Most companies do not have an AI problem. They have an operations problem.
Most AI initiatives fail for the same reason. They start with tools instead of work.
If the goal is reliable ROI from AI and automation in 2026, there is a method that consistently works across industries and company sizes.
Shadow the work.
Document it as an SOP.
Map inputs, outputs, and systems.
Identify repetitive steps.
Prioritize with a scorecard.
Build.
That is the full loop.
What fails in practice is not the idea of automation. What fails is stopping halfway. Teams automate one slice of a workflow, leave the rest untouched, and then live in the painful middle where some steps are automated, some are manual, and no one has clarity on where responsibility begins or ends.
This article lays out the complete end to end method.
The Outcome This Process Produces
When done correctly, this process produces clarity, not just automation ideas.
By the end, organizations should have:
Documented SOPs for the workflows that actually matter
A list of automation candidates tied to real work
A prioritized automation roadmap for the next six to eighteen months
Cost bands for implementation showing what a smaller build versus a larger build looks like
An ROI model for each initiative
Clear understanding of which systems are being integrated and why
This is the foundation required before implementation begins.
Step One: Start With the Freed Up Hero
The highest ROI AI opportunities usually come from freeing up top performers.
The right starting questions are not about tools. They are about work.
What outcomes matter this quarter
What is blocking those outcomes
Who is closest to the work
Which tasks consume their time and attention
The list of tasks that drain time becomes the backlog. The critical follow up is converting that backlog into a roadmap that can actually be built.
Step Two: Shadow the Work Instead of Interviewing It
This is the most important step and the one most teams skip.
The work must be observed directly.
Not a hypothetical walkthrough. Not a whiteboard explanation. The real workflow, in real systems, with real edge cases.
During observation, the following questions surface the complexity that kills automation projects later:
Why is this done this way
What exactly is happening at each step
How is the work executed
When is the process skipped or altered
What output is expected
What happens when the output is wrong
What inputs are required
What happens when inputs are incomplete or incorrect
These answers expose where automation will succeed and where it will fail if assumptions are wrong.
Step Three: Turn Observation Into a Real SOP
After shadowing, the workflow is documented as a real SOP.
A useful SOP includes:
The trigger that starts the process
The inputs and what valid input looks like
The exact steps including decision points
The outputs and where they go
Exception handling for unexpected inputs or outputs
The systems involved such as CRM, email, ERP, documents, or spreadsheets
Ownership for execution, approval, and notification
This document is not bureaucracy. It is the specification.
Even without automation, this step often produces time savings by forcing process reengineering. Many workflows were created during emergencies and never revisited. Writing them down exposes unnecessary steps.
Step Four: Identify Repetitive Work That Is Suitable for Automation
With the SOP in hand, automation candidates become obvious.
Common candidates include:
Data gathering and normalization
Extraction and structuring from documents
Repetitive analysis and validation
Drafting summaries, reports, and emails
Routing, approvals, and notifications
Reconciliation and discrepancy detection
Exception triage and flagging
AI is most useful when the work is repetitive but messy. Unstructured text, variable formats, and ambiguous inputs paired with clear success criteria create the best conditions for automation.
Step Five: Map Systems and Data Flows
Before building anything, systems and data flows must be mapped.
This includes:
Source systems where inputs originate
Destination systems where outputs must land
Access constraints such as APIs and permissions
Data formats including structured and unstructured
Triggers and frequency such as event based or scheduled
This is where tool demos fail and real roadmaps survive.
Step Six: Decide Between Tools and Custom Code
Some workflows can be handled with off the shelf tools.
However, for core operational workflows with unique data, edge cases, and multiple systems, custom code is usually more durable.
Code tends to offer:
Fewer brittle failure points
Better testability and reliability
Improved long term maintainability
Greater control over exceptions and audit requirements
If the workflow matters, it should be treated like software.
Step Seven: Prioritize With a Scorecard
Once automation candidates are identified, they must be prioritized.
A practical scorecard includes:
Strategic alignment with existing plans
Revenue impact or cost reduction
Operational effectiveness gains
Stakeholder impact and workload reduction
Market demand if applicable
Competitive differentiation
Cost versus benefit
Each initiative can be scored from one to five and ranked. The goal is not precision. The goal is clarity and agreement.
Step Eight: Produce the Automation Roadmap
At the end of the process, organizations should have:
A prioritized list of automation initiatives
Estimated timelines
Budget ranges per initiative
ROI models tied to time savings and impact
Dependencies and risks
This roadmap becomes the operating plan for AI and automation over the next several years.
Example: Commercial Real Estate Underwriting
Commercial real estate underwriting is a strong example of high ROI automation.
The workflow includes:
Large volumes of properties
Structured public data
Unstructured documents such as rent rolls and financial statements
Repeatable analysis and assumptions
A clear output in the form of an investment memo
A feedback loop where most properties fail quickly
This work is repetitive, document heavy, and high volume. That makes it ideal for:
Automated data collection
Document extraction into structured fields
Repeatable calculations
Drafting investment memos with assumptions
Exception handling when data is incomplete
The key requirement is defining inputs, outputs, expected values, and exception behavior through the SOP first.
If This Is Needed as a Packaged Engagement
This is the process used in an Automation Mining and Deep Dive engagement.
It includes:
Shadowing and documenting workflows
Producing SOPs
Identifying automation candidates
Mapping systems and integrations
Building an ROI backed roadmap with cost bands and prioritization
This work happens before implementation begins and is designed to ensure that automation investments deliver measurable returns.