How an Agency Automated 200 Videos/Month

Key Takeaways

• A 5-person digital marketing agency scaled monthly client video output from 30 to 200 pieces (6.6x) in 90 days without hiring.
• Per-video production time fell from 2-3 hours to roughly 20 minutes end-to-end, a 78-83% reduction.
• Client roster grew from 12 to 28 retained accounts; average monthly retainer per client rose from $1,800 to $2,650.
• The agency rebuilt its creative stack around Sora 2, Veo 3 and Flux2 avatars orchestrated through AIReelVideo, with Gemini 2.5 Flash for script generation.
• Gross margin on video services climbed from 41% to 68% because the marginal cost of an additional video is now almost entirely token spend, not labour.
• Client approval rate on first draft improved from 62% to 89%, mostly because brand-consistent avatars eliminated the subjective "does this look like us" debate.

The Starting Point

Meridian Social (name changed at the agency's request) is a 5-person digital marketing shop based in a mid-sized European city. At the end of 2025 the team was structured as you would expect for a studio at that scale: one founder doubling as strategist and new-business lead, two account managers running client relationships, one in-house videographer-editor, and one junior creative producing graphics and cutdowns.

The book of business was 12 retained clients: three restaurants, two boutique fitness studios, one regional e-commerce brand selling home goods, one B2B SaaS startup, three real estate agents (two residential, one commercial) and two professional services firms (an accountant and a physiotherapist). Average retainer was around $1,800 per month and every contract included "short-form video content" as a line item, typically priced as 8-12 TikTok or Reels videos per month per client.

That commitment sold well, and it is what was quietly killing the agency.

The Problem

Production was the bottleneck, and it was structural rather than tactical. Each short-form video required, on average:

1. Topic research and trend checking -- 15 to 30 minutes per video, often more when the account manager was researching a niche outside their comfort zone.
2. Script writing -- 20 to 30 minutes per video, usually drafted by an account manager and reviewed by the founder.
3. Shooting or sourcing footage -- 30 to 90 minutes, depending on whether the videographer was on site, the client was sending phone footage, or the team was stitching stock b-roll.
4. Editing, captioning, colour, music -- 45 to 90 minutes per video in Premiere Pro or CapCut.
5. Client review and revisions -- 20 to 40 minutes of async back-and-forth, averaging 1.4 revision rounds per video based on the agency's project management data.
6. Publishing and scheduling -- 10 to 15 minutes split across platforms, including thumbnail selection and copy.

Total: 2 to 3 hours of blended team time per video. At 12 clients requesting roughly 10 videos each per month, the monthly load worked out to roughly 300 video-hours, which is already more than a single editor can carry even at full utilization. In practice the team was hitting about 30 delivered videos per month after accounting for revisions, sickness, new-business work, strategy calls and internal overhead. The rest was either rolled into the next month, delivered late, or silently scoped down to "we'll focus on the three best ideas this month."

There was a second, sharper constraint. The agency's videographer had a hard production rate limit: one person with a camera and a 16-inch MacBook cannot physically cut more than 8-10 polished short-form videos per day, and that is only on days with no client meetings. Hiring a second editor was financially viable but the founder had tried it twice in the previous two years and churned both hires inside six months. Short-form video editing at agency scale is a role with high skill requirements, tight deadlines and low perceived status, and retention was brutal.

The founder described the economic picture bluntly in a call that later became the basis of this case study: "We were selling a product we couldn't deliver, running at a 41% margin that felt like 10% after revisions, and losing two clients per quarter because production slipped."

Why Agencies Are Adopting AI Video

Before covering what Meridian specifically did, it is worth framing why agency automation has become a defensible operational strategy rather than a novelty. HubSpot's State of Marketing benchmark reported that more than half of surveyed agencies had integrated generative AI into at least one production workflow, with short-form video being the fastest-growing category. TechCrunch's AI coverage and other outlets have documented similar trends across creative services, driven by the commoditization of long-form text generation and the more recent availability of controllable video models like Sora 2 and Google's Veo 3.

The commercial logic for an agency is almost identical to the one that drove copywriting studios to adopt generative text tools two years earlier. Clients do not pay for hours; they pay for outcomes and volume. If the marginal cost of producing one more polished video drops by an order of magnitude, then the agency that rebuilds its workflow first captures margin that the rest of the market will only be able to match a year later. For Meridian, that was the opening.

The Evaluation Process

The founder did not choose tools by watching Twitter demos. Before rebuilding anything, the team ran a structured four-week evaluation against five criteria:

• Output quality, graded blind by three existing clients on a 1-5 Likert scale against samples of the agency's current human-produced work.
• Per-video cost, calculated as API or subscription cost plus a fully-loaded estimate of the 20 minutes of human time required to run the pipeline.
• Client approval rate on first draft, measured as percentage of videos accepted without revision.
• API reliability and throughput, measured by running 50 generations per tool and tracking failures, queue times, and degraded quality outputs.
• Brand consistency across a single client, which turned out to be the decisive criterion. A tool can produce one spectacular video and ten misaligned ones and still average above a competitor that produces eleven on-brand 3.5/5 videos.

The evaluation cycled through six stacks: a single-tool approach with Runway alone, another with Pika, a DIY ComfyUI pipeline using open models locally, a monolithic platform the founder declines to name publicly, OpenAI's Sora directly through ChatGPT, and finally AIReelVideo as an orchestration layer wrapping multiple providers.

The DIY ComfyUI pipeline produced the best cost-per-video in raw GPU terms but failed on reliability and brand consistency. The single-tool approaches were either strong at avatars and weak at product b-roll, or the inverse. The orchestration approach won on brand consistency because it let the team lock a single AI avatar per client and route different scene types to different underlying models without the client ever being exposed to that complexity.

The Stack They Built

The production stack Meridian settled on has four layers. Every layer is exposed through a single operator interface so the account managers, who are not technical, can run it end-to-end.

Script generation. Gemini 2.5 Flash generates a three-scene script structured as voiceover text plus visual directions, running through the agency's house style guide and the client-specific category rules. Each script is 140-180 characters of voiceover, producing 18-22 seconds of final video. Cost per script is a fraction of a cent. The team uses the AI script generator as the entry point because it enforces the voiceover character limit, which matters downstream.

Avatar image generation. For clients whose content strategy includes a spokesperson, the agency generates 3-5 brand-locked AI avatars per client using Flux2 running locally on the videographer's GPU workstation. Avatars are approved by the client once, then reused across hundreds of videos. This was the single biggest unlock for consistency: the client sees the same face, outfit, lighting and background every time, which is what makes a feed feel like a brand rather than a stock library.

Video generation. Video output is routed based on shot type. Spokesperson shots and lip-synced avatar videos run through Sora 2 I2V at 20-second duration. B-roll and product-forward shots run through Veo 3 when photorealism matters, or through cheaper variants when it does not. OpenAI's own Sora announcement and Google's Veo documentation are the primary sources the team references when a new model release warrants a pipeline update.

Publishing and scheduling. Once a video is approved the agency uses the video publishing scheduler to queue posts across TikTok, Instagram Reels, YouTube Shorts and, for selected B2B clients, LinkedIn. Captions are generated automatically from the voiceover text and styled per client.

The entire stack is billed in a single token currency, which solved the finance side of the problem. The founder knows exactly what a video costs to produce because every API call settles against the same meter.

Workflow Redesign

The old workflow had six sequential steps, each performed by a human, with handoffs between them. The redesigned workflow has the same six steps on paper but only two of them now require human judgment: strategy and approval. The rest is orchestrated.

1. Discovery. The account manager selects a client and a weekly content theme. The trend discovery tool surfaces competitor videos and trending topics in the client's niche. This step replaces the 15-30 minutes of manual research.

2. Script generation. Gemini produces a batch of 8-12 script drafts per session. The account manager reads them, usually approves 6-9 as-is, rewrites a headline or CTA on 1-2, and kills the rest. Time per batch: roughly 12 minutes, down from 3-4 hours of sequential writing.

3. Avatar selection. Client-specific avatars are already approved and stored, so this step is a dropdown choice. For b-roll-only videos this step is skipped.

4. Video generation. Scripts are submitted to the video generator with the selected avatar and routed to Sora 2 or Veo 3 based on scene type. The batch runs in the background. A single operator submits 10-15 videos in under 10 minutes and moves on to the next client.

5. Review and captioning. Once generations finish, the account manager reviews each video at 1.5x speed. Acceptance rate is 89% on first draft. Rejected videos are regenerated with a revised prompt or routed to a different model.

6. Publishing. Approved videos are bulk-scheduled across platforms with per-client caption templates.

The critical architectural decision was that the videographer no longer sits inside the production path for most client work. He now focuses on two categories of output: physical-location shoots for restaurants and real estate, and premium "hero" videos for client campaign launches. Those still command premium billing. Everything else runs through the AI pipeline.

The Numbers

The agency tracked metrics for 90 days before the migration and 90 days after. The comparison is as follows.

Production volume. 30 delivered videos per month before, 200 delivered videos per month after. Output per team member rose from 6 to 40 videos per month. No new hires.

Per-video time. 2.0 to 3.0 hours before, blended across research, scripting, shooting, editing, revisions and publishing. 18 to 24 minutes after, of which roughly 6 minutes is script review, 3 minutes is video review, and the remainder is client-facing communication. Pure machine time (generation queue) is not counted because it runs asynchronously.

Per-video cost. Roughly $78 fully loaded before, dominated by the editor's time. Roughly $9 after, of which approximately $4 is API and token spend and approximately $5 is the account manager's 20 minutes at fully-loaded cost. The $69 per-video savings compounds quickly at 200 videos per month.

Team utilization. The videographer's week used to be 90% production and 10% client-facing. It is now 40% production (shoots and hero videos), 30% client-facing strategy and 30% internal R&D on the pipeline itself. The founder describes his role as unchanged on paper but notes that strategy time per client roughly doubled because the hours are no longer being consumed by revision triage.

Roster growth. The agency grew from 12 to 28 retained clients over the same 90-day window, entirely through referrals. No paid acquisition was run. Average retainer rose from $1,800 to $2,650 per month because the agency raised prices on new contracts while keeping existing clients at legacy pricing for the first renewal cycle.

Margin. Gross margin on video services rose from 41% to 68%. Net margin at the agency level rose from roughly 14% to 31%.

Client Results

The client-side picture matters as much as the internal economics because a case study that improves margin while degrading client outcomes is a case study of a business about to churn.

Approval rate on first draft. 62% before, 89% after. The founder credits this almost entirely to the locked avatars and client-specific style guides. Clients were previously rejecting drafts because the spokesperson on-screen felt inconsistent between videos; with a fixed avatar that failure mode disappeared.

Turnaround time. Average time from brief to published video fell from 4.1 business days to 0.9 business days. Several clients specifically cited turnaround as the reason they expanded scope, which in turn drove the retainer increases.

Client engagement metrics. Across the twelve legacy clients, average views per video on TikTok rose 34% in the first 60 days after migration. The agency attributes this to volume and consistency rather than any single video being better: posting 40 videos per month versus 10 gives the platform more surface area for trend discovery.

Client retention. Zero churn in the 90-day window after migration. The agency lost two clients in the 90 days before migration, both citing "you keep missing deadlines."

Challenges and Learnings

The migration was not clean. Three categories of problem surfaced in the first 30 days.

Model selection was harder than expected. The team spent the first two weeks defaulting every video to Sora 2 because it produced the best avatar work. Product-focused e-commerce clients came back unhappy because Sora 2 was rendering their products with subtle but visible inaccuracies: wrong logos, slightly off colours, incorrect packaging. Routing product shots to Veo 3 with tighter prompts, and in some cases falling back to real footage from the client, resolved this but added a routing layer the team had not originally planned.

Revision loops got shorter but more opinionated. When revisions dropped from 1.4 rounds to 0.3 rounds on average, the remaining revisions concentrated on a small set of clients who had strong aesthetic preferences. Those clients were producing disproportionate support load. The agency solved this by offering those clients a "premium creative" tier with guaranteed videographer involvement at a 40% price premium.

Client expectations moved. Within 60 days, two clients had begun asking for video volumes the agency had not historically delivered: one requested 60 videos per month, another requested daily posting. The infrastructure could handle it, but the account manager capacity to strategize at that volume could not. The agency introduced a maximum of 40 videos per month per client with a capacity upgrade available on a custom basis.

The broader learning, which the founder now repeats to other agency owners who ask for advice, is that AI video automation does not remove the creative bottleneck; it moves it. The bottleneck is no longer production hours, it is strategic judgment and client communication. An agency that automates production without investing in strategy and communication capacity will hit a new ceiling almost immediately.

Playbook for Other Agencies

Agencies in a similar position can follow the same template. The steps below are the practical checklist Meridian's founder now shares when peers ask how to replicate the result.

Tools. Pick an orchestration layer that can route between models rather than betting on a single video model. Meridian uses AIReelVideo. Direct competitors exist; the important criterion is multi-model routing with a single billing surface, not brand. Pair it with Gemini 2.5 Flash or a comparable fast, cheap LLM for scripts, and a local or hosted image model for avatars.

Pricing. Do not drop prices to reflect your lower costs, at least not on the first renewal cycle. The value to the client is volume and consistency, not your internal unit economics. Raise prices on new contracts and keep legacy pricing for existing clients to avoid churn during migration.

Margins. Target a gross margin in the 60-70% range on video services once the pipeline is stable. Below 60% the pipeline is not yet optimized; above 70% you are probably under-serving clients on strategy and will see churn within two quarters.

Roles. Keep your videographer but shift the role toward hero videos, physical shoots, and pipeline R&D. Do not expect account managers to become prompt engineers overnight; invest in a clear operator interface and script templates so the learning curve is measured in days, not months.

Governance. Lock avatars per client and require explicit approval to change them. Lock brand voice per client and require explicit approval to change it. These two locks are what make AI output feel like a brand rather than a stock library.

Communication. Tell your clients what you are doing. Meridian made the choice to disclose the AI-generated nature of its production pipeline to every client. Exactly zero clients objected; two asked for a small discount, both were declined, both renewed anyway. Transparency on AI content disclosure builds trust that will matter when platforms inevitably tighten labelling rules.

FAQ

How many videos per month can an agency realistically produce with AI?

Based on the Meridian case study: a 5-person agency scaled from 30 to 200 delivered videos per month in 90 days without hiring. Per-video time dropped from 2-3 hours to 18-24 minutes. The ceiling becomes strategic judgment and client communication, not production.

Which AI models should agencies route between for client video work?

Sora 2 for avatar and spokesperson shots, Veo 3 for photorealistic product b-roll, Flux2 for locked client avatars. Single-model approaches fail because each model has a weakness. Use an orchestration layer that routes different shot types to different underlying models rather than betting on a single model.

Should agencies lower prices after adopting AI video production?

No, at least not on the first renewal cycle. The value to the client is volume and consistency, not your internal unit economics. Raise prices on new contracts and keep legacy pricing for existing clients to avoid churn during migration. Meridian lifted retainers from $1,800 to $2,650.

What gross margin can an agency expect after AI migration?

Target 60-70% gross margin on video services once the pipeline is stable. Below 60% the pipeline is not yet optimized; above 70% you are likely under-serving clients on strategy and will see churn within two quarters. Meridian went from 41% to 68%.

Do clients object to AI-produced videos?

Rarely, if you disclose the pipeline openly. Meridian disclosed to every client — zero objections, two asked for a small discount (both declined, both renewed anyway). Transparency on AI content disclosure builds trust that matters when platforms tighten labeling rules.

Conclusion

Meridian Social's story is not exceptional in its details. It is a 5-person agency that hit the production ceiling every services business hits eventually, and it rebuilt its creative stack around a set of AI tools that happened to reach production quality at the right moment. What makes the case study worth documenting is how quickly and cleanly the economics shifted once the migration was committed to: a 6.6x volume increase, a 27-point gross margin improvement, and a 133% increase in client roster in 90 days without hiring.

Agencies watching this space often ask when the right time to migrate is. The honest answer is that there is no waiting move that pays off. The tools are production-ready now. The agencies that migrate first capture the margin window before client pricing normalizes to reflect lower production costs. The agencies that wait will migrate later under price pressure, with fewer of the operational wins that come from being early.

Ready to scale your own video production? Try AIReelVideo today or review our pricing to model the economics for your own agency.