Approach
audit before automate
The method, in five steps
The order is the argument. Steps three and five are the ones most engagements skip.
AI made building cheap, so the value is not in building. It is in judgment: picking the right problem, scoping it to a measurable outcome, shipping it to production, and driving adoption. Most initiatives fail by going straight to building on top of a workflow that was already broken.
Real data, real systems, the workflow as it runs - not as documented.
A ranked map, including the explicit finding that a case is not worth doing.
Fix the broken process first. Automating it as-is only makes it faster.
In phases, into your environment, instrumented against a baseline.
Against the baseline we established, and documented for your team.
The audit that did not buy a camera
Told as method, not as a sector story. The point is the sequence, and it transfers.
On a bottling operation with three high-speed lines, end-of-line quality relied on manual spot checks above twenty thousand units an hour. Every vendor in the room wanted to sell a vision suite that caught everything. We ran a defect-data audit first.
It found that three specific failure modes - underfill, cocked caps, skewed labels - accounted for roughly 90% of the real cost, against twelve listed in the quality manual. So we deployed on the highest-throughput line first, targeting only those three, and flagged rather than auto-rejected while operators validated every flag for weeks before automated diversion was enabled.
Defect escape to distribution fell 84%, detection went from hours to under two minutes, and quality-related customer returns fell 61% over the following quarter. None of that came from a better camera. It came from having counted first.
Every vendor wanted to sell us a suite catching everything - but three defects were 90% of our real cost, and solving those first is what made the business case obvious.Plant Quality Manager
The opportunity map that came out of it: impact against effort, with data readiness in the marker
- Data is ready
- Data needs work first
- Data does not exist yet
- 1
- Underfill, caps, labels - the three that mattered
- 2
- Full twelve-mode inspection suite
- 3
- Downstream complaint linkage
- 4
- Shift-level defect reporting
- 5
- Predictive line maintenance
A diagnostic, not a questionnaire
The distinction this page exists to draw, and the one a buyer is entitled to press on.
Assessments in this market range from a five-minute self-serve form to a half-day workshop, and the published floor is about 270 dollars for five business days. The answer to that is not indignation. It is a concrete list of what gets counted.
Five business days and a template does not buy anyone counting your invoice volumes.
What a diagnostic examines
- Interviews with the people who run the process, not only the people who own it
- Real data samples, pulled from the live system rather than described
- The workflow as it actually runs, including the steps nobody documented
- Document volumes counted, per type, per month
- A baseline established where nothing is currently measured
- An explicit benchmark disclosure: where no real external benchmark exists, we say so
What a questionnaire examines
- What the respondent believes the process is
- A self-scored maturity level with no evidence behind any score
- Volumes as remembered, which are reliably wrong in both directions
- A score presented as calibrated when it is self-referential
- A recommendation that happens to be the assessor's product
What runs around the work
Thirteen phases. The shaded nine are what a single build-me-a-thing transaction does not have.
Three phases before any contract exists, a formal change process for the reality that real findings require a scope conversation, and a close-out that most engagements skip. Each of them is a point where your input is captured formally and the work can still change direction cheaply.
All thirteen phases, across the whole relationship from the first diagnostic to ongoing oversight - not one tier.
- A structured engagement: all 13
- A one-off build: the 9 shaded rows are missing
- 1 First Contact 1-3 days Engagement only
- 2 Discovery Call 1 session Engagement only
- 3 Recap & Internal Analysis Same day - 3 days Engagement only
- 4 Proposal 2-5 days to prepare Engagement and one-off build
- 5 Proposal Review & Negotiation 3-10 days Engagement and one-off build
- 6 Contracting 1-5 days Engagement only
- 7 Onboarding & Kickoff First 5-7 days Engagement only
- 8 Delivery Phase(s) Weeks to months, phased Engagement and one-off build
- 9 Milestone Reviews Per phase Engagement only
- 10 Change Management (as needed) As triggered Engagement only
- 11 Final Delivery & Acceptance 1 session Engagement and one-off build
- 12 Close-Out 1 session Engagement only
- 13 Post-Engagement Ongoing Engagement only
Start here
The method is checkable. Ask us what we would count in your first two weeks.
It is a specific answer, not a brochure: the interviews, the systems, the document types and the volumes, named for your operation.
You get a reply within one working day, from the engineer who would do the work - not a sales sequence.