Prepared with AI assistance and reviewed for clarity, relevance and unsupported claims.
An automation discovery checklist should help a small business explain its work, not create a paperwork project. The aim is to bring enough evidence to distinguish a repeatable task from a decision that still needs a person.
For a Moroccan merchant or service SME, that evidence may live in message threads, notebooks, spreadsheets and staff habits. You do not need to reorganise everything before the meeting. You do need to show where information comes from, who checks it and what happens when the usual process breaks down.
1. Describe one workflow and its boundaries
Start with a short statement: “We want help turning incoming food orders into structured drafts for staff confirmation.” This is more useful than “We want AI across the business” because it identifies both the task and its approval boundary.
Use a hypothetical food merchant as an example. Customers send mixed orders by message, a staff member copies them onto an order sheet, and the kitchen lead confirms availability before anyone promises preparation. The proposed workflow stops at an internal draft; it does not automatically accept the order.
Bring a simple current-process outline:
- Where the request arrives.
- What someone copies, checks or rewrites.
- Which source they consult.
- Who makes the next decision.
- What marks the end of this task.
If the starting point or endpoint is unclear, note that as a discovery question. The guide to defining workflow triggers and states can help you make those boundaries concrete.
2. Gather examples that reveal different decisions
Do not bring only clean, successful cases. A handful of deliberately varied examples can reveal more than a large folder of routine messages.
For the hypothetical food merchant, useful samples would include a straightforward collection order, an order missing quantities, a request for an unavailable item, a change after staff confirmation and a message that appears to repeat an earlier order.
For every example, attach the outcome and a short explanation of why staff acted that way. If the outcome is unknown, label it unknown rather than reconstructing a convenient ending.
- Input: what the customer actually supplied.
- Interpretation: what staff understood, including uncertainty.
- Decision: what was accepted, queried or declined.
- Output: the order sheet, response or internal task produced.
- Exception: what prevented the normal sequence.
Anonymise examples before sharing. Remove names, phone numbers, precise addresses, payment details and identifying information in attachments. Preserve relationships using neutral labels such as “Customer A” and “Order A,” so repeated messages still make sense. Never include passwords or access tokens.
3. Bring templates and explain which version counts
Existing templates show what the business considers a complete output. Bring the current order sheet, approved confirmation wording, product references and any checklist staff actually use.
Mark the owner and revision date where known. If two menu files disagree, do not quietly choose one for the meeting. That conflict is important: the workflow needs an agreed source before it can prepare reliable drafts.
In our hypothetical example, “two lunch boxes for tomorrow” is not enough to confirm an order. The sheet might require the menu choice, quantity, requested collection time and contact reference. Staff must still check whether the request fits the current preparation plan.
Separate fixed wording from judgement. “Thank you, we are checking your request” may be reusable. “Your order is confirmed” should depend on the merchant’s approval rules. If cutoffs matter, bring the actual internal rules and review how preorder cutoffs relate to preparation batches.
4. Inventory tools without assuming they connect
List the tools involved, what each one stores and who controls access. Include informal tools: a shared spreadsheet or paper preparation sheet may be more important than an unused CRM.
For each tool, record the account owner, business purpose, relevant fields and any known import or export method. Describe restrictions you have encountered, but label unverified capabilities as questions. Do not assume a connector or subscription includes every action the project needs.
A screenshot or anonymised export may be sufficient for discovery. Granting broad production access is usually unnecessary just to explain the process. Any later access request should specify its purpose and scope.
Also explain language and format variation. If customers mix French, Darija, voice messages and product nicknames, include representative anonymised material. A workflow designed only around tidy typed messages would miss an important part of the task.
5. Name decision owners and non-negotiable boundaries
The session needs someone who can decide how the workflow should behave, not only someone who knows the software. Invite the person doing the work and the person authorised to change the process; in a small business, this may be the same person.
Prepare explicit questions: Who can confirm availability? Who resolves conflicting quantities? Who handles late changes? Who covers those decisions during an absence?
Hypothetical discovery note: “Automation may prepare the order draft. The kitchen lead confirms availability. Only the designated sales colleague sends the final confirmation. An unclear quantity remains a question, not an estimate.”
State preferences and hard limits separately. “We prefer one shared sheet” is negotiable. “No customer confirmation before availability review” defines an essential boundary.
6. Leave with a scope, not a vague promise
A useful discovery outcome identifies the first task, required inputs, expected output, review owner, exceptions, open technical questions and acceptance checks. It should also identify what remains manual.
Discuss how to assess usefulness without inventing a baseline. If nobody knows how long order entry currently takes, agree to observe it rather than claim a saving. A draft that is fast to create but slow to correct may not be worthwhile.
Use the evidence pack to choose a narrow, reversible pilot. More preparation can reduce uncertainty, but polishing every document before discovery delays learning. Prioritise representative cases and unresolved decisions.
For work spanning order sheets, internal checks and multiple tools, a custom n8n automation can be quoted against an agreed scope, subject to the required connections and review steps. You can bring your workflow examples to a discovery discussion with FlowAgent.