Lean Daily Management + AI: UNFI’s Playbook

UNFI reported a 9% throughput increase, a 3% lift in on-time delivery, and a 16% shrink reduction after rolling out a standardized lean operating model across 34 of its 49 distribution centers in roughly a year. Those aren’t “nice-to-have” gains. They’re the kind of improvements that show up in customer conversations, in DC morale, and in operating expense.

Here’s what I like about this story for our AI in Supply Chain & Procurement series: UNFI didn’t start with a shiny technology promise. They started with operational truth. A disciplined cadence of reviewing the right metrics every day—SQDC: Safety, Quality, Delivery, Cost—and building site-level buy-in.

And that’s exactly why AI belongs in the conversation. AI doesn’t replace lean. It scales it. If lean daily management is the habit, AI can be the engine that keeps the habit consistent, fast, and measurable across dozens of sites—especially during peak season planning, labor volatility, and supplier disruption.

What UNFI’s results tell us about lean done right

Lean daily management works when it’s standardized enough to compare, but local enough to be owned.

UNFI’s leadership explicitly warned against “fake lean”—the kind of rollout where corporate sets a deadline, sites comply on paper, and the floor quietly ignores it. I agree with that stance. In distribution, the fastest way to poison a continuous improvement program is to make it feel like a compliance exercise.

UNFI’s approach—daily review of SQDC metrics, a standardized operating model, and a focus on the working environment—signals three practical truths:

• The best operational metrics are leading indicators, not just end-of-month scorecards.
• Consistency beats intensity. Daily cadence drives compounding gains.
• Adoption is cultural work. You can’t spreadsheet your way into buy-in.

The results they shared map cleanly to lean fundamentals:

• Throughput +9%: fewer bottlenecks, better flow, tighter process control.
• On-time delivery +3%: more stable execution, fewer last-minute firefights.
• Shrink −16%: better quality, handling, inventory discipline, and root-cause follow-through.

Now the interesting part: once you’ve got the lean cadence, AI can remove friction and widen the impact.

Where AI fits: turning SQDC into real-time operations

Answer first: AI strengthens lean daily management by making SQDC metrics more predictive, more automated, and more actionable.

Most DCs already track safety incidents, pick rates, on-time shipments, and labor cost. The issue is that teams often spend their day arguing about whose report is “right,” or reacting after the fact.

AI helps in three ways.

1) From lagging metrics to early warnings

If SQDC is reviewed daily, the next step is spotting patterns before performance drops.

Examples of AI-driven early warnings that align directly to SQDC:

• Safety: detect rising near-miss risk by correlating congestion, overtime hours, equipment downtime, and task switching.
• Quality: predict damage risk by SKU attributes, handling path, dock-to-stock time, and temperature exposure windows.
• Delivery: identify which outbound waves are likely to miss cutoffs based on labor staffing, pick density, and replenishment gaps.
• Cost: forecast labor variance by function (receiving, replenishment, picking, loading) and recommend rebalancing.

Lean says “go see the work.” AI says “here’s where to look first.”

2) Faster root cause, fewer meetings

Lean daily management lives or dies on problem-solving. The catch is that root-cause work gets slow when data is scattered.

A practical pattern I’ve found works: pair lean’s structured thinking (like 5 Whys) with AI-assisted investigation.

• AI clusters exceptions (late orders, shorts, mispicks) into likely drivers.
• Supervisors validate on the floor.
• Teams implement countermeasures.
• AI monitors whether the countermeasure actually moved the metric.

That last step is underrated. Many operations “fix” problems and never verify durability.

3) Standardization without suffocating the site

UNFI emphasized site buy-in. Good. AI can support buy-in by making standards feel useful, not bureaucratic.

For multi-DC networks, AI can create a consistent “operating picture” while still allowing local decision-making:

• A shared SQDC dashboard definition across sites
• Local thresholds tuned to site realities (SKU mix, automation level, labor market)
• Cross-site benchmarking that highlights transferable practices

If you want adoption, don’t force sameness. Force clarity.

Procurement’s role: lean operations start upstream

Answer first: procurement performance improves when lean daily management is paired with AI that reduces variability in supply, specs, and inbound flow.

A lot of companies treat lean as a warehouse thing. That’s a mistake. Some of the most expensive “waste” in distribution is purchased variability:

• inconsistent case pack or labeling
• supplier ASN inaccuracies
• late inbound arrivals that create labor spikes
• quality escapes that become rework, returns, or shrink

Here’s how procurement teams can connect to SQDC in a way that actually helps the DC.

Make supplier performance measurable in operational terms

Instead of only tracking OTIF at a high level, connect supplier behavior to DC pain:

• Receiving cycle time variance by supplier
• Shorts and damages by lane and packaging spec
• Appointment adherence and dwell time
• Claim rate and dispute cycle time

Then use AI to flag which suppliers are driving the most operational cost—not just the most obvious late shipments.

Use AI to prioritize supplier development (not just penalties)

Penalties can work, but they don’t build resilient supply.

AI helps you answer: “If I can only fix five suppliers this quarter, which five will reduce cost and improve service the most?”

A simple prioritization model can combine:

• volume impact
• defect/shrink contribution
• inbound variability
• critical SKU coverage
• recovery difficulty (how fast you can qualify alternates)

That’s lean thinking applied to procurement: focus effort where it returns the most.

A practical roadmap: blending lean daily management and AI

Answer first: start with a stable lean cadence, then add AI where it removes the most manual effort and improves decision speed.

If you’re trying to replicate UNFI-style gains (or exceed them), don’t start by buying software and hoping it changes behavior. Start by deciding how work will run every day.

Step 1: Lock the SQDC “truth” (2–4 weeks)

Define:

• the 8–15 metrics your teams will review daily
• standard definitions and data sources
• who owns each metric and what actions are allowed

If metrics aren’t trusted, lean becomes debate club.

Step 2: Build the daily management rhythm (4–8 weeks)

Operationalize:

• tiered daily huddles (frontline → area → site)
• visible problem logs and escalation rules
• closed-loop follow-up (countermeasure, owner, due date)

This is where most companies get stuck because it feels “too basic.” Basic is the point.

Step 3: Add AI in the “pain seams” (8–16 weeks)

Pick one or two use cases tied to SQDC outcomes:

• delivery risk prediction for wave planning
• shrink/damage prediction and handling interventions
• labor planning optimization tied to inbound/outbound variability
• automated exception triage for shorts/mispicks

Success criteria should be operational, not technical: fewer exceptions, less overtime, improved service, lower shrink.

Step 4: Scale across sites without creating “fake lean”

UNFI’s warning is the lesson: avoid scaling pressure that produces theater.

What works instead:

• scale the system (definitions, governance, training)
• scale the capability (coaching, site champions)
• scale the tech after the process is stable

If a site isn’t ready, forcing the rollout makes the whole network weaker.

People also ask: common concerns before you blend AI and lean

“Will AI make frontline teams feel monitored?”

It can—if you position it as surveillance. It shouldn’t be. The healthiest framing is: AI reduces guesswork and rework. Put it on the side of the operator.

“Do we need perfect data before using AI?”

No. You need reliable definitions and a plan to improve data quality as you go. Waiting for perfect data usually means waiting forever.

“What’s the fastest place to start?”

Start where daily management already feels painful:

• too many exceptions to triage
• too much manual reporting
• recurring late waves
• chronic shrink categories

AI should remove friction from the daily cadence, not add a new program.

The stance I’m taking: lean is the prerequisite, AI is the multiplier

UNFI’s improvements are a reminder that operational discipline still wins. A 9% throughput lift and 16% shrink reduction don’t come from slide decks. They come from daily management that people actually follow.

But it would be a missed opportunity to stop there. When you pair lean daily management with AI—especially for real-time analytics, exception prediction, and supplier variability control—you get a system that’s harder to break during peak season and easier to improve week over week.

If you’re leading supply chain or procurement in 2026 planning cycles, here’s a useful question to take into your next ops review: Which SQDC metric do we still manage by hindsight—and what would it take to manage it by prediction instead?