The longer-form companion to the AI Reality Check. These are the patterns that repeat across the businesses we work in: the ones that decide whether AI compounds or quietly fizzles. Each ends with the move it implies.
The audit told you where you stand. These notes are about what happens next, and the mistakes you no longer need to make first-hand, because other businesses have already made them for you.
The pain shows up in delivery: late reports, slow quotes, a team drowning in admin. The cause is almost always upstream, in how work arrives, how it is scoped, or how information is captured the first time. Point AI at the pain and you get a faster version of a broken flow. Point it at the upstream constraint and the pain downstream often disappears without being touched.
The move: before automating anything, trace the painful task back two steps and ask where the work actually became hard.
Automating a task returns the hours that task used to take, once. Amplifying what your business uniquely knows, your pricing judgment, your delivery patterns, your accumulated precedent, improves every job that flows through it. Most tools sold to you are automation. Most of the durable value we see comes from amplification, because it is built on something competitors cannot buy off the shelf.
The move: list what your firm knows that a competitor with the same tools would not. That list is your amplification map.
If your numbers live in systems that cannot talk to each other, or in spreadsheets that disagree, AI sits on sand. Businesses that buy the tool first pay twice: once for the tool that cannot land, and again for the plumbing they should have fixed first. The plumbing work is unglamorous and almost always smaller than feared.
The move: pick the two systems whose disconnect costs most, usually finance and operations, and get a real quote for connecting them.
AI output quality tracks the quality of context you can feed it. A firm whose pricing logic, delivery runbooks and past decisions are written down can hand a system real fuel, and gets work back that sounds like the firm. A firm that runs on know-how in heads gets generic output, because generic input is all it can give. This is why the cheapest preparation for AI is documentation, and why most stalled projects stalled there.
The move: write "how we price" and "how we deliver" as two working documents a smart outsider could follow.
Every business we meet has tried something: a subscription here, an experiment there, a demo that impressed and then faded. The graveyard is rarely a tooling problem. It is what happens when experiments are started in parallel, owned by nobody, and never taken through to a result anyone keeps. One finished thing beats ten pilots, and the discipline of finishing is itself the capability being built.
The move: kill the zombie experiments deliberately, keep one, and drive it to a result still in use thirty days later.
If checking the output takes as long as doing the work, the gain is zero, and this is the most common silent failure in professional settings. The businesses that win design for cheap verification: structured outputs, citations back to source, reviews at the level of exceptions rather than everything. When verification is cheap, trust grows with evidence, and the system earns more responsibility over time.
The move: for any AI output you rely on, measure how long checking it takes. If checking rivals doing, redesign the output, not the model.
Every task belongs behind one of three gates. Autonomous: the machine does it and nobody checks each instance, because the cost of a rare miss is small. Assisted: the machine drafts, a person approves, because judgment or liability is concentrated there. Human-only: the machine stays out, because the task is the relationship. Most AI failures are gate failures: an assisted task treated as autonomous, or a human-only task automated into somebody's resentment.
The move: take your ten biggest workflows and assign each a gate, on paper, before any rollout.
The best system nobody uses returns nothing. Adoption is not a personality trait of your team; it is managed, like anything else that matters. The pattern that works: one named owner with standing, one visible early win, role-specific training rather than generic demos, and saying plainly what the efficiency gains are for, because the team has already guessed, and silence confirms their worst version.
The move: name the owner before you build, and give them the first win to carry.
Hours saved get all the attention. The quieter, often larger prize is the time between a question arising and a decision being made: quotes that go out same-day instead of Friday, issues caught in the week they emerge instead of at quarter end. Speed of decision compounds through everything downstream of it, and it is usually a founder or one key person who sets the pace.
The move: for your most frequent decision, measure the gap between "asked" and "answered". Then aim AI support at exactly that gap.
Platforms promise everything and require everyone. One workflow taken end to end, quoting, onboarding, reporting, whichever concentrates the most cost, proves value in weeks, teaches you how AI behaves in your business, and builds the appetite that makes the second workflow easier. The platform decision, if it ever comes, is better made after you know what actually works for you.
The move: choose one workflow with a real number attached and take it all the way through before starting anything else.
A demo that impressed the partners is not a result. A result is something the business still uses, with a number attached: hours, dollars, days of latency, error rates. The discipline of measuring is what separates programs that compound from programs that fade, because measurement tells you what to double down on and what to kill, and both decisions create value.
The move: give every AI initiative one number it must move, agreed before the build starts.
In founder-led businesses, the biggest single bottleneck is usually the founder: the approvals that queue, the decisions only they can make, the know-how only they hold. The instinct is to fix the team first. The higher-leverage move is often to write down how the founder actually decides, and build support around those patterns, so the business can move at the founder's quality without waiting on the founder's calendar.
The move: write one page on the decision that most often waits on you. That page is worth more than most software you could buy.
A 30-minute conversation puts these patterns against your actual map: no deck, no pitch, just the sequence that fits your business.
