Describe a task, rather than an entire department
List the tasks your team repeats and consider one workflow at a time. Looking up shipment information, preparing a contact for follow up or moving data from a file into a report are tasks you can observe. Automating logistics, sales or administration is still too broad a goal.
For each candidate, write a short description using a real example:
- Starting point: the event or file that triggers the task.
- Current work: what a person looks up, copies, interprets or updates.
- Output: the information or action needed by the next person in the process.
- Owner: who uses the output and who reviews an exception.
- Tools: where information arrives and where it needs to end up.
Compare manual work and consequences
Observe the same type of task over a period that reflects normal operations. Record how often it occurs, the working minutes it requires and what needs to be repeated when information is missing. Separate working time from the time a case spends waiting.
- Frequency: does the task occur regularly or only in isolated situations?
- Intervention: does someone look up, copy or assemble the same data in each case?
- Consequence: does a delay prevent a delivery, conversation or report from moving forward?
- Verification: can someone recognize a correct output and detect an error?
A useful first candidate combines a problem worth solving with a workflow you can verify. If a task repeats but no one can define a correct output, clarify the process first.
Check the data and decide what needs review
Collect a typical case and another that needs a different approach. Check whether the information exists, whether the team can access it and whether it may be used for this work. Do not confirm an automation based only on the cleanest example.
- Available input: identify the file, record or message and the required fields.
- Possible access: check permissions and tool capabilities before promising a connection.
- Recognizable exception: define what happens when data is missing, two records disagree or a source does not respond.
- Assigned review: specify who receives the case and what they need to decide.
If access is missing, rules change in every case or no one can review the output, resolve that dependency before automating. A decision that requires judgment can stay with a person while the information they need is prepared for them.
Define a first workflow using these examples
These cases describe specific work by ZaTek. Each example separates what was developed from what to consider when assessing a similar process in your business.
Shipments: bring information together for operational review
The case: In the Klozer example, shipment processes required manual responses. ZaTek worked to centralize that management in Slack, with a summary and information organized in one place.
For your business: Choose a recurring query and define what information the team needs, where it comes from and which missing data requires review. The first test should make that workflow verifiable before covering the entire logistics operation.
Contacts: prepare the next commercial step
The case: The Delma Group system brings together lead capture, enrichment and qualification, conversations and campaigns with an embedded chatbot.
For your business: Choose one segment, such as preparing a contact for follow up. Define the data needed by the person continuing the conversation, what might be missing and who validates the information. This example does not require automating the entire sales journey.
Documents: move from an input to a verifiable output
The case: For Effect Doctors London, ZaTek developed PDF and CSV processing, structured biomarker extraction and PDF clinical report generation.
For your business: Choose a file type, the required fields and an output the team can compare with the original. Define what happens with an incomplete file.
Agree on how to verify the first version
Before building, write down what should happen with the typical case and the exception you chose. The owner should be able to verify that the output allows work to continue, missing data stays visible and they know which cases need review.
- Compare the same workflow before and after, using the same type of example.
- Record manual work, corrections and cases awaiting review.
- Check whether the team finds and uses the output without reconstructing information elsewhere.
- Review exceptions before adding another task or tool.
Fewer manual steps do not by themselves demonstrate cash savings. Expanding requires observing how the process works and what happens to the team’s time. For a first conversation, bring the task description, a typical example, an exception and the expected output.
Common questions
What if we have several candidate tasks?
Compare their frequency, manual work and consequences. Then check whether you have data, access and someone who can validate the output. Start with a useful workflow you can observe and review.
When should we organize the process before automating?
When the starting point, required output or person responsible for exceptions is unclear. The same applies when data or access is missing. Clarifying these dependencies helps identify which part can be automated.
When is AI useful?
When the work requires interpreting content or context and you can define how to review the result. Connecting tools or following a defined sequence may not require AI. The task and its validation guide the choice.