01
Current-workflow map
The process from its trigger to the business outcome, including where AI is already used.
01Autonomy Assessment
Where the workflow stops. And what would move it.
GreyShacks examines one workflow from its trigger to its outcome, including where people still intervene and why. You leave with a diagnosis, an autonomy boundary and a blueprint. Implementation is a separate decision.
Center
Workflow
TriggerSystemsRulesExceptionsApprovalsOutcome
TriggerSystemsRulesExceptionsApprovalsOutcome
Scroll
02What it is
A diagnosis of one workflow. With a boundary.
We investigate the actual workflow: its trigger, the applications and data it uses, the rules, the decision points, the exceptions, the permissions, the failure modes and how completion is verified. The written result separates facts supported by your evidence, assumptions that still need validation, and scenarios that still need testing. It is a diagnosis and a blueprint. It is not an automated software product, and it does not promise that the workflow can run without people.
Build
When a system has to be engineered into real work.
Buy
When a product already does the job, and your systems can hold it.
Wait
When data, permissions or ownership are not ready.
03Why companies stall
AI is in the building. The process is still manual.
Open each pattern. These are the ways a company can adopt AI and still leave most of the work with people.
- AI is present, but only a slice of each process has moved.
- Pilots prove a task and never reach the systems the live process needs.
- The loudest use case gets funded. The highest-value workflow does not.
- Autonomy, approvals and data access are left for the build to discover.
04Opportunity assessment
Eight lenses. One conversation.
Click a criterion. This is how we rank what is worth automating — so a CIO and a business owner can disagree with the same facts.
Scoring lens
Business value
Does this change a decision, a cycle time, or a cost the business already measures?
What “high” looks like
Tied to a named P&L or operating metric.
05How we work
From the live workflow to a blueprint.
Five moves. The written assessment can stand on its own.
Step 01
Map the workflow
From the trigger to the business outcome, including steps that already run and steps a person still finishes.
06What GreyShacks leaves you with
A brief you can act on.
02
Intervention and blocker analysis
Where a person still steps in, and whether the cause is data, a system, a rule, a permission or a missing test.
03
Exception and failure register
Failure scenarios marked as evidenced, assumed, or still untested.
04
Autonomy boundaries
What can run on its own, what needs approval, and what is not ready.
05
Remediation plan
The changes in priority order, including a recommendation to wait when the workflow is not ready.
06
Implementation blueprint
Integrations, permissions, testing and monitoring the next level would require.
07
Business case
Stated assumptions, and the metrics to measure after implementation. No promised percentage.
In the strategy
Every recommended workflow includes the access it would need, the actions it may take, and the approvals it requires. Safety is not a later chapter that stops the work. It is how a higher level of automation becomes operable.
- Data and systems in scope
- Actions that may run without a person, and actions that may not
- What must be true before a pilot is allowed to act
- How uncertainty and failure are escalated
Where the opportunity often sits
These are illustrations of the kind of work we examine. They are not claims about current GreyShacks customers or guaranteed outcomes. Automating one step is not the same as completing the case.
01
Sales
Today a seller researches an account, prepares notes, updates the CRM and drafts follow-up, even if a copilot already summarized a call. AI could assemble the account picture, draft the next note and prepare the CRM update. That needs CRM, mail and a rule for what may be written. A person should still approve outbound messages and any change to a commercial commitment.
02
Customer operations
Today a request is classified by hand, facts are looked up in several systems, and the case is updated after someone decides. AI could classify the request, retrieve the permitted facts and resolve the routine path. That needs the case system, the knowledge source and a policy for what may be closed. Anything outside the policy, or anything that changes a customer promise, should escalate.
03
Finance and administration
Today documents are read, figures are checked against a source, and exceptions are routed by email. AI could extract the fields, compare them to the record and route the mismatch. That needs the documents, the ledger or ERP, and validation rules. A person should approve anything that changes money, a posting, or a customer balance.
04
Manufacturing and supply chain
Today status is rebuilt from mail, the plant system and a tracker before anyone can answer. AI could consolidate the picture, watch for the defined exception and prepare the next action. That needs those systems and a rule for what counts as an anomaly. A change to a delivery promise or a financial record should stay with a named person.
05
Professional services
Today status, risks and client documents are assembled from inboxes and slides. AI could compile the status, draft the routine document and flag delivery risk against an agreed definition. That needs the engagement files and a permission model for commercial terms. Client communication and partner judgment should remain with the firm.