Choosing an AI automation consultant: twelve questions that matter.
The ones that separate real practices from resold templates. A checklist you can use tomorrow, before you sign a diagnostic or sit through another demo.
- A serious consultant talks about workflow, data, permissions, handover, and measurement before tools.
- Ask for a written diagnostic, named custodian, test data plan, red lines, and post-handover ownership.
- The fastest warning sign is a vendor who promises a general AI system without spending time inside your operation.
Most owners do not lack options. They lack a way to separate a considered automation practice from someone reselling a prompt library, a chatbot template, or a subscription with nicer words around it.
The difference shows up before the proposal. It shows up in the questions they ask, what they refuse to promise, and whether they can explain how the system will keep working after the launch meeting is over.
1. What will you observe before you propose?
A real diagnostic begins with the work as it happens. Inbox patterns, WhatsApp volumes, spreadsheet ownership, escalation habits, approval loops, exceptions. If the consultant can price and scope the engagement after one sales call, the scope is probably a template.
2. Which workflow will you refuse to automate first?
Every business has attractive but wrong first projects. They are visible, politically important, and too broad. A serious consultant will narrow the start, often more than the owner wants.
3. Who is the custodian on our side?
AI systems need an internal owner. Not a committee. One person who understands the workflow, reviews exceptions, and notices when the system drifts. If the consultant does not ask for this person, the handover has not been designed.
4. Where will the data live?
Ask this literally. Which folders, systems, inboxes, tables, or APIs. What is read-only. What is write-enabled. What is never sent to a model. Good answers are boring, specific, and written down.
5. What permissions will the agent have?
An agent that can draft is different from an agent that can send. An agent that can read Tally exports is different from one that can edit the books. The permissions model is the system's spine.
6. What are the red lines?
For a CA firm, it might be tax advice without partner review. For a coaching institute, it might be fee discounts. For a resort, it might be medical or safety commitments. Red lines must be explicit before the first demo.
7. How will you measure success?
Look for one sentence with a number. "Reduce first-pass reconciliation review from two hours to twenty minutes." "Respond to parent enquiries within thirty seconds after 9pm." If success cannot be counted, the rollout cannot be governed.
8. What happens when the model is wrong?
The answer should include confidence thresholds, escalation, logs, review queues, and a human owner. If the answer is "the model is very accurate," keep asking.
9. What will my team see every week?
Weekly demos should use your data, not sample data. The team should see the system becoming real while there is still time to correct it.
10. What do we own at the end?
Ask about credentials, source code, documentation, prompts, workflow diagrams, logs, and vendor accounts. If the answer is unclear, you are buying dependency, not automation.
11. How do you handle Indian-language work?
For Hindi, Marathi, Gujarati, Bengali, Tamil, Telugu, Kannada, or mixed English-Hindi WhatsApp, ask for native-speaker review and clear boundaries. Language support is not just translation. It is tone, idiom, and escalation.
12. What do you maintain after launch?
The frontier moves. So do staff habits, documents, policies, and customer questions. A serious proposal includes custodianship: reviews, updates, logs, and a way to retire or expand the system without drama.
If the first half of the conversation is about tools, models, and demos, be cautious. If it is about your work, your data, your people, and your boundaries, keep listening.
A checklist you can use tomorrow
- Ask for the diagnostic plan.What will they observe, who will they speak to, and what artifact will you receive?
- Ask for the first workflow.One workflow only, with a number attached.
- Ask for the permissions map.Read, draft, approve, send, edit, delete.
- Ask for the handover list.Everything you own when the engagement ends.
- Ask for the first-month operating rhythm.Who reviews exceptions and how often.
A good consultant will welcome these questions. They make the work cleaner. They reduce ambiguity. They protect the business on both sides of the table.
Want us to answer these questions for your first workflow?
Book a 30-minute conversation. We will tell you what we would scope, what we would avoid, and what has to be fixed first.
Book a discovery call ↗