How to Choose Your First Automation Project
Start with a task someone already repeats and can explain. Before choosing software or an AI agent, find out how often the task happens, how long it takes, what information it needs, and what happens when it goes wrong. Those answers give you a basis for deciding whether automation is worth the investment.
Choose a recurring task your team can measure, define where a person needs to review it, and calculate whether the time recovered justifies the cost. A worked retail example shows how.
Compare the work your team actually repeats
For one working week, record recurring tasks in a short list. Include the number of times each task occurs, minutes spent, systems involved, and the person responsible. Use observed work or existing records rather than a guess about an ideal week.
A retailer might compare answering order-status questions, approving returns, and preparing stock reports. Order-status replies could be a useful first project if the order and carrier data are accessible. Return approvals may need more judgment, while stock recommendations depend on reliable sales and inventory data. The choice comes down to your actual process.
Use the same approach for a restaurant checking supplier orders, a manufacturer preparing quote information, or a service business collecting intake details. Start with one task that has enough volume to matter and a result you can check.
- 01Frequency: how many times does it happen in a normal week or month?
- 02Effort: how many minutes does each instance take, including corrections?
- 03Inputs: is the required information available and current?
- 04Ownership: who can explain the process and review the result?
- 05Exceptions: which cases need a person, and how often do they occur?
Define one workflow from trigger to review
Consider an ecommerce customer asking where an order is. A focused first version retrieves the matching order and the latest carrier update, then prepares a reply for a customer-service person to approve. The project ends at an approved draft. It does not also change an address, issue a refund, or make a delivery promise.
This is an illustrative workflow, not a claim about a client result. Whether it is feasible depends on access to the shop, carrier data, and support system.
- 01Trigger: a customer sends an order-status question.
- 02Inputs: a verified order match, current tracking information, and approved response rules.
- 03Output: a factual draft with links to the information used.
- 04Review: a person checks the order, wording, and next action before sending.
- 05Exceptions: missing tracking, an uncertain order match, complaints, refund requests, and address changes go to a person.
Write these rules before connecting tools. If nobody can explain what a correct reply looks like, review a few real cases together first.
Use AI where the task calls for it
A fixed rule can retrieve an order by its identifier or route a message to a queue. An existing integration might already handle that step. Check what your current software can do before paying for a custom build.
AI may help interpret differently worded requests or draft a reply from verified facts. Keep the order lookup and permission checks explicit. An agent should flag missing information rather than invent a shipping update.
For the first version, choose the simplest approach that meets the requirement. You can expand it once you have evidence about accuracy, time recovered, and the review effort.
Calculate the value after review and upkeep
Here is an illustrative calculation in US dollars. These are planning assumptions, not Interzekt prices or a measured customer outcome. Replace every number with your own volume, costs, and observed timings.
Suppose a shop receives 500 order-status questions a month. At four minutes each, manual handling takes about 33.3 hours. If a draft still needs one minute of review, that leaves 8.3 hours of review and frees 25 hours a month.
At an assumed labor value of $30 an hour, those 25 hours represent $750 of monthly capacity. Subtract $125 for software and two hours of upkeep at $30 an hour. The remaining capacity value is $565 a month.
With an illustrative setup cost of $1,800, the capacity-based payback would be about 3.2 months if all of that recovered time is useful. If only half is useful, monthly value falls to $190 after the same recurring costs, and payback rises to about 9.5 months.
Freed time does not automatically become cash savings or profit. Salaries may stay the same. To count financial savings, identify an expense you can actually reduce. To count additional capacity, specify useful work the team will do with the time. Include implementation, maintenance, review, and exception handling in the calculation.
- 01Monthly hours recovered = task volume × minutes saved per task ÷ 60.
- 02Monthly capacity value = useful hours recovered × hourly labor value.
- 03Net monthly value = capacity value minus software, upkeep, and other recurring costs.
- 04Capacity-based payback = setup cost ÷ positive net monthly value.
Run a pilot with a clear decision at the end
Choose a limited queue or task type, keep drafts under review, and compare them with the current process. Include ordinary cases and known exceptions. Record the time spent checking drafts, corrections required, wrong matches, and requests that still need manual handling.
Set your acceptance criteria before the pilot. For example, you might require every reply to match the right order, missing tracking to reach a person, and total handling time to fall after review and upkeep. Use a time target that makes the cost calculation work for your business.
At the end, decide whether to continue, revise the scope, or stop. A draft that takes longer to verify than a manual reply is useful evidence that this version needs to change.
- 01Measure total handling time, including review and exceptions.
- 02Track factual errors and the edits reviewers make.
- 03Check whether the team can recover when a system or data source is unavailable.
- 04Compare recurring cost with the capacity you can actually use.
What to keep
- 01Choose a recurring task with accessible inputs, a clear owner, and a result you can check.
- 02Scope the first version around one trigger, one output, and explicit exceptions.
- 03Check existing rules and integrations before adding an AI agent.
- 04Calculate value after review, upkeep, and software costs.
- 05Use the pilot to decide whether the project deserves further investment.
Frequently asked
What should a small business automate first?
Compare recurring tasks using observed volume, time spent, access to data, review effort, and exception frequency. An order-status draft, intake checklist, or recurring report can be a candidate when the inputs and expected result are clear. The right choice depends on your process and costs.
Does the first automation project need an AI agent?
No. A rule, template, or existing integration may cover the task. AI can be useful when requests vary in wording or a draft needs to be prepared from verified information. Choose it for a specific requirement and measure the review effort it creates.
How much should a first automation project cost?
There is no single useful price without a defined scope. Estimate setup, software, maintenance, review, and exceptions, then compare those costs with capacity you can use or expenses you can reduce. The $1,800 setup figure in this article is an illustrative assumption, not a quote.
Interzekt Editorial Team
Editorial Team at Interzekt
The Interzekt Editorial Team publishes practical reporting and guidance on AI systems, automation, and the operating decisions behind useful technology.