The GM has done the work. Three vendors evaluated, demos taken, a spreadsheet with tabs. They bring it to the owner, who looks at it for forty seconds and says let's talk about it next month. Next month does not happen.
The problem was not the analysis. It was that the analysis required the owner to do the reasoning themselves, in a meeting, from a spreadsheet built by someone else. A one-page memo shifts that work to you and asks them for one thing: a yes or a no.
What the owner is actually deciding
They are not deciding whether voice AI is good technology. They are deciding whether to add a recurring monthly expense in a business where recurring monthly expenses have a way of surviving long after their usefulness does.
That is the frame the memo has to answer. Every owner who has been in the business ten years has a subscription they are still paying for and cannot cancel because nobody remembers what it does. Your memo is competing with that memory, not with a rival vendor.
Which means the memo needs a number for what stopping costs, an honest statement of what you do not know, and a date by which the decision gets revisited. Those three things do more for credibility than any projection.
Start with what the phone costs you right now
Start with the loss, not the product. The first sentence of the memo should be a dollar figure with the arithmetic visible.
The arithmetic is simple enough to fit in two sentences. You take X calls during your peak hours, you miss Y percent of them, your average phone ticket is $Z, so you are leaving roughly X times Y times Z on the table each month. If you take 900 calls a month during dinner service, miss 18 percent, and average a $34 ticket, that is about $5,500 a month walking to whoever answered on the second try. Method for getting an honest miss rate is in the real cost of a missed restaurant phone call, and it is worth measuring rather than estimating, because owners test the input they distrust most.
Then state the number you are least sure about and why. If your miss rate came from one week of manual counting, say so in the memo. An owner who catches you overstating confidence stops reading. An owner who sees you flag your own weak input starts trusting the strong ones.
Then what changes, in dollars
Two components, and keep them separate because they behave differently.
Recovered revenue is the calls that now get answered, valued at contribution margin rather than gross ticket. This is where most memos overreach. If you recover $5,500 of gross sales at 25 percent contribution margin, the memo says $1,375, not $5,500. Owners know food and labor come out of a ticket, and a memo that pretends otherwise reads as a pitch. Being conservative here costs you nothing, because the number usually still works.
Labor is the second component and it is smaller than vendors suggest. You rarely cut a shift. What you get back is a host who is not on hold at 6:40pm while four people wait to be seated, and a manager who stops taking catering calls during prep. Value it in hours at a real wage, and if you cannot honestly claim a schedule change, say the benefit is service quality and leave it out of the dollar total. The method for doing that without inflating it is in calculating labor hours saved.
If you want a fuller model to check your own arithmetic against before you compress it to one page, the ROI calculator walkthrough covers the inputs.
Cost and payback belong together
State the full annual cost, not the monthly one. Owners think in monthly for operating expenses and annual for decisions, and the memo should give them the annual number so nobody feels a small number was used to slip something past them.
Include setup, twelve months of base, and an estimate of overage during your busiest months. If a vendor quotes $250 a month, the honest annual line is somewhere above $3,000, and saying so protects you when the January invoice arrives higher than the December one. What to look for in the fee schedule is in the costs that don't appear on the quote, and current plan structure is on the pricing page.
One more line belongs here and is usually missing: what it costs to stop. If the term is month-to-month, write "cancel with 30 days' notice, no exit cost." An owner evaluating a reversible decision applies a different standard than one evaluating a permanent one, and most people never tell them which kind they are looking at.
Then payback, in months, using the conservative revenue number above. If it is under four months, write the number and stop. If it is over a year, you should probably not be writing this memo at all. Payback conventions for this category are in how to think about payback period.
Answer the objections before they get raised
This is the section that separates a memo that works from one that gets tabled, and most people leave it out because it feels like arguing against yourself.
Three objections come up almost every time. What happens when it fails on a real customer, and the answer is that calls transfer to a person under rules you set. What happens to guests who want a human, and the answer is that they ask for one and get one. And what if we hate it in six weeks, which is the real question underneath the other two, and the answer needs to be a specific term length and a specific exit.
Write each objection as the owner would say it, in their words, then answer it in two sentences. An owner reading their own objection in your memo concludes you have already thought about it. An owner who has to raise it out loud concludes you have not.
The thirty-day review, named in advance
The last third of the page is the part that gets a hesitant owner to yes, and it costs you nothing to write.
Name two metrics and a date. Answered-call rate during your peak three hours and order accuracy measured by pulling twenty tickets, reviewed on a specific day thirty days after go-live. Write down what number would mean it is working and what number would mean you turn it off. Do that before you start, because a threshold set afterward is a threshold set to justify the decision you already made.
This does two things. It converts an open-ended subscription into a bounded trial, which is the frame owners accept most readily. And it puts you on record, which is what makes the second memo you write easier to get approved than the first.
Keep it to three sentences. A review plan that reads like a project charter defeats the purpose.
What to leave out
No vendor logos. No feature list. No comparison table. No mention of the technology working, because the owner does not care how it works, they care what happens when it does not.
And no ask for a decision in the meeting. End the memo with a recommendation, a proposed start date, and a proposed thirty-day review with two metrics named. Send it two days before you talk about it. The decision an owner makes alone at their kitchen table is the one that sticks; the one you extract in a meeting is the one that gets reversed the following week.