AI, automation or a person? Deciding how to fix a bottleneck

When something in the business is slow, the first instinct is to reach for a tool. The better first question is: what kind of work is this? The answer usually points to the right fix.

Step 1: Is the process clear?

Can you write the steps on one page, including who does each one and what “done” looks like? If not, fix the process first. Automating a process nobody can describe produces a faster version of the confusion. Often a named owner and a checklist resolves most of the problem.

Step 2: Does software you already own do this?

Before buying or building, check whether your CRM, helpdesk or project tool already has the feature, switched off or never configured. Configuration is the cheapest rung after process.

Step 3: What kind of work is it?

Sort the task into one of these:

Type of work	Example	Usually best served by
Rule-based and repetitive	Copying a form submission into the CRM and notifying the owner	Workflow automation
Reading or classifying text and documents	Sorting incoming enquiries, extracting fields from CVs	AI-assisted workflow, with human review
Needs judgment, empathy or conversation	Handling an upset customer, qualifying a complex lead	A trained person
Distinctive to your business and not available off the shelf	A client portal built around your method	Custom software
High volume, stable, and not your core work	Tier-one support, data entry	A managed service

Step 4: What happens when it goes wrong?

Ask what an error costs. If a wrong answer is cheap and visible, automation or AI with spot checks is reasonable. If an error is expensive or hard to reverse, keep a person in the loop for that decision, whatever the technology can do.

Step 5: Is the volume big enough?

A workflow that saves ten minutes a week is not worth building. As a rough check, estimate hours per month the task consumes, multiply by the cost of the person’s time, and compare with the cost of the fix over a year. If the fix is more expensive than the problem, choose a lower rung.

A worked example

A small agency spends two days each month building client reports by hand.

• The process is unclear: each client gets a slightly different report. Step one is to standardize the report.
• Once standardized, assembling the data is rule-based. Automate the data pull.
• Writing the commentary on what changed needs judgment. A person writes it, possibly with an AI draft to edit.
• The result: automation for the repetitive part, a person for the part that carries the value.

The principle

Always recommend the simplest effective option, and use more than one when the problem has more than one kind of work in it. Most real bottlenecks do.

If you would like help sorting your own bottleneck, a Business Operations Review is a good place to start.