Insight
AI assistant or simple automation: what does your business need?
If your business keeps repeating the same administrative task, start by mapping the task. You may need a straightforward automation, an AI assistant or a clearer process before either will help.
- Published
- Reading time
- 3 minutes
- Written by
Map the Task Before Choosing AI
If your business keeps repeating the same administrative task, start by mapping the task. You may need a straightforward automation, an AI assistant or a clearer process before either will help.
The useful question is what the work requires. Does it follow predictable rules, or does it involve understanding different questions and choosing an appropriate response?
01Use rules when the next step is predictable
A form submission can trigger a confirmation, create a lead record and notify the right person without an AI conversation.
The same applies to many reminders and routing tasks. If a known input should always produce a defined action, a rules-based workflow can be easier to understand and maintain.
For example, sending a new website enquiry to the correct sales inbox may need a clear form field and a routing rule. Adding a conversational layer would not automatically improve it.
02Consider AI when language varies
An AI assistant may help when customers ask similar questions in different ways or need guidance through approved information.
The work still needs boundaries. Define the information it can use, the actions it can take and the situations that require a person.
A booking assistant might explain the process and direct a customer to an approved reservation system. Confirming availability is a separate capability that requires a reliable connection and clear rules.
03Put the handoff on paper
Before building anything, describe the trigger, required inputs, expected output and person responsible when the task cannot be completed.
Decide what happens if information is missing or a connected service is unavailable. A useful failure path might save the request and alert staff rather than continue with an uncertain answer.
For customer conversations, make it easy to reach a person. Keep commitments, exceptions and sensitive decisions within the authority you have explicitly assigned.
04Start with one measurable workflow
Choose a narrow problem: enquiries copied manually between systems, repeated service questions or requests waiting for assignment.
Record how the task works now and what improvement would matter. That could be fewer manual steps, quicker routing or less time spent answering routine questions.
Test realistic examples and awkward cases before release. Include incorrect information, unavailable tools and requests outside the agreed scope. Review actual conversations or task logs after launch.
05Choose by the decision the task requires
| Task | Rules-based route | Where AI may help |
|---|---|---|
| Assign a structured form | Route by service or location field | Interpret varied free-text requests, then validate |
| Answer service questions | Approved FAQ lookup | Explain the same approved information conversationally |
| Confirm an appointment | Check calendar and booking rules | Guide the conversation; still use the real calendar |
| Approve an exception | Send to an authorised person | Summarise the request, without granting authority |
06Pilot the handoff before the conversation
Illustrative pilot: a form captures service and postcode, creates a CRM record and alerts the correct person. Add AI only if free-text questions create a real unresolved need. Test a missing contact detail, a request outside the service area and an unavailable calendar. Define whether each case stops, asks for information or reaches staff. Measure completed, correctly routed requests rather than impressive-looking answers.
Our AI and automation service scope starts with the workflow and its connections. For the sales process behind it, see how agency work connects CRM and follow-up.
NIST's generative AI risk guidance describes confident but false outputs as confabulation. Verify answers against approved information before they trigger a commitment.
If a known input should always produce a defined action, a rules-based workflow can be easier to understand and maintain.
Hugo Filipe
07Include the work after setup
Someone needs to maintain approved information, check failures and review costs. Software subscriptions, usage charges and support should be included in the scope discussion.
A small pilot can help reveal whether the proposed workflow fits your operation before you extend it. Do not promise savings until you have measured the actual result.
Studio 17 can help assess one workflow and define whether rules, an AI assistant or a combination makes practical sense. Start with the task you want to improve and the systems your team already uses.
Start with one workflow
Tell us about a repetitive task and the systems your team uses. We’ll help assess a practical next step.