The height in a picture does not settle the servings
Someone asks for a three-tier cake, mentions forty guests and later writes: “Make all of them chocolate.” Yet the initial enquiry described two dummy tiers. Before quoting, there is a more useful question than choosing a filling: which parts do they want to eat, and which do they want to display?
This guide proposes an AI trial to prepare that clarification: separate each visible tier, its requested composition and any separate cake for serving. You will find a fictional file, a manual worksheet and a trial prompt. We do not calculate servings or report results from an executed trial.
Map two things: what is displayed and what is served
Ms. Robinson describes the option of adding dummy tiers to a tall cake in its own offering. Dolça Fantasia sells cake dummies for display or practice. These pages establish that those options exist; they do not establish what your customer’s order contains. A cake’s appearance or a reference image does not replace that confirmation.
For this exercise, I propose two sections. Under “display”, record each tier by position — top, middle or bottom — and copy whether edible, decorative or still unspecified is requested. Under “serving”, retain the requested portions and separate items. A kitchen sheet cake, if requested, gets its own line: it is not automatically another tier in the visible cake.
Ma Petite Patisserie explains that it tailors wedding cake quotes to guest numbers and location. That is the approach published by that business, not a general tariff. In our worksheet, guests, requested servings and servings the bakery agrees to prepare are separate fields. Only the professional can define the proposal and confirm quantities, design, feasibility and price.
The invented enquiry: three tiers and a change with more than one reading
All text, references and rules in this case are fictional. Use only these data for the trial; do not submit real conversations, names, customer photographs or celebration details without permission and suitable confidentiality and processing conditions.
F0 · Trial rule: “Prepare a worksheet and questions without confirming an order or producing a quote. Changes require clarification from the enquirer and assessment by the bakery. Do not calculate servings, dimensions, recipes, structures or storage conditions.”
F1 · Initial enquiry: “I want three visible tiers. The top one edible, vanilla; the middle and bottom ones dummy. There will be 40 guests. Please also explore a separate sheet cake for serving from the kitchen: I have not yet decided how many servings to request. Smooth ivory finish.”
F2 · Later message, still unclarified: “And make all of them chocolate.” It does not specify filling, coating or colour, or confirm a change to the dummy tiers. There is no accepted order, price or agreed serving quantity.
The manual worksheet: what you can copy and what needs asking
This reference was written by the editor to check a future AI response. It is neither assistant output, a production order nor a technical proposal from the bakery. Numbers belong to the invented case.
| Item | What is documented | Clarification pending |
|---|---|---|
| Visible top tier | F1 requests edible vanilla cake. | F2 mentions chocolate without specifying the change. Keep both texts; do not replace the flavour. |
| Visible middle tier | F1 requests a dummy. | Clarify F2; do not make it edible or assign servings to it. |
| Visible bottom tier | F1 requests a dummy. | The same unresolved F2 question. Do not infer a filling from the word chocolate. |
| Separate kitchen sheet cake | F1 asks to explore this option; it does not confirm it. | Servings undecided. Do not list it as a fourth visible tier or an already ordered item. |
| Serving and finish | F1 states 40 guests and an ivory finish. | No agreed servings are documented. Clarify whether F2 changes the finish; invent no quantities, price or acceptance. |
Ask for a clarification you can review, not a cake nobody ordered
First prepare that worksheet manually. Then copy F0–F2 in full into an assistant with this trial prompt:
“Organise this cake enquiry for human review. Separate the three visible tiers from the separate kitchen sheet cake. For each item, retain requested composition, source and exact quotation; write pending where undocumented. Keep guests and servings in different fields. Present F2 as unresolved ambiguity: do not choose between filling, coating or colour, or override F1 yourself. Return missing information and a short draft question for the enquirer. Do not calculate servings, dimensions or prices; confirm no production, availability or order. Send no messages and change no records.”
As an editor-written draft, the question could read: “When you say chocolate in all the tiers, do you mean the filling of the edible part, the coating or the finish colour? Does the request for an edible top and two dummy tiers still stand? The separate sheet cake and servings remain undecided.” This text has not been sent to anyone.
Claude’s documentation proposes admitting uncertainty and checking claims through citations. That guidance inspires the trial without guaranteeing that an assistant will follow it. Compare its response with the manual worksheet: reject a change from dummy to edible without clarification, 40 servings inferred from 40 guests, or a sheet cake presented as confirmed. Check quotations against the original text, not just reference labels.
Resolving one clarification does not close the entire order
In a separate second trial, add F3 · Fictional clarification: “Chocolate refers only to the finish colour on all three tiers. I still want an edible vanilla top and two dummy tiers. The separate sheet cake is still to be assessed, and servings remain undefined.”
The reference now retains vanilla at the top and two dummy tiers; it records the chocolate colour requested in F3 instead of the initial ivory. That document resolves the meaning of F2. Do not ask the same question again as though F3 were absent. The sheet cake, servings and bakery assessment remain pending: clarifying one word does not confirm production or the quote.
This variant checks something beyond spotting doubts: incorporating a valid clarification without retaining obsolete questions or closing matters that still have no answer. Save both versions with their sources.
Measure the pilot with cases that challenge the worksheet
Prepare more invented enquiries: all decorative tiers with separate serving cake; a tier whose composition is unspecified; a colour change without a flavour change; an enquiry with specified servings but no bakery acceptance yet. Write each manual reference first.
Count composition changes without a source, confusion between guests and servings, omitted or duplicated separate items, and questions already resolved by a clarification. An order marked confirmed without acceptance should invalidate the trial. Also record time spent preparing data, checking quotations and correcting the worksheet, and compare it with doing the work manually. These are proposed pilot criteria; no savings or accuracy gains have been demonstrated.
The worksheet organises a commercial enquiry. It should not decide ingredients, allergens, food suitability, storage, transport or assembly. Those matters require the relevant professional process and review outside this exercise. Do not infer composition from a generated image either: the cover only illustrates a consultation.
From a manual trial to a bounded integration for your cake business
If your current spreadsheet or application already separates tiers, serving items and acceptance of changes, start by using those fields properly. Copying fictional data into an assistant and reviewing a proposal is a manual trial. It does not connect a calendar, reserve a date or update orders.
An integration to assess might retain source text, each tier’s position, requested composition, separate items and open questions. The AI proposal would remain separate from the staff-approved worksheet, with a clarification history. Actual application exports or connections, permissions, error recovery and authority to accept each change would need checking. We claim no compatibility with the bakeries or tools cited.
Adaris develops custom software, CRM and AI automation case by case. We can assess a first delivery limited to preparing and reviewing enquiry sheets, with agreed examples and acceptance criteria. This is a development proposal, not an existing proprietary cake-shop application. Commercial support is in Spanish.
Related services
Sources
Does your worksheet mix design, servings and unclarified changes?
Tell us how you receive and confirm enquiries. We can assess a worksheet with human review before connecting your business systems. Commercial support is in Spanish.