Most firms we meet do not lack AI. They lack direction. One partner drafts in ChatGPT. Another team pays for Copilot. A third built a prototype that impressed everyone in a demo and then quietly died. Meanwhile the information the firm actually runs on — projects, costs, references, templates — sits scattered across drives, inboxes, and legacy systems.
The failure mode is consistent. Individual experiments produce individual wins, and none of it compounds. There are no shared rules for data, usage, or security, so cautious people abstain and incautious people improvise. Nothing is measured, so nobody can say what worked. And no one owns the outcome — tools get bought, but adoption, policy, and return have no accountable owner.
The result is familiar: margins tighten while the tool count grows.
Direction is a function, not a purchase. In our transformation engagements it takes three concrete forms.
First, a diagnostic. Interviews with every role — practitioners, operations, leadership, finance — and a full inventory of systems. The deliverable is an operational audit: how AI is really used today, where information lives, and a catalogue of tasks that could become agents. About a month of work, ending in a GO / NO-GO that belongs to the client.
Second, a foundation. Scattered information becomes one reliable, shared source — we call it the Vault — surrounded by the rules that make AI safe to use: data, usage, and security policies, and decision loops of try, measure, decide. The first agent ships here, proven on real work and validated by the firm’s own professionals.
Third, a cadence. Seven to ten candidate workflows run through the same loop, with monthly deliverables and an ROI report covering usage, time saved, adoption, and cost. Direction is not a strategy deck; it is a monthly discipline.
Our diligence work taught us one thing that transfers directly: an output no one signs is an output no one trusts. Every agent we deploy is tested against real work and signed off by a named professional before it is adopted. AI becomes useful when it survives contact with real work — and someone accountable says so.
We have now given this work a name and a page: AI Transformation at LiquidDocs. If your firm’s AI efforts feel like islands without a map, the diagnostic is where we would start.