Business Process Automation: What to Automate, What to Keep Human, and How to Calculate the Value
A practical guide to choosing the first workflow, calculating its current cost, deciding what stays human, and sequencing a 30-day first project.
— Craig Major
Business automation works when it removes a repeated constraint from a process people already understand. It fails when a company buys a tool before defining the workflow, the exception, and the person responsible for the outcome. This guide gives you a practical way to choose the work, estimate its current cost, and decide which steps software should handle.
What business process automation actually means
Business process automation is the use of software to carry out the repeatable steps of a defined business process, from the event that starts it to the outcome that ends it. That includes moving data between systems, applying agreed rules, generating documents or messages, scheduling work, and routing exceptions to a person. AI extends this by handling unstructured inputs such as email, documents, and notes, and by drafting outputs a person reviews. The important word is process. A tool automates a task. A system changes how a whole piece of work runs, who is responsible for it, and how you know it worked.
Automating a task is not the same as redesigning a workflow
Task automation removes a click: a form posts to a spreadsheet, a message triggers a notification. It is cheap, useful, and easy to abandon. Workflow automation redesigns the path from trigger to outcome: what starts it, what data is required, what decisions are made, who reviews what, what happens when something is wrong, and what evidence remains afterwards. Most disappointing automation projects are task automations sold as workflow projects. The steps get faster and the outcome does not change, because the constraint was a missing decision, unclear ownership, or unusable data rather than manual effort.
Good candidates for automation
- Frequent. The work happens often enough that a small improvement compounds.
- Rules-based. The decisions can be written down and tested by someone else.
- Measurable. A before-and-after number exists: time, delay, error rate, cost, or conversion.
- Accessible data. The inputs live somewhere a system can read with permission.
- Manageable exceptions. Unusual cases are a minority and can be routed to a person.
- Named owner. One person is accountable for the outcome after launch.
Poor candidates, at least for now
- Unstable process. Two people describe it differently, or it changed last month and will change again.
- Low volume. It is annoying but rare, so the build cost cannot be recovered.
- High judgment. The value of the step is the human reading of context, relationship, or risk.
- Unavailable data. The inputs are in someone's head, a locked system, or an inconsistent spreadsheet.
- Unclear owner. Nobody will maintain the rules, review the exceptions, or defend the result.
Map the process from trigger to outcome
Before estimating value, write the process down in one page:
- Trigger. The named event that starts the work.
- Inputs. The data and documents required, and where they live.
- Steps. What happens, in order, and who does it today.
- Decisions. Each choice made, the rule behind it, and who is allowed to make it.
- Exceptions. What goes wrong, how often, and what people do about it.
- Outcome. The end state, and the record that proves it happened.
If you cannot complete this page, that is your first finding. The process needs to be understood before it can be automated.
Calculate the current cost
Use a transparent model rather than a vendor estimate. This is the same model behind the Automation Opportunity Calculator:
annual_hours = people × minutes_per_occurrence ÷ 60 × occurrences_per_week × 48
annual_labour_cost = annual_hours × loaded_hourly_cost
recoverable_hours_low = annual_hours × repetitive_percentage × 0.35
recoverable_hours_high = annual_hours × repetitive_percentage × 0.65
Two deliberate choices keep the number honest. We use 48 working weeks rather than 52 to allow for vacation and statutory holidays, and the 35 to 65 percent recovery band is an explicit planning assumption, not a performance claim. Change it if you have your own evidence.
Illustrative calculation (hypothetical, not a client result)
Two coordinators spend 20 minutes per inbound request, 30 requests per week, at a loaded cost of $45 per hour. That is 20 ÷ 60 × 30 × 48 = 480 hours per year per coordinator-equivalent of handling, and roughly $21,600 in annual labour cost at that rate. If 70 percent of the work is repetitive, the planning band suggests recovering somewhere between 118 and 218 hours a year. Those are directional figures for prioritisation, not guaranteed savings.
Decide what stays human
Write the human boundary before the build, not after the first complaint. In most workflows the person should keep:
- judgment calls where context matters more than rules;
- sensitive exceptions, complaints, and anything with legal or safety weight;
- important relationships and the tone of what a customer receives;
- final accountability for the outcome.
Software should prepare the information, apply the agreed rules, draft the output, coordinate the steps, and make the state of the work visible. That is the difference between a system that makes people more capable and one that quietly makes decisions nobody can explain.
Compare your four real options
| Option | Best when | Watch for |
|---|---|---|
| Fix the process | The steps are inconsistent or duplicated and no software is required | Nothing sticks without an owner and a written standard |
| Use a tool you already own | Your CRM, help desk, or finance system already covers the workflow | Configuration effort and adoption are still real costs |
| Build a custom system | The workflow crosses systems, uses your own rules, and matters commercially | Scope, testing, ownership, and support must be defined up front |
| Add ongoing leadership | Several opportunities compete and no senior person owns the roadmap | Leadership is not implementation; budget both |
A 30-day first project sequence
- Days 1 to 5. Choose one workflow, name its owner, and complete the one-page map.
- Days 6 to 10. Establish the baseline with real numbers and calculate current cost.
- Days 11 to 15. Agree the future state, the human decision points, and the exception paths.
- Days 16 to 25. Build the smallest version that handles the common cases end to end.
- Days 26 to 30. Test normal, edge, and failure cases, train the team, and hand over ownership.
Where to go next
Run one workflow through the Automation Opportunity Calculator. If the process has enough volume, clear rules, usable data, and a named owner, the next step is either an AI Success Audit to rank it against the alternatives, or an AI Automation System to build it properly.
Frequently asked questions What is the difference between business process automation and AI automation?
Business process automation runs defined steps and rules across systems. AI automation adds the ability to work with unstructured inputs such as email, documents, and notes, and to draft outputs a person reviews. Most useful systems combine both. How do I choose the first workflow to automate?
Choose the workflow that happens often, has rules you can write down, has data a system can read, has limited exceptions, and has one person accountable for the outcome. Calculate its current annual cost before choosing a platform. What should stay human in an automated workflow?
Judgment calls, sensitive exceptions, important relationships, and final accountability. Document those boundaries before the build is approved so the system supports people rather than replacing decisions nobody can explain. How do I calculate the value of automating a process?
Multiply people by minutes per occurrence, divide by 60, multiply by occurrences per week and 48 working weeks to get annual hours, then multiply by loaded hourly cost. Apply a stated recovery range to the repetitive portion and treat the result as directional.
Related services
- Find the automation work worth doing first.
- Build the workflow your team should not have to carry by hand.
Continue with Flowgrammer
- Start with an AI Success Audit to identify the highest-value workflow opportunities.
- See the US manufacturing outbound system for an example of the work in practice.
- Estimate the opportunity with the Automation ROI Calculator.
- Book an AI Success Audit.