Most automation projects start in the wrong place. Someone sees an impressive AI demonstration, signs up for a platform, and then looks around the company for something to do with it.
That is how businesses end up with another subscription, three half-built workflows, and employees quietly returning to the spreadsheet that already worked.
Start with the work instead. Find a process that happens often, consumes real time, follows a recognizable pattern, and has an owner who wants it fixed. Then decide whether the answer is a checklist, a built-in feature, traditional automation, AI, or a custom application.
You can identify the best first candidate in about 30 minutes. Here is the workflow audit we use.
First, Define a Workflow
A workflow is not a department and it is not a vague complaint such as "sales takes too long." It is a repeatable path from a trigger to an outcome.
- Trigger: What starts the work?
- Input: What information arrives, and in what format?
- Steps: What does someone read, enter, compare, approve, or send?
- Outcome: What must be true when the work is finished?
- Owner: Who is responsible when the process stalls or produces the wrong result?
For example, "onboarding" is too broad. "Create accounts and assign equipment after a manager submits an approved new-hire form" is a workflow. It has a clear beginning, a visible finish, and decisions that can be examined.
If the team cannot describe the process at this level, do not automate it yet. First make the work understandable.
Minutes 0 Through 5: Collect the Friction
Ask the people doing the work one question: What do you repeatedly copy, chase, retype, check, or explain?
Do not start with executives guessing what employees find inefficient. Ask the person who opens the shared mailbox every morning, prepares the weekly report, creates user accounts, reconciles invoices, or tracks down missing documents.
- "I have to check three places"
- "We wait for someone to email us back"
- "I copy this into another system"
- "It depends on who submitted it"
- "I rebuild the same report every Friday"
- "Only one person knows how to do it"
- "We usually catch the mistake later"
Those sentences reveal the handoffs, delays, and hidden knowledge that automation can address.
Minutes 5 Through 10: Pick Three Candidates
Choose three workflows that happen at least weekly or create a meaningful delay when they occur. Give preference to work that affects customers, cash flow, security, compliance, or the ability of another employee to continue.
Do not choose the most ambitious idea. Your first automation should be easy enough to understand and important enough that people notice when it improves.
- Lead or customer-request intake
- Invoice and document processing
- Employee onboarding and offboarding
- Recurring management reports
- Compliance evidence collection
- Support-request classification and routing
- Approval reminders and status updates
"Build an AI assistant for the company" is not a candidate. "Read support emails, identify the affected customer and system, and draft a categorized ticket for technician review" is.
Minutes 10 Through 18: Map What Actually Happens
Pick one candidate and map the current process from beginning to end. Use a whiteboard, paper, or a plain document. The format matters less than getting the real sequence out of people's heads.
For every step, record: 1. Who performs it 2. What system they use 3. What information they need 4. What decision they make 5. What happens when the information is missing or wrong
Pay special attention to waiting. A task may require only ten minutes of labor but take three days because it sits in inboxes between each step. Automation often creates more value by removing that delay than by eliminating the ten minutes.
Also mark every place information changes format. Email becomes a spreadsheet row. A PDF becomes data in an accounting system. A phone call becomes a ticket. These transitions are common sources of retyping, mistakes, and lost context.
Minutes 18 Through 25: Score the Opportunity
Score each category from zero to two. Zero means weak, one means mixed, and two means strong.
Frequency: Does this happen daily or many times each week?
Time: Does each occurrence consume meaningful employee time or create substantial waiting?
Consistency: Do most cases follow the same steps and rules?
Handoffs: Does information move repeatedly between people, inboxes, or systems?
Error cost: Do missed steps or bad data create rework, delayed billing, security exposure, or customer frustration?
Data readiness: Are the required inputs available digitally and reasonably consistent?
Control: Can the risky decisions remain behind a clear human approval?
- 10 to 14: Strong pilot candidate
- 6 to 9: Improve the process or data first, then reconsider
- 0 to 5: Leave it alone for now
The score is not a financial model. Its job is to stop the loudest complaint or newest software demo from automatically becoming the next project.
Minutes 25 Through 30: Establish the Baseline
- How often the workflow runs
- Average hands-on time per case
- Average elapsed time from trigger to completion
- Number of corrections or reopened items
- Number of people involved
- Cost of the software already supporting it
Without a baseline, every automation looks successful because the team remembers the old process as worse than it was. Measure first. Review the same numbers after 30 days.
Decide Whether the Workflow Needs AI
Not every automation problem is an AI problem. Most reliable systems combine fixed rules with AI only where the input is messy.
Traditional automation is best when the instructions are exact: create an account after approval, copy an approved field into another system, send a reminder after two days, or generate the same report every Friday.
- Reading emails and documents to identify intent or extract fields
- Summarizing a long record for review
- Categorizing requests that arrive in everyday language
- Drafting a response from approved source material
- Comparing evidence against a defined checklist
- Answering internal questions while citing the underlying documentation
The best design is often straightforward: AI reads or drafts, ordinary software applies the rules, and a person approves the consequential action.
Keep Human Approval Where Consequences Are Real
Speed is not the only goal. A useful workflow also makes responsibility clearer.
- Transfer money or change payment instructions
- Make a legal or contractual commitment
- Disable critical accounts or infrastructure
- Send sensitive information to an unverified recipient
- Make a final hiring, disciplinary, medical, or compliance determination
- Apply destructive security remediation without a recovery path
The system can assemble the information, flag exceptions, and recommend the next step. The accountable person should still approve the decision. As the workflow proves itself, low-risk cases can become more automatic while exceptions receive more attention.
Six Strong First Automations
Lead intake and routing
A form or email arrives. The workflow verifies required information, identifies the requested service, creates the record, assigns an owner, and drafts the first response. A person reviews unusual or high-value requests.
Document intake
Invoices, applications, contracts, or evidence files arrive in different formats. The workflow extracts the required fields, names and files the document, identifies missing information, and sends questionable items for review.
Employee onboarding
An approved request triggers account creation tasks, equipment preparation, licensing, training assignments, and manager reminders. The workflow tracks completion without granting access that has not been approved.
Recurring reporting
The workflow collects data from the systems already in use, applies consistent calculations, creates a draft report, and highlights changes that deserve explanation. The owner reviews the narrative before distribution.
Compliance evidence collection
The workflow requests recurring evidence, checks whether submissions match the expected period and system, records what was received, and flags gaps. An assessor still determines whether the evidence is sufficient.
Service-request triage
Incoming requests become structured tickets with the customer, affected system, urgency, and relevant history attached. A technician starts with context instead of spending the first fifteen minutes gathering it.
What Not to Automate First
Some processes should wait, even if they are frustrating.
Avoid workflows that happen only a few times a year, change every time, rely on information nobody records, or have no clear owner. Do not automate a broken approval chain simply to make bad decisions happen faster.
Also be cautious when the promised savings depend on removing expert judgment. A good first project gives experts better information and less clerical work. It does not pretend their experience is a sequence of obvious rules.
Buy, Configure, or Build?
Once you have selected the workflow, check the software you already own. Microsoft 365, Google Workspace, your CRM, accounting platform, and line-of-business systems may already include the trigger, approval, or integration you need.
Use an existing feature when it fits the process cleanly. Use a general automation platform when the workflow is mostly moving structured data between common systems. Consider a custom application when the process is specific to your business, crosses several disconnected systems, handles regulated information, or needs an interface your team will actually use.
If you are deciding whether custom software is justified, our article Stop Renting AI: Why Custom Business Applications Beat Another SaaS Subscription provides a deeper build-versus-buy test.
Start With One Measurable Win
The first project should run alongside the existing process long enough to catch surprises. Limit it to one team, keep an approval step, log what the system did, and review exceptions weekly.
After 30 days, compare the new numbers with the baseline. Did hands-on time fall? Did work finish sooner? Were fewer items corrected? Did customers or employees get a faster response? If the answer is yes, expand carefully. If not, fix the workflow before adding more technology.
Automation should not begin with a promise to transform the company. It should begin with one piece of work that everybody understands and nobody wants to keep doing manually.
KaselTech helps businesses map these workflows, secure the data involved, and build the right combination of automation, AI, and human approval. Explore our AI and custom application services or contact us to bring us a process your team wants off its plate.