Law Practice
The COO search was the symptom, not the diagnosis
An owner-led law practice was already looking for a COO when we met them to talk about where AI could help. The firm had grown the way good firms grow, by winning cases until it was twice the size it used to be, and the weight that came with the growth was real. Every operational decision still routed through the owner, attorneys were carrying administration alongside full caseloads, and the COO search had become the conventional answer to a problem the firm had not yet fully diagnosed.
What was at stake
The firm worked harder than its results explained, and nobody could say where the loss sat. A search and a salary were about to be spent on a job description nobody had earned the right to write, and the operator they hired would inherit a firm whose leaks had no names. On contingency, every signed case starts with an inquiry the firm already paid for. Spending a year of executive search while the leaks kept leaking would compound the cost the firm could not see.
What the day found
We asked one question of the COO plan: what would the operator find? The firm could not say. There was no named bottleneck, no baseline, no metric anyone watched going the wrong way. We proposed the part of the COO job that mattered most, done first, for a flat fee: a day inside the practice watching the work as it happens, and a playbook two weeks later naming the leaks and what they were worth. If the day said the firm still needed a COO, the job description would get written from evidence.
The day split the practice in two, and the split was the finding. Downstream of a signed case the firm ran with discipline. Statutes of limitation and court calendars enforce a rhythm no consultant could impose: filings moved, hearings got made, the staff was good. Upstream of a signed case there was no rhythm at all. A receptionist answering between transfers, a website form feeding an inbox the office administrator cleared when she had time, and messages taken on faith because reception is in no position to ask the questions that decide everything downstream.
The senior attorney we shadowed came back from a morning in court to the day's message stack. In the middle of it, a slip from two days earlier: a caller hurt in a crash, exactly the case the firm exists for. The callback found the caller already signed with another firm. The attorney's face showed no surprise. That callback happens every week, and everyone had learned to live with it, because no report anywhere showed what it cost. The firm's own numbers told the rest. About one hundred fifty inquiries a week, roughly a third arriving outside office hours straight to voicemail, web forms waiting a business day for a first response, and roughly a third of evaluation hours spent on cases the firm could never sign because the first screening happened in the attorney's chair. In a market where the insurer's adjuster calls the injured person within days, often first, the firm's front door was its whole business and nobody owned it.
The strategic shift
Instead of treating the COO search as the answer, we tested what the firm was trying to hire around. The firm did not have a general operations problem. It had an unowned front door in a market where the first helpful answer signs the client, and a second leak behind it that the doubling had quietly built: the firm's knowledge had dispersed into senior heads and personal folders instead of compounding. The COO instinct was right about the weight and wrong about the organ. The trade-off was visible: a smaller, more specific first move than a new executive, but one the firm could prove on its own funnel in weeks.
Before anything got built, the funnel got interrogated. We shopped the practice the way a stranger meets it, calls and forms at all hours, and the gap held. Then the test that settled the argument: for two weeks, one person answered web inquiries within the hour, business days only, nothing else changed. Evaluations booked from that channel moved sharply. Speed alone, no AI, no system, moved the number. Two conventional alternatives died in the same weeks. An answering service covered the nights without lifting bookings: speed without screening is half a fix. A screening script at reception collapsed in the busy hours when most inquiries arrive: screening that depends on a person being free is no screening at all. The build's requirement got written by those failures: speed, screening, and constant availability at once, which the firm could not get from its existing staffing model.
How we built it
Full implementation ran six months. The front door went first, on web inquiries before phones, with every decline reviewed by a human before it was final. The screen answers in minutes at any hour, gathers the facts reception could never ask for, screens for merit, venue, the limitation clock, and conflicts, and books qualified cases onto the right attorney's calendar by case type and capacity. Two lines were drawn in its architecture before its first conversation and neither has moved. It never answers the legal question and never says what a case is worth, even asked directly, even asked desperately, because that line is the practice's license and it is enforced in design and logged on every interaction. A caller in crisis, still in treatment or describing an unsafe situation, reaches a human immediately at any hour, ahead of everything else the system does.
The phones joined the screen by stages, after-hours first, then overflow, then first-line, with reception repositioned as the human gate the system escalates to instead of the bottleneck it overwhelmed. The intake owner who now holds the decline queue and the weekly audit came from the people who used to take messages on faith between transfers.
The firm's knowledge got written down alongside the front door, case type by case type, curated by the senior attorneys, versioned like law, attributed, current. Not a wiki. The doctrine serves the practice and the screen both, so the firm's judgment reaches the first phone call without costing an attorney's afternoon. Over the systems the firm already ran, none of which got replaced, one picture of the practice got assembled: caseload, progress, rulings, deadlines, status, readable by the attorneys, the clerks, and operations, with nobody entering data to feed it.
Live was not the finish line; governed was. Trust on the screen was granted per decision type, on evidence. Scheduling and routing graduated to light review on catch rates near zero. Declines graduated last and never fully, holding a permanent sampled review, because a wrongly declined client is the one error no aggregate would ever surface. The advice line never graduated at all. After one doctrine update tightened acceptance criteria, the screen began quietly declining a borderline category the firm in practice signs. Every aggregate held. The weekly decline audit, run at the category level, caught the shift inside a week. The rule was corrected the same day as a versioned change, and category-level decline rates joined the weekly read permanently.
What changed
Signed cases from the same inquiry flow rose without new marketing spend. Attorney hours on casework are up across the practice. Legal research costs are down because the firm's own work is findable and stops being bought twice. The same inquiry flow now produces more screened evaluations on the right calendar.
A new inquiry now gets a first response in minutes, at any hour, on both channels, ahead of the adjuster instead of behind. The roughly third of inquiries that arrived outside office hours stopped meeting voicemail. Evaluation calendars arrive screened and conflict-checked, and the fifteen to twenty attorney hours a week that went to cases the firm could never sign went back to casework. The owner reads the practice in minutes instead of assembling it from asks.
The firm runs cleaner today than it did when it was half the size.
What stands as proof
The Day One Audit recorded the baseline before any build. The cheap tests tested the diagnosis on the firm's own funnel before money was spent on construction. The drift incident showed the governance working: the failure mode the weekly audit exists for showed up, the audit caught it inside a week, and the fix went in the same day as a versioned change. The Audit, the Playbook, and the Charter stand as the engagement record.
The intake owner who runs the gates today came from the firm's own front of house. The doctrine is the firm's, curated by the firm's senior attorneys. The system is live, the firm runs it, and the owner who had been looking for a COO got the part of the operating help the practice actually needed: the part that finds the leaks and closes them, built into the practice itself, with the judgment exactly where the clients pay for it to be.
A hire is a diagnosis you have not made yet. Spend a day before you spend a salary.