Back to Portfolio Case Study

From status chasing to exception control.

The client: a $12M D2C specialty food brand. 28 employees, Shopify, an outsourced 3PL, a small customer-service team. The founder and the Head of Operations came in asking for automation. The audit found something else first.

The Ask

"We probably need automation or AI."

The full sentence was: "We probably need automation or AI because the team is spending too much time chasing status." Reasonable request. The team was drowning in status questions. But before recommending any tool, the audit traced one order through the workflow to see where the time actually went.

Current State

The workflow, as it actually ran

  1. A customer places an order in Shopify.
  2. The order passes to the 3PL.
  3. An inventory discrepancy appears.
  4. The 3PL notes the issue in its own portal. Nowhere else.
  5. Customer service never sees it.
  6. The customer emails asking for status.
  7. CS asks Ops in Slack.
  8. Ops manually checks Shopify, the 3PL portal, and an inventory spreadsheet.
  9. The founder gets tagged, because no one knows who owns the decision.
  10. The customer update goes out late.

EXHIBIT 01 · CURRENT-STATE WORKFLOW

Current-state swimlane exhibit: one inventory exception crosses six lanes before anyone owns it, with failure tags on each step FROM THE ACTUAL AUDIT EXHIBITOpen full exhibit →

One inventory exception crosses six lanes before anyone owns it. The full document adds the evidence reviewed, the ranked risk table, and the stabilization logic.

Failure Points

Five failure points, one pattern

Visibility gap

No unified exception list. The issue lived in one portal, invisible to everyone who needed it.

Manual work

Every status check meant a cross-system lookup: Shopify, the 3PL portal, a spreadsheet, Slack.

Ownership gap

No named owner. The question bounced between CS and Ops until it found the founder.

Reactive service

The customer found the issue before the team did. Every update was an apology.

Escalation failure

The founder was pulled into an order-level decision. That is not an escalation rule. That is the absence of one.

WHAT THAT COSTS, MEASURED

80-120exceptions a month needing manual review
15-25 minper status lookup, across three systems and a spreadsheet
20+ hrsof ops labor lost to status chasing, every month
0escalation rules. Founder attention substitutes for missing operating rules

Illustrative figures from the audit's evidence review: 42 delayed-order samples traced from order creation to customer response.

The Diagnosis

Not an order-status problem. An exception-control problem.

The client did not need to eliminate every exception. It needed to control exceptions before customers discovered them.

The first fix was not AI. It was an exception-control layer: one queue, one owner, one customer-safe status, one SLA clock, one escalation rule, one daily operating cadence.

Automation came after the workflow was stable, and only the safe pieces: owner routing, SLA aging alerts, a daily digest, customer-update prompts with human approval. Not AI-generated customer replies.

The scale of the leak, from the audit's evidence review: 80 to 120 exceptions a month, at 15 to 25 minutes of cross-system lookup each. That is 20+ hours of ops time every month spent finding out what is happening before anyone fixes anything. A full work week every two months, producing nothing.

EXHIBIT 02 · FUTURE-STATE WORKFLOW

Future-state swimlane exhibit: the exception becomes visible, owned, and time-bound before it reaches the customer FROM THE ACTUAL AUDIT EXHIBITOpen full exhibit →

The same exception, redesigned: visible, owned, and time-bound before it reaches the customer. The full document adds the owner rules, escalation thresholds, and automation readiness table.

The Roadmap

Targets set in the 30-day roadmap

The impact table from the audit report. Current state measured, future state targeted.

MetricNowTargetHow
Monthly exceptions needing manual review 80-120 80-120, controlled The goal is not zero exceptions. It is visible, owned, time-bound exceptions.
Lookup time per exception 15-25 min 3-5 min One shared queue with standard fields replaces the cross-system hunt.
Monthly labor lost to status chasing 20+ hours 5-8 hours Same exception volume, far fewer cross-system investigations.
Founder involvement in routine issues Frequent, ad hoc Down 60-80% Founder sees threshold hits and trend summaries, not queue noise.
Customer update timing Reactive, after the customer asks Same-day visibility CS sees a customer-safe status before the customer follows up.
  • Every exception carries a named owner.
  • CS sees a customer-safe status before sending any update.

EXHIBIT 03 · THE 30-DAY ROADMAP

30-day roadmap exhibit: four phases from hidden exceptions to a running operating cadence, automation in the final week FROM THE ACTUAL AUDIT EXHIBITOpen full exhibit →

Four weeks from hidden exceptions to a running cadence, with automation gated to the final week. The full document adds the phase plan by workstream, the automation gates, and the handoff.

After the Roadmap

What gets built, in operation

The roadmap's implementation options are five packages. Not five brochures: here is each one doing its job, in order. All of it is priced in the audit's implementation quote.

PACKAGE 01 · EXCEPTION TRACKER

Full spec →

The queue, on a normal Tuesday

Eight fields, owner rules, statuses, and resolution codes, installed in the tools the team already uses. Illustrative rows:

IDCategoryOwnerSLA ageCustomer-safe statusNext action
EX-0141Inventory mismatchOps Lead3h On track"Order confirmed, ship date unchanged."Confirm live count with 3PL
EX-0142AddressCS Lead2h On track"Awaiting your address confirmation."Email customer to confirm unit number
EX-0143DamageCS Lead1h On track"Replacement on the way."Send replacement, open 3PL claim
EX-01443PL delayOps Lead6h On track"Slight delay, new ship date shared."Get committed ship date from 3PL
EX-0138Partial shipmentOps Lead28h Aging"Remainder ships separately at no cost."Book remainder; first item in today's review
EX-01363PL delayOps LeadBlocked 5 bus. hrs VIP threshold"Your order has our full attention. Update by end of day."Escalated to founder per VIP rule; 3PL call booked

Four on track. One aging, worked first in the daily review. One VIP row, the only thing the founder sees today. The queue absorbs the volume so people work the exceptions that matter.

PACKAGE 02 · DASHBOARD AND CADENCE

Full spec →

Fifteen minutes a day against one screen

The dashboard reads from the queue. The cadence guarantees someone acts on it every business day. Illustrative snapshot, then the actual daily agenda:

34open exceptions
5aging past SLA
3blocked
4customer-impacting
  1. Triage new exceptions. Category, owner, next action. Nothing sits unowned. 3 min
  2. Review aging, oldest first. Owner confirms next action and due time on each item. 4 min
  3. Unblock or escalate blocked items. By threshold rule. No silent waiting. 3 min
  4. Check customer-impacting items. CS confirms each one has a current customer-safe status. 3 min
  5. Rebalance owner load. If one owner carries the aging risk, work moves. 2 min

PACKAGE 03 · SOP AND COMMUNICATION KIT

Full spec →

The words CS actually sends

Internal truth translates to customer-safe language. Both versions are true; the customer-safe one removes the noise the customer cannot act on.

3PL inventory mismatch, investigating→"We found a stock check issue during preparation and are confirming availability today."
Partial pick, one SKU short at 3PL, split shipment created→"Part of your order ships today. The remaining item ships separately."

PACKAGE 04 · SAFE AUTOMATION SPRINT

Full spec →

Automation with a human hand on every send

Starts only after the queue and cadence run stable for two consecutive weeks. Four automations ship; here is the one closest to the customer, spec'd the way all four are:

Customer-update prompts, human approval

TriggerA status change on a customer-impacting queue entry.
ActionA draft update from the matching Package 03 template is queued for CS review.
GuardrailNothing sends without a human click. Drafts come only from approved templates, never generated language.
Human roleCS reads the draft, edits or discards it, and sends only what they approve.
Failure modeThe wrong template is matched to the status change.
RollbackDrafts are disabled. CS reverts to the manual templates.

What stays human: AI-written customer replies (not first), predictive inventory (later, when the data has history), and every judgment call on refunds and exceptions (always).

PACKAGE 05 · FRACTIONAL OPS SUPPORT

Full spec →

The week-four readout

Systems decay without an owner of the system itself. A weekly readout keeps adoption honest and turns resolution data into prevention. An illustrative healthy week:

31open, vs 34 last week
3aging past SLA, all owned, all blocked on 3PL
2founder threshold hits all week
4 minstatus lookup spot check, inside the 3-5 minute target

Support ends when the team runs two stable months on its own. Graduation, not dependency.

Complete documents, for the deep read:  Current-state workflow · Future-state workflow · 30-day roadmap · Tracker · Dashboard · Comms kit · Automation sprint · Fractional ops

Your workflow has its own version of this.

Send 3 to 5 recent examples where work got stuck, escalated, or required manual status chasing to dan@diazovate.com. I'll tell you whether the audit is the right fit.

Book the Audit Starts at $2,500, scoped on the intro call.