A business asking to “use AI” for incoming enquiries is usually describing more than one job. The message may need to be captured, interpreted, routed, summarised, drafted into a reply and tracked for follow-up. Some of those steps suit rules-based automation. Some suit AI. Some should not be touched until the process itself is clearer.
That is the practical decision for SME leaders. Do not start with the technology label. Start with the work, the risk if something goes wrong, and how much judgement the task actually needs. If you automate a messy process too early, you can lock the confusion in and move it faster.
Start with the work, not the tool
The first question is simple. What is the process really doing?
If the work is repetitive, structured and rule-driven, straightforward automation is usually the better fit. If the work involves varied wording, inconsistent documents or interpretation, AI may help. If the process has duplicate approvals, unclear ownership or unnecessary steps, the right answer may be to simplify it first.
That is why the same request can lead to different solutions. A reminder after a missed deadline is not the same as understanding a free-text email from a customer. A clean hand-off is not the same as an approval chain that nobody fully owns.
Rules-based automation in plain English
Rules-based automation follows instructions you define in advance. If this happens, do that.
Examples include:
- moving completed form fields into a CRM
- sending a reminder after a fixed number of days
- creating a task when a status changes
- routing a request to a named person when a clear category is selected
- updating a record from a known source
These tasks often need integration, clear logic and someone responsible for maintenance. They do not usually need language understanding. For an SME, that can make them quicker to implement and easier to control.
A useful test is whether the step can be described as a fixed rule without much ambiguity. If it can, automation may be enough.
Where AI adds value
AI is most useful when the task involves interpretation rather than a fixed rule. That often means free text, inconsistent formats or information that arrives in different shapes.
Examples include:
- classifying an enquiry by topic when the wording varies
- extracting contact details, dates or order information from differently formatted documents
- summarising a long email thread for a colleague
- drafting a reply using approved information and wording
AI can speed up those tasks, but it is not a guarantee of correctness. It can miss context, misread a request or produce a confident answer that is not quite right. That is why validation matters. The higher the consequence of a mistake, the more human review you need.
A practical enquiry workflow example
Here is one illustrative SME enquiry workflow, showing where each approach fits.
| Step | Appropriate approach | Why |
|---|---|---|
| Capture the enquiry | Automation | The message can be logged, timestamped and given a reference automatically. |
| Identify the subject | AI | The wording may be varied, so interpretation helps. |
| Route it to the right owner | Automation | Once the category is known, a clear rule can assign it. |
| Draft a reply | AI with approved content | AI can prepare a first draft, but it should stay within controlled information. |
| Check commitments or unusual requests | Human review | A person should confirm anything that could create a promise, exception or risk. |
The point of the table is not to make AI do everything. It is to place each step where it adds the most value. In many workflows, that means a mix of automation, AI and human oversight rather than one approach on its own.
When the process needs fixing first
Some slow processes are not slow because they are manual. They are slow because they are poorly designed.
Common issues include:
- duplicate approvals
- unclear ownership
- unnecessary data entry
- inconsistent definitions
- steps that exist only because “that is how we have always done it”
If those problems are still in place, technology will not remove them. It will usually reproduce them more quickly.
For example, if three people believe they own the same enquiry, automation will not resolve the confusion. It may simply route the confusion faster. That is why process review should come before software choice. Simplify the workflow first, then decide what should be automated and what, if anything, needs AI.
A simple decision framework for one workflow
Use these questions to assess a single process. They are a guide to judgement, not a rigid scoring system.
- Is this step necessary, or could we remove or simplify it?
- Are the inputs structured and the rules clear?
- Does the task involve interpreting varied text or documents?
- What happens if the output is wrong?
- Which decisions need human judgement or approval?
- Is the volume of work enough to justify implementation and maintenance?
A sensible next step is a small test. Pick one part of the workflow, define what success looks like in plain terms and check whether the change reduces manual effort without creating new problems. If the risk is low, keep the control lighter. If the mistake could affect customers, cash flow or compliance, build in more review.
Hally AI starts with the business problem
At Hally AI, we begin with the process, not the pitch. We look at what the business needs to achieve, where time is being lost and which steps are worth changing. Then we assess whether simplification, automation or AI is the most useful fit.
That approach keeps the solution proportionate. It also helps SMEs connect existing systems, make better use of their data and implement practical solutions that save time, improve decision-making and create capacity without adding unnecessary complexity.
Have a process that takes more time than it should? Book a discovery call with Hally AI to explore whether simplification, automation or AI is the most useful next step.
