AI Without the Hype · Part 3 of 10
The monthly subscription is the easy part to price. The harder question is what it will cost your business to get useful work out of it.
For an employee drafting messages in a chat tool, that may be mostly the subscription, training, and review time. For an AI workflow connected to your inbox, CRM, and scheduling system, add implementation, monitoring, access management, and maintenance.
There is no honest single price for both. To compare options, price a defined job through to an acceptable result.
Start with the work, then ask for a price
“AI for customer service” leaves too much open. “Draft replies to routine delivery-status inquiries for an employee to approve” gives a provider something concrete to estimate.
Specify the approximate volume, where the information comes from, what the tool may do, and which exceptions go to a person. Find out whether a quoted price includes those requirements or only a demonstration using sample data.
Ask about minimum seats, usage charges, required software tiers, overages, support, and contract terms. These vary by product and plan, so get current written terms for the version you are actually considering.
Count the work around the subscription
Your first-year estimate should include initial configuration, preparing source material, testing, training, employee review, and ongoing upkeep. Connected workflows may also need changes when another system or its permissions change.
Some of that work may be handled internally. It still consumes time. Some may be included in a provider’s fee. Ask which parts, how much support is covered, and what becomes an extra charge.
Also check what happens at cancellation. Can you export the information and templates you need? Can your team return to the previous process while you decide what comes next?
A worked example with assumptions you can replace
Imagine an office that produces 200 routine customer updates a month. Each takes eight minutes to prepare and approve. That is 1,600 minutes, or about 26.7 staff hours.
In a trial, AI-assisted preparation plus human review takes four minutes per update. At the same volume, the business uses about 13.3 hours. The difference is roughly 13.3 hours a month.
For illustration, assume the employee’s total employment cost is $30 an hour. That gives the recovered capacity a value of about $400 a month. Assume another $80 a month in software charges and two hours of administration at $30 an hour. Ongoing costs consume $140 of that estimated value, leaving about $260 before recovering setup costs.
If setup and training cost $780, that simplified calculation suggests roughly three months to recover the initial investment. These are invented inputs to show the method. They are not typical prices or a prediction of your results.
Recovered time needs somewhere useful to go
If the employee works the same hours for the same pay, the business has not automatically reduced payroll by $400. It has made time available for other work.
That can be valuable. The office may clear a backlog, answer inquiries sooner, avoid overtime, or handle growth without immediately adding administrative hours. Name the use for that capacity before treating it as a return.
Keep uncertain revenue gains separate. “We might win more jobs” is a hypothesis to test. It should not quietly become guaranteed income in the spreadsheet.
Find the point where the numbers stop working
Repeat the calculation with less favorable assumptions. What if volume is half what you expected? What if review takes six minutes rather than four? What if someone needs an extra hour each week to maintain the workflow?
In the same example, six minutes per update recovers only about 6.7 hours a month, worth about $200 at the assumed rate. Against $140 in ongoing costs, that leaves about $60 before setup recovery. The decision looks quite different.
This is why measuring the approved result matters more than timing how quickly a draft appears.
What I would ask before signing
I would want a written scope, current pricing, an explanation of additional charges, and a short trial with a defined review date. I would also want the employee doing the work to confirm whether the proposed improvement feels useful in practice.
If a provider promises a dramatic return, ask to see the assumptions. The FTC has brought enforcement actions involving allegedly deceptive AI business-growth claims. You do not need to settle for a sales forecast that cannot survive a few basic questions.
Buy when the expected benefit still makes sense after you count the checking, upkeep, and transition work. If the case only works when every assumption is optimistic, keep testing before you commit.
Source
FTC: Crackdown on deceptive AI claims and schemes. All costs, volumes, and returns in the worked example are hypothetical.