Audit report · 01 Readiness score
Ready with conditions.
- Workflow and systems are in place.
- Data and ownership need work first.
- Two fixes stand between this and a pilot.
AI Audits & Readiness
Before you scale AI,
find out what will break.
Most AI projects don’t fail on the AI. They fail on data, ownership, access, and processes nobody wrote down. We check all of it before you spend more.
Assess our readiness
Pre-scale inspection
Support reply drafting
- WorkflowsMapped end to endReady
- DataTwo sources disagreeFix
- SystemsAccess confirmedReady
- ControlsNo owner for exceptionsHold
- Operating conditionsReview time not budgetedFix
Hold
Exceptions have no owner. Fix before scaling.
Why pilots stall
It worked in the demo. Then it met your business.
Most failed AI projects are not model problems. They fail on the parts nobody puts in a demo.
Where is order 1042?
Shipped Tuesday. Tracking link sent.
Looks great
- Order numbers arrive in three formats.
- Nobody owns the refunds queue.
- The tool can’t reach billing.
- Security hasn’t reviewed it.
- No one has time to check the drafts.
How the audit works
From “we think we’re ready” to a clear call.
-
01
Pick the initiative
Start with the use case you want to scale, or the one you are deciding on.
-
02
Inspect what it depends on
Workflows, data, systems, access, security, and who reviews the output.
-
03
Flag what will break
Every gap gets a severity, an owner, and a fix.
-
04
Give you the verdict
Go, go with conditions, or not yet, with the fixes in order.
What you leave with
You get a verdict, not a pile of findings.
Every audit ends with a clear call, the fixes in order, and an owner for each one.
- GoReady to pilot, with the measures to judge it.
- Go, with conditionsFix these first, in this order.
- Not yetBuilding now would ship the gaps.
Unsure if you’re ready? That’s exactly what the audit is for.
Before hiring developers, buying software, or launching another AI experiment, check the ground it will stand on.
Assess our readinessUnder the hood
What usually gets missed, the areas we review, the report you receive, and which audit fits your question.
What usually gets missed
Most failed AI projects are not model problems.
An audit finds these issues before they become expensive.
The modelWhat everyone looks at
What they usually are
- bad data
- unclear ownership
- weak processes
- missing permissions
- no evaluation standard
- poor integration
- undefined risk boundaries
- a use case nobody really needed
What we review
We walk through
the whole building.
Depending on the engagement, we open some or all of these rooms. Choose one to see what we ask inside.
Business use cases
“Is this a problem worth solving, and does anyone own the outcome?”
Process maturity
“Is the work written down, and done the same way twice?”
Data availability
“Does the data exist, is it current, and are you allowed to use it?”
System access
“Can the AI reach the systems it needs, with the right permissions?”
Integrations
“Where does the output go next, and what breaks if it arrives wrong?”
Security
“What data leaves your environment, and who can see it?”
Governance
“Who approves the use case, and which policy does it sit under?”
Human review
“Where does a person check the work, and do they have time to?”
Vendor dependencies
“What happens if the model, the price, or the vendor changes?”
Measurement and monitoring
“How will you know it is working, and when it stops?”
What you get
A decision,
not a dashboard.
Every audit ends in a report your leadership can act on. Open a tab to read a page.
Audit report · 02 Risk flags
Customer data leaves the approved tool list.
- Where: step 3, drafting
- Why it matters: no agreement covers that tool
- Owner: IT security
Audit report · 03 Workflow gaps
Exceptions are handled from memory.
- No written rule for refunds over the limit
- Two people route the same case differently
- Nobody sees the queue after hours
Audit report · 04 Data / integration
Two systems disagree on account status.
- CRM and billing are updated at different times
- The draft would quote whichever it reads first
- Fix: one source of truth for status
Audit report · 05 Governance gaps
No one owns model changes.
- Prompts are edited without review
- No log of what changed or why
- Name an owner and a change process
Audit report · 06 Next steps
Pilot one queue, with a person on every send.
- Start after fixes 1 and 2
- Measure time per reply and error rate
- Review the results together at four weeks
Audit report · 07 Priority fixes
Three fixes, in order.
- 1 · Assign an owner for exceptions
- 2 · Make billing the source for account status
- 3 · Move drafting into an approved tool
Audit report · 08 Go / no-go
Not yet. Go after fixes 1 and 2.
- The use case is worth doing
- The blockers are fixable in weeks, not months
- Building now would ship the gaps
Not all audits are the same
Which question
are you asking?
Independent by design
The audit is useful even if the answer is “do not build yet.”
We would rather find the blocker before implementation than discover it halfway through production.
Your next move
Find out what is ready, what is not, and what needs to change first.
Tell us about the initiative and where it stands. We’ll help you decide what to check before you scale it.
Tell us what’s slowing you down.
A few lines is enough. We’ll reply by email with what’s worth doing first.