[{"data":1,"prerenderedAt":139},["ShallowReactive",2],{"\u002Fwork\u002Fcyphon":3},{"id":4,"title":5,"anonymized":6,"body":7,"client":104,"cover":105,"description":95,"draft":6,"engagement":106,"extension":107,"liveUrl":108,"meta":109,"metrics":110,"navigation":126,"order":99,"path":127,"role":128,"seo":129,"stack":130,"stem":135,"summary":136,"year":137,"__hash__":138},"caseStudies\u002Fwork\u002Fcyphon.md","Building an AI agent practice from zero to $100K+ monthly value",false,{"type":8,"value":9,"toc":94},"minimark",[10,15,19,22,25,29,32,35,38,42,45,48,71,74,78,81,84,88,91],[11,12,14],"h2",{"id":13},"context","Context",[16,17,18],"p",{},"In late 2023, Cyphon was a small team with good engineers, a big idea, and no operating spine. The thesis was right — AI agents were about to replace not just tasks but entire roles inside professional services firms, and someone needed to be the implementation partner for enterprises that could not build it themselves. But thesis without delivery is a pitch deck, and pitch decks do not send invoices.",[16,20,21],{},"The engagement started with a specific problem. Cyphon's parent company ran client delivery with 10 project managers, 5 QA reviewers, and 70+ analysts. The margin was getting thinner every quarter. Every new client pulled linearly on staffing. The team sensed AI could unlock the equation, but did not know where to start, what to build first, or how to turn experiments into a billable product.",[16,23,24],{},"What they needed was not a strategy. They needed someone to install operating structure while the team kept shipping.",[11,26,28],{"id":27},"moved-fast","Moved Fast",[16,30,31],{},"Week one, we picked a single workflow that would never make a PR splash but would prove the model: outbound lead welcome emails. PMs were spending 20 minutes per email, drafting in a blank screen, re-using their best lines inconsistently. We built an agent trained on ten years of their top-converting welcome emails, wired it into the CRM, and shipped it to five PMs as a pilot.",[16,33,34],{},"Within two weeks, welcome email turnaround dropped by 90 percent and response rates rose 38 percent. We had our first internal case study. More importantly, we had a pattern: pick a workflow that is costly, repetitive, and already has training data in the client's own history. Ship an agent in two to four weeks. Measure against a hard before\u002Fafter. Expand.",[16,36,37],{},"By week eight, five agents were in production across the parent firm — welcome emails, proposal generation, QA review, lead nurture, and inbox triage. The same team of 85 was servicing the same book of business with two PMs, ten analysts, and zero dedicated QA staff. Margin opened up by a factor we are not at liberty to publish.",[11,39,41],{"id":40},"failed-forward","Failed Forward",[16,43,44],{},"The first external client engagement was a mess. We had proven the pattern internally, and we walked into a Fortune 500 consumer goods company assuming the same playbook would land. It did not. The client's internal data was not clean. The workflows we scoped turned out to be three workflows in a trench coat. And the champion who hired us moved teams eight weeks in.",[16,46,47],{},"We lost two months and a healthy chunk of scope. The failure taught us three things that now shape every Cyphon engagement:",[49,50,51,59,65],"ol",{},[52,53,54,58],"li",{},[55,56,57],"strong",{},"Discovery is delivery."," We now spend the first two weeks of every engagement mapping the actual workflow — not the org chart's version of it — before any agent is scoped. Cheaper to change a plan than an agent.",[52,60,61,64],{},[55,62,63],{},"Champions churn."," Every engagement now onboards at least two stakeholders from day one, with explicit escalation paths. Single-champion engagements are refused.",[52,66,67,70],{},[55,68,69],{},"Training data is the work."," If the client does not have historical workflow data we can train against, phase one is setting up that collection. No shortcuts. Without data, the agent is a chatbot.",[16,72,73],{},"The second external engagement — a mid-market SaaS company — closed in six weeks and shipped three agents. The third closed in four. The pattern that emerged from the failure is now the firm's playbook.",[11,75,77],{"id":76},"built","Built",[16,79,80],{},"In fourteen months we went from two-person experimental team to a 14-person AI agents consultancy with 40+ enterprise clients across finance, healthcare, SaaS, logistics, and consumer goods. 150+ agents deployed in production. A repeatable four-phase delivery model: Discovery & Consultation → Design & Development → Secure Deployment → Ongoing Support. Model-agnostic architecture running on OpenAI, Claude, Gemini, Grok, and private deployments as the use case demanded. OpenClaw framework integration for skill-based, memory-carrying, tool-using agents that operate inside live workflows rather than beside them.",[16,82,83],{},"Internally: a full leadership team with a CTO, a COO, a Chief Experience Officer, a Chief Creative Officer, and seven directors spanning ML, delivery, research, security, analytics, prompt engineering, and predictive modelling. Hiring pipeline, delivery pipeline, sales pipeline, and quality pipeline — all documented, all running.",[11,85,87],{"id":86},"compounds","Compounds",[16,89,90],{},"The thing that makes this case study compound is that the work did not end when we handed the COO seat off. Every process we installed — the discovery-to-delivery pattern, the dual-stakeholder requirement, the training-data phase, the internal-first validation model — is still running. Cyphon's growth since we stepped back has been higher, not lower. That is the test of an operator engagement: can the system run without you?",[16,92,93],{},"The second compound: the internal use of Cyphon's own agents inside the parent firm now runs a client base that would have needed 70+ analysts with 10 analysts. The savings throw off cash that funds R&D on the next generation of agents. Every dollar saved internally pays for the agents we sell externally. That is a flywheel that a strategy deck would never have built — it had to be installed.",{"title":95,"searchDepth":96,"depth":96,"links":97},"",3,[98,100,101,102,103],{"id":13,"depth":99,"text":14},2,{"id":27,"depth":99,"text":28},{"id":40,"depth":99,"text":41},{"id":76,"depth":99,"text":77},{"id":86,"depth":99,"text":87},"Cyphon AI","\u002Fbrands\u002Fcyphon.png","transformation","md",null,{},[111,114,117,120,123],{"label":112,"value":113},"Monthly value unlocked for clients","$100K+",{"label":115,"value":116},"AI agents deployed","150+",{"label":118,"value":119},"Enterprise clients served","40+",{"label":121,"value":122},"Workload reduction across firm","70%",{"label":124,"value":125},"Delivery speed improvement","4x",true,"\u002Fwork\u002Fcyphon","Chief Operating Officer",{"title":5,"description":95},[131,132,133,134],"OpenClaw agent framework","Multi-model orchestration (OpenAI, Claude, Gemini, Grok)","Custom prompt engineering pipeline","Client data ingestion and integration layer","work\u002Fcyphon","COO and build-partner engagement that architected an AI agents consultancy from a team of two to a 14-person practice serving 40+ enterprise clients.",2024,"JzSHL5UVzHgzEn8imQD_7zp9P22y1Q1BZf2ng9sq02w",1776610212872]