The Most Profitable AI Play Isn't Building Platforms

01 /The Concept

Multiply is an AI agency that helps B2B companies run more effective advertising.

In principle, any company can put in the effort and launch a decent ad campaign. The problem is that campaigns burn out – effectiveness declines over time, and the returns that looked promising at launch eventually flatten.

Multiply's core offer is a "self-updating" advertising platform: campaigns that automatically refresh to keep performance above the client's acceptable threshold.

The model is hybrid – a combination of human experts and AI agents. Human experts set strategy; AI agents execute it.

Client engagements run in two phases.

Phase one is a sprint with Multiply's strategists to build the advertising framework. The team develops and tests hypotheses about what will work, and the output is a campaign blueprint – documenting strategy, audience definitions, channels, formats, and offers that perform best.

Phase two is continuous optimization. AI agents create, test, and refresh ad creatives within that framework. The goal is preventing performance decay, not just monitoring it.

Multiply also runs monthly strategy sessions with each client to review results, set targets for the next month, and generate new hypotheses for improving the existing framework.

Inside the platform, different AI agents handle different jobs:

- Execution agents handle creative work: writing new ad copy, generating visuals, running A/B tests, researching new keywords, tracking performance metrics, and optimizing bid prices.

- Intelligence agents collect the inputs that inform strategic decisions: extracting insights from customer calls and emails, analyzing won deals to build and refine Ideal Customer Profiles (ICP), and so on.

Multiply is still in beta with pilot clients, but has already helped those clients close $25 million in new deals. Some clients have seen 300–500% improvements in campaign efficiency.

Those results were persuasive enough that Multiply just closed its first funding round at $9.5 million.

02 /Why It Matters

The AI agency model is gaining real momentum as a business structure. The premise: a startup builds an AI platform but doesn't sell access to it – it uses the platform to deliver services directly to clients.

The economics are striking. Charging for outcomes can yield 100x more revenue than selling the same capability as a subscription. Y Combinator has added the AI agency model to its list of startup categories it actively wants to fund.

That said, AI agencies can run on different internal architectures.

The simplest version has human experts verify and improve AI-generated outputs. The canonical example is Crosby ([related review](/review/bolee-prostaja-model-dlja-sozdanija-perspektivnogo-ii-produkta)), a legal AI firm that raised $25.8 million last year. It reviews contracts and legal documents – with the AI doing the heavy lifting and human lawyers checking and refining the results before delivery.

Multiply runs on a different architecture: humans own strategy, AI agents own execution. Several other AI agencies use the same split:

GrowthX ([related review](/review/novaja-biznes-model-dlja-bystrogo-i-pribylnogo-rosta)) raised $12 million for an AI agency applying this model to organic traffic acquisition rather than paid advertising.

Imagine AI ([related review](/review/to-zhe-samoe-no-v-100-raz-dorozhe)) helps companies scale social media presence by activating all senior staff as content publishers. A human strategist defines the content framework; AI agents then write draft posts for each participant to publish on the agreed schedule.

Fifr ([related review](/review/ty-doverish-ii-upravljat-svoimi-dengami)) raised $1.5 million in November for an AI agency applying the same split to personal investing: a human expert agrees on investment strategy with the client, then AI agents execute it automatically – buying and selling securities within the agreed parameters.

In all of these cases, strategies can be refined through regular reviews or at the client's request.

03 /Opportunities

The obvious general direction: build an AI agency. There's no clear reason a startup should settle for subscription revenue when the same platform could generate 100x more through delivered services.

Within that general direction, the most compelling variant is the strategy-plus-execution split: human experts own the "what and why"; AI agents own the "how and when."

This architecture seems more defensible than simple AI verification, because the human contribution – strategic judgment – carries higher intrinsic value than proofreading outputs. That means it commands a higher price.

It also ages better. As AI quality improves, the value of verification alone will shrink – except in genuinely high-stakes domains where a single error can be catastrophic. That's the opening Amigo ([related review](/review/u-tvoego-ii-agenta-est-sertifikat)) is pursuing with purpose-built AI agents for healthcare, where reliability is non-negotiable.

So – what other domains lend themselves to this model? Where can you maintain strong service quality while dramatically increasing speed and efficiency by separating human strategy from AI execution?