The cost of a manual process is not limited to minutes spent entering information. It includes active work, waiting between people, corrections, supervision, tools, and delayed opportunities. Measure one complete cycle using real cases instead of recalling an ideal task from memory.
The result does not need perfect precision. It needs enough credibility to compare three decisions: keep the process, improve it without software, or automate a clearly defined part.
Start with an outcome, not an isolated task
“Enter an invoice” is only one step. The process may begin when a document arrives and end when payment is approved, recorded, and traceable.
Define observable start and finish points. Then record every actor, system, decision, handoff, and exception. The Government of Canada Digital Standards, updated May 25, 2026, recommend understanding needs and problems with users, then starting small and improving iteratively. That discipline prevents a team from optimizing one step while moving the wait elsewhere.
Measure six kinds of cost
Active work
Time the minutes when someone reads, copies, classifies, checks, or decides. Use several normal cases and a few difficult ones. One clean file almost always understates reality.
The basic formula is monthly volume × average active time × loaded hourly cost. A loaded cost can include compensation, employer costs, and a reasonable share of role-related overhead. Document the assumption rather than looking for a universal number.
Waiting
A file may require fifteen minutes of work and remain blocked for three days. Measure the gap between stages, then identify what the delay prevents: invoicing, delivery, a customer response, resource booking, or a decision.
Waiting does not always convert directly into dollars. It still matters when it delays revenue, breaks a service commitment, or causes repeated follow-up.
Corrections and rework
Count incomplete records, duplicates, entry errors, and returns to a previous stage. For each category, measure frequency and correction time.
One average can hide an expensive exception. Separate simple fixes from cases that require investigation, a customer call, or an accounting correction.
Coordination
Add follow-up email, status meetings, internal messages, and searches required to find the work. These activities often exist only because the process has no shared state.
Tools and subscriptions
Include licences, forms, storage, isolated automations, and connectors already in use. A manual process may depend on several paid tools even when none carries the full journey.
Opportunity cost
Ask what the team would do with recovered hours. Freed time creates value only when it can be reassigned, avoid a hire, absorb growth, or improve customer response. Do not present every saved hour as immediate cash.
A transparent worked example
Consider an illustration, not a market statistic. A team processes 400 requests per month. Each requires 12 minutes of active work. The loaded planning rate is $45 per hour. Monthly active cost is therefore $3,600.
Eight percent of requests require a 15-minute correction. Those 32 cases add $360. The process already costs roughly $3,960 per month, or $47,520 per year, before follow-up, subscriptions, and delay impact.
This figure does not prove custom software will pay for itself. It creates a baseline for comparing the cost of change, ongoing ownership, and the share of work a solution can truly remove.
Observe the process for two weeks
Create a simple log and ask participants to record:
- case type;
- start and finish of every stage;
- actual active time;
- the person or system being awaited;
- the exception encountered;
- the cause of correction;
- final outcome.
Two weeks often reveals the main categories, but may miss seasonal peaks. Add historical data where volume changes sharply through the year.
Statistics Canada's June 11, 2026 research shows Canadian businesses using AI for areas including data analytics, text analytics, and virtual agents, while also identifying cost, privacy, and cybersecurity barriers. That supports a proportional diagnostic: technology should respond to an observed cost and include its own risks.
Separate three kinds of problem
The process is unnecessarily complicated
Remove approvals that add no value, unused fields, and unread reports first. Automating an unnecessary step only makes it faster.
Information moves poorly
An integration or shared view may be enough when current tools perform their jobs but exchange data badly. A new central application is not always required.
A repetitive decision consumes judgment
Rules fit explicit conditions. AI becomes relevant when variable documents, messages, or images need interpretation. Our rule-based automation versus AI guide explains the distinction.
Build a credible change scenario
Do not assume 100% of the work disappears. Estimate separately:
- the portion eliminated;
- the portion shortened;
- cases that remain manual;
- human review time;
- software, integration, and maintenance cost;
- adoption effort.
Prepare cautious, expected, and favourable scenarios. If a project works only in the favourable case, revisit the scope or assumption.
Choose a proportional next step
Start with a process whose volume, friction, and owner are identifiable. Measure the baseline, simplify the rules, and test the smallest change that removes one complete loop.
Project Factory can connect a focused capability through AI integration or build a custom web application when a new tool must carry the process. The Workind project illustrates a platform bringing distinct roles and journeys together. To compare the investment over several years, read our custom software cost guide.