Creative agency, subsidiary of a media group · four-week implementation, then adoption · Client details anonymized. Time savings are the team's own estimates, not measured over a year.
Situation
A creative agency owned by a media group, with a leadership team of five and an IT partner. The agency wins work by pitching: every new client means a proposal, a deck and a price. The pricing took two to three hours per proposal, built from memory and ten years of past jobs sitting in a file-sharing system. A first-draft deck took four to eight hours. The team already used ChatGPT and Claude individually, with no shared knowledge, no governance and a chat history nobody could find again.
The CEO wanted the back office faster. The head of marketing wanted to stay six months to two years ahead of competitors as AI became a commodity. The growth officer wanted visibility across every project. The AI operations lead asked the only question that mattered: will the team actually use it day to day?
Method
Crawl, walk, run. We implemented an LLM-agnostic platform, so the agency is not locked into one model vendor, and connected it to the three systems that held the agency's intelligence: the messaging workspace, email and calendar, and the file store with ten years of proposals.
Four custom agents, built from the team's own workflows:
- Pricing agent. Trained on the agency's past jobs, with budget tiers, seasonality and negotiation context. Answers the two-to-three-hour question in minutes.
- Pitch deck agent. Interviews the account lead, researches the brand and the moment, proposes a structure, writes in the agency's voice and drops out a branded deck. Seventy to eighty percent of a first draft in twenty to thirty minutes.
- Daily briefing agent. Morning recap, prospect research and end-of-day notes for each leader, from their own inbox and calendar.
- Master agent. Lets any team member build their own agent, so the platform grows without us.
Timeline. Weeks one and two: technical scoping with IT, single sign-on, governance, access, and two short business interviews on the pricing and deck workflows. Weeks two and three: platform setup and agent development, with no user access until the agents were ready. Week four: launch with power users and admins, two training sessions, then fifteen business days of testing and iteration.
Results
- Pricing a proposal: from 2 to 3 hours to minutes.
- First-draft deck: from 4 to 8 hours to 20 to 30 minutes.
- Fifteen users live on one platform, with shared knowledge and governance, replacing individual chat accounts.
- The team builds its own agents after the handover.
The savings are the team's estimates from the first weeks, not a twelve-month measurement. We publish them as such.
What this means for a services business
Start from the workflow that wins revenue. Pricing and pitching were where the hours went and where speed changes outcomes. Two short interviews were enough to specify both.
No user access until the agents work. The team saw finished tools on day one, not an empty platform. Adoption followed.
Tool-agnostic by design. The platform runs any model. When a better one ships, the agency switches without rebuilding.