Is AI a Fad or a Profit Door? Can a Consultant Show Concrete Returns?

That tools bought on excitement gather dust; that “everyone’s doing it” is a sales script, not evidence; that the invoice is certain and the benefit is a story. These lessons are assets. The reflex only errs when it generalizes from sellers to tools — bad sellers don’t make every tool bad.

When it rejects the category instead of the pitch. Somewhere in the AI noise sit mundane, unglamorous uses — drafting, summarizing, sorting, answering the same twenty questions — that behave less like fashion and more like electricity: boring, metered, useful. The reflex should filter pitches, not outlets.

Hours — yours and your team’s — spent on tasks a metered tool would eat, month after month, compounding silently the way the dark room’s invoice does. A wrong “yes” costs a subscription; a wrong “no” can cost a workweek monthly. Both errors have prices; only one appears on a statement.

The hour count: does this use save countable time or reduce countable errors, in my shop, this quarter? Everything in this guide serves that test. Fashion cannot pass it, and real value cannot fail it — which makes it the only referee you need.

Quotes, follow-up emails, product descriptions, social drafts — first versions produced in minutes, your expertise editing rather than authoring. The gain isn’t literary; it’s the difference between facing a blank page and correcting a draft. For a business that writes daily, this family alone often funds the whole exercise.

Count one day honestly: how often does your team answer the same questions — hours, prices, availability, directions, “do you also…”? A well-built answer layer handles the repetitive tier and hands humans the real conversations. The win is double: hours returned, and faster replies — which, as the media guide notes, is itself a sales advantage.

Wherever reading is the bottleneck: long email threads before a meeting, supplier documents, meeting notes into action lists, records into monthly one-pagers. This family serves the very routines this series keeps installing — the light page and the account meeting get cheaper to produce. Order and AI feed each other.

Because errors are hours in disguise: the mistyped order, the missed line in reconciliation, the quote sent with last year’s price. Uses that check, compare and flag reduce the redo-tax. Fewer errors rarely make announcements; they make quieter months — thrifty owners recognize the sound.

A week of honest observation: which tasks eat which hours, done by whom, how often, with what error rate. No tool names yet — tasks first, tools later is the discipline that separates consultants from resellers. The inventory alone usually surprises: the biggest eater is rarely the one the owner guessed.

By multiplication: hours eaten × frequency × ease of piloting. High score, low disruption, quick readability — that’s a pilot spot. The temptation to pilot something impressive is resisted on principle: pilots exist to read, not to impress. Impressive comes later, funded by boring.

The task’s old time and error rate, measured before; the new ones, measured after; the tool and setup cost between them. One page, arithmetic a skeptic can check. If the saved hours don’t clear the costs, the pilot says so plainly — and a pilot that can say “no” is the only kind whose “yes” means anything.

Everything that failed the hour test, everything requiring your data to leave safe boundaries without cause, everything whose pitch needs urgency to survive scrutiny. Postponed is not banned; the list gets revisited as tools mature. The wastebasket and the waiting room are different rooms — good consultants know which door is which.

Because “everyone tries it” means nobody measures it. The pilot lives with one willing person who runs the task the new way, logs the minutes, and reports the friction honestly. Willingness matters: a conscript pilot produces conscript results. Volunteers are cheap to find when the task chosen is one everyone hates.

Nothing machine-drafted reaches a customer without human eyes during the pilot — quotes, answers, documents, all reviewed before sending. The gate protects your name while the error profile gets learned. Gates loosen later, by evidence, task by task; they never start open.

Less than feared, more than zero: a short, task-specific session — how to instruct the tool, what to check, what never to paste into it. Generic “AI literacy” seminars are the fashion version; twenty minutes on the actual task is the profit version. Rules about customer data go in writing on day one.

Because open-ended pilots become unowned subscriptions — the exact failure mode this whole series exists to prevent. On the date: the one-page reading, then one of three honest verdicts — scale it, fix and re-run it, or drop it with a small, known cost. All three are wins over guessing.

With a written rule before the first prompt: what may be pasted into which tool, what never leaves your systems — customer identities, financials, anything sensitive. Tool choices respect the rule, not the reverse. This file and the compliance guide are neighbors; a consultant fluent in one should be conversant in the other.

By admitting the machine’s confident wrongness up front and gating accordingly. The pilot measures the error rate instead of debating it; some tasks pass with light review, some demand the full gate permanently, some fail outright. Priced risk is workable; denied risk is expensive — the sentence applies to consultants too.

Two clauses keep it small: the task’s manual procedure stays documented — the shop must survive the tool’s bad day — and subscriptions stay monthly until the pilot verdict, in your name, cancellable by you. Deeds again: accounts, prompts and outputs belong to the business, not to the provider or the enthusiast who set them up.

A modest, honest one: while you wait for perfect clarity, competitors bank ordinary hours. The answer isn’t panic adoption; it’s the metered trial you’ve now read — small, gated, dated. Official SME digitalization programs tracked by KOSGEB occasionally defray these costs; letting the state share a metered experiment is thrift, not fashion.