AI Capabilities for Advertising Agencies: The Complete Guide to Choosing the Right Tools and Services

Every agency has already had this conversation: a client asks whether you “do AI”. The honest answer is almost always “partly” — someone in the office generates beautiful frames, and there is no process that carries a beautiful frame through to an approved, broadcast-ready deliverable. The gap between those two things is the whole story.

This guide is written for the agency side, not for the end client. It answers three practical questions: which AI capabilities for advertising agencies are worth holding in-house, which production services for agencies are better bought outside, and how to bring AI video tools into the pitch process without committing to something you cannot deliver. If you are looking for the guide written from the point of view of the business owner choosing an agency, that one lives elsewhere — here you are the side being chosen.

What are you actually buying when you buy “AI capability”?

The most common mistake agencies make is treating AI as a software purchase. The production lead gets a budget, opens subscriptions to four tools, and six months later nobody can tell a client how long a video takes or who is accountable if it comes back weak. Real AI capability is made of three layers, and only the first one can be bought with a credit card.

Layer 1: tools

Image-to-video models, motion models, avatar platforms and an editing layer. This is the cheap layer, the fastest to swap and the least differentiating — any agency can open the same subscriptions tomorrow morning. The full breakdown of which model does which job is in our guide to the best AI video tools, and there is no point duplicating it here.

Layer 2: process

Who writes the prompt, who signs off a mood board, how many revision rounds the client gets, how you lock a character so it stays identical across eight shots, what happens when the model returns a hand with six fingers an hour before a presentation. This is the layer that separates an experiment from a service. It cannot be purchased — it is either built or rented.

Layer 3: accountability for the deliverable

Who signs off that the file meets broadcast spec, that audio is normalised, that the subtitles do not break, and that the client receives a master rather than “a file that came out nicely”. An agency selling AI without this layer is really selling risk, and finds out the first time a campaign falls over.

What to keep in-house and what to send out

The rule that works: keep in-house everything that differentiates you in front of the client, and send out everything that is measured against a standard. Creative, brand knowledge and strategic judgement are the reasons the client pays you, and they must never be handed to a supplier. Visual consistency, sound design, grading and delivery standards are measured objectively, which makes them natural candidates for outsourcing.

A numbers-level comparison of the routes — cost, speed and quality of AI production versus traditional production — is in our dedicated analysis. The commercial model for working under the agency’s name is set out on the agency partnerships page.

How to bring AI into the pitch process

This is where AI gives agencies their biggest advantage, and it is precisely where it is least used. A traditional pitch is presented on a static storyboard; a pitch with AI is presented in motion. What changes in the room is the kind of conversation: the client stops imagining and starts reacting to something they can see.

1. DAY 0Lock the visual language before you touch motionThree or four still frames that define lens, lighting, colour palette and world. If the client does not approve the frame, there is no point producing a shot.
2. DAY 1An animatic in motion, not a film10-15 seconds that show the pace and the one central transition. The goal is to convey a feeling, not to present a deliverable. Label it clearly as pitch material.
3. DAY 1Two extreme versions, not three mediocre onesOne restrained concept and one bold one. AI makes the difference cheap enough to show both, and that is what turns the discussion from “yes/no” into “which one”.
4. PRESENTATIONSay explicitly that the material is AI-producedIt does not weaken the pitch — it prevents the uncomfortable conversation later, at the production stage. Clients react badly to surprises, not to technology.
5. PRESENTATIONPresent the production route alongside the creativeTimeline, approval points, who actually produces. An agency that shows a process is read as safer even when the idea resembles a competitor’s.
6. AFTER THE WINDo not promise the animatic as the final deliverableThe distance between pitch material and an approved master is consistency work, sound and delivery. Price it up front instead of absorbing it.

What breaks when you try to do everything with DIY tools

Every AI video tool has an impressive demo. It is impressive because it is one shot. An agency does not sell one shot — it sells 40 seconds with a recurring character, a product that looks identical in every frame, and a voiceover that stays in sync. That is where the failures start.

The important point: not one of those five is solved by a better tool. All of them are solved by process — which is why an agency that invests only in buying tools pays the same price again on every campaign. A walkthrough of what a structured production route looks like from brief to master is in our process breakdown.

The fit test: five axes

Before you decide to build an in-house capability, score yourself on the five axes below. Four out of five fully green — build in-house. Two or fewer — buy outside and put the difference into creative.

Disclosure, labelling and platform policy

This is the section agencies skip and pay for afterwards. Platform policy on content produced or edited with AI is no longer theoretical, and responsibility falls on whoever uploaded the ad — that is you, not the supplier.

• YouTube requires disclosure of realistic content that is synthetically created or altered: a real person appearing to say something they did not say, an alteration of a real event or place, or a realistic scene that did not occur. Clearly fantastical content, filters and colour correction do not require disclosure (YouTube policy).
• Meta labels content it detects as AI-generated using industry-standard metadata, and expects advertisers to disclose realistic content themselves (Meta announcement).
• Google Ads prohibits, under its misrepresentation policy, manipulation of media intended to deceive, as well as presenting an unlikely outcome as an expected one (Misrepresentation policy).
• Content Credentials — the C2PA standard for signing provenance into files — is gradually becoming something enterprise clients ask for in the brief (contentcredentials.org).

What is different about an Israel-based market

Three practical differences that change the build-versus-buy calculation for an agency operating in Israel.

Hebrew is not just translation. RTL subtitles break easily inside motion layers, Hebrew voiceover requires choosing a voice that pronounces foreign brand names correctly, and the rhythm of a Hebrew sentence is shorter than an English one — which changes the timing of every shot. A supplier who has not worked in Hebrew will hand you back a file that looks good and sounds foreign.

Short, urgent approval cycles. A typical Israeli campaign closes fast, and “we need it tomorrow” is the default rather than the exception. A young in-house capability falls apart exactly there; a partner with a committed timeline does not.

Local invoicing and payment. Israeli clients can be billed in shekels with a tax invoice, which removes friction that exists with a foreign supplier. It sounds like a technicality, but in practice it is what decides whether procurement at a large organisation approves the engagement this quarter or next. The Israel Association of Advertising Agencies (iaaa.org.il) brings together around 30 agencies representing most of the country’s advertising activity — and most of them are facing exactly this question in exactly the same quarter.

What people are saying publicly

Three points from the public conversation that are directly relevant to agencies — all of them verifiable.

Adoption is no longer the question, but it is partial

According to IAB’s 2025 Digital Video Ad Spend report, as covered by eMarketer, 22% of video ad creative was built or enhanced using generative AI in 2024, and the share is expected to reach 39% in 2026. The leading uses are tailoring ads to specific audiences (42%), changing visual style (38%) and contextual adaptation (36%). Note what is not on that list: creating the campaign from scratch. Even advertisers who adopt AI mostly use it for processing and adaptation, not to replace the creative.

The resistance is real, and so is its brand impact

Coca-Cola’s AI-produced holiday ad in November 2024 — the work of three AI studios across four different models — drew sharp public criticism and was described by creators as “soulless”. The company responded that it was a combination of human storytellers with generative AI. The lesson for an agency is not “don’t use AI” but that emotional and nostalgic categories carry the highest risk, and that is where human direction has to be visible.

In the developer community: tools churn fast, craft persists

In the large Hacker News discussion around Veo 3 and Google’s Flow tools — 832 points and 532 comments — three claims recurred that match exactly what an agency experiences in practice: the single shot is impressive, but consistency between shots breaks and environment details change across cuts; the few-seconds-per-shot length limit dictates the cinematic language rather than the other way round; and the use case commenters found most convincing was advertising creative at volume and storyboarding, not long-form work. In a separate thread following the shutdown of Sora, a heavy user described a monthly bill of 130 dollars on Google Veo and noted that the AI video conversation is dominated by other models anyway. The practical conclusion for an agency: any bet on a single tool is a short-term bet, and the two breaking points described above — consistency and length — are precisely what the technical community identifies. What survives is the process and the creative judgement above it.

Cost and delivery timelines

Two numbers worth holding in your head when you build a quote for a client.

Standard delivery time for outsourced production at ArcaneWiz is 7–10 working days from brief approval to the final master. Israeli clients can be billed in shekels with a tax invoice. Usage rights are defined per project in the quote. The entry figure and the scope breakdown are on the pricing page and in the questions below, and you can build an initial scope through the project planner.

For an internal comparison, price the in-house route at its full cost rather than at the cost of the tools: subscriptions, team hours, management time, repeated revision rounds and schedule risk. Calculated that way, the break-even point usually lands at around eight videos a month — below that, an external partner is cheaper even before you talk about quality.

Checklist before you commit to a client

• You have defined in writing which parts of the deliverable are AI-produced, and the client has approved it.
• You have two extreme versions to present, not three mediocre ones.
• You locked the visual language in still frames before producing any motion.
• You agreed a ceiling on revision rounds and what counts as a round.
• Someone owns sound, normalisation and Hebrew RTL subtitles.
• The delivery spec — formats, aspect ratios, lengths — is closed with the client up front.
• You checked the labelling requirements of the platform the campaign will run on.
• The deliverable contains no replication of a real person’s likeness or voice.
• The timeline includes the client’s approval round, not just production time.

Where ArcaneWiz fits — and where it does not

We are a production studio that works with agencies as an execution arm. We use generative AI tools as part of the process, and we say so openly — alongside direction, editing, grading and sound design done by people. The full model is described on the AI video for agencies page, and examples of work are in the portfolio.

We are a fit when you need a cinematic bar with a committed timeline, when version volume is weighing on your team, or when you want to offer a capability you are not yet ready to build in-house. We are less of a fit when the need is dozens of short social videos a month at minimal cost — there, DIY tools with a fixed template will do the job, and we will tell you so up front.

What we will not do: we will not write your creative and we will not take over the client relationship. Specific formats such as an AI commercial and AI video ads sit on dedicated pages. If you want to understand how an agency’s own content appears inside AI engines, we have a GEO guide for creative agencies, and a tools buying guide at CMO level.