. Strong agencies will tell you to keep brand strategy, first-party data infrastructure, and product-led growth experiments in-house. They will take performance creative, channel execution, and measurement. Weak agencies will promise to run everything, which usually means they run nothing well. Our rule of thumb: - Hire an agency for: paid acquisition at scale, creative testing velocity, attribution and incrementality testing, and channel expansion you have never run before. - Keep in-house: positioning, pricing, lifecycle messaging, and any workflow that depends on deep product context. - Split it: content and SEO, where AI accelerates production but your team owns the point of view. ## Our take: what we recommend If you want a shortlist to benchmark against, here is what we would actually put in a bake-off in 2026. - For performance media with real AI infrastructure: **Smartly.io** for creative and media automation at scale, or **Motion** if you need creative analytics that tie directly to spend. - For measurement and incrementality: **Measured** or **Northbeam**. Both will tell you what the platforms will not. - For AI-native creative testing: **AdCreative.ai** is fine for volume, but pair it with a human creative director or you will drown in sameness. - For SEO and content at scale: **Surfer** plus an internal editor. Do not outsource the point of view. - For full-service with genuine AI ops: look for shops in the 30 to 80 person range with an in-house data engineer. Bigger is not better here. Specialists beat generalists. The pattern: pick tools that expose their logic, and agencies that can explain theirs. ## 7. What does month one look like, in writing? Ask for a 30-day plan with named deliverables. Not a roadmap. A plan. You want to see: audit scope, data access requirements, first experiments, reporting cadence, and the specific metrics that trigger a pivot. If the plan reads like a template, it is one. The agencies worth hiring will push back on your brief. They will tell you what they cannot do. They will ask for access to your CRM before they ask for your ad account. That friction is a feature. ## FAQ ### How long should we give an AI-forward agency before judging results? Ninety days for leading indicators (CTR, CPA trends, creative win rates), six months for lagging ones (CAC, LTV by cohort). Anything faster is noise. ### Is a smaller AI marketing agency better than a large one? Often yes, if they have a data engineer on staff and senior operators on your account. Size matters less than the ratio of judgment to automation. ### What is the single biggest red flag when evaluating agencies? A pitch deck full of AI claims and zero named humans, tools, or metrics. If they cannot show you the machine, they do not have one.

How to Choose a Marketing Agency in the AI Era

Frequently asked questions

Related articles