Layer 9 — Solution and system design engineering

This is what the job actually produces. Hai's Solution Engineer JD defines the deliverable in one sentence; the 88 case studies show which parameters get sized and which outcomes get promised. Everything in layers 1–8 exists to serve this.

Layout, material flow and process design

What it is. Turning a customer's operation into a proposed physical and process design — the pre-sales engineering deliverable.

Why it matters here. It is the named output of Hai's Solutions function: the path from a customer's order file to a defensible layout is the whole job.

Fundamentals — a method, end to end:

1. Characterise the demand. Order lines per day, lines per order, units per line, peak-to-average ratio, cutoff times, seasonality. Ask for a year of order data; averages hide everything that matters.
2. Characterise the inventory. SKU count, dimension and weight distribution, velocity distribution (ABC), storage volume required, growth expectation.
3. Derive throughput requirements. Peak-hour lines/hour, not daily totals. Add the receiving and replenishment load — a system sized only for outbound will fail on a heavy inbound day.
4. Choose the handling method. Does the case envelope fit? Do cartons need decanting? Is there a long tail that justifies deeper storage?
5. Size the resources. Workstations first (they set the human rate), then robots to feed them, then chargers, then racking and floor area.
6. Lay out the flow. Receiving → storage → picking → consolidation → despatch, minimising cross-traffic and back-tracking.
7. Verify by simulation. See simulation verification.
8. Cost it and state the assumptions the design depends on, explicitly.

The bottleneck discipline. A GTP system has exactly one binding constraint at a time — usually workstation human rate, robot fleet capacity, or aisle congestion. Identify which, size around it with headroom, and be able to say what the second constraint is and when you would hit it.

How Hai applies it.

"Making automated proposals that may include layout design, data analysis, material flow design, and process design" — /join-us/recruitment/solution-engineer

"Work with the sales team to understand customer needs and design the right automation solution" — /join-us/recruitment/solution-engineer

"Computer skills: AutoCAD, Microsoft Office (especially Excel)." — /join-us/recruitment/solution-engineer

"Ability to analyze and understand customer business processes." — /join-us/recruitment/senior-system-designer-workstations

Take AutoCAD and Excel seriously. They are named as required skills. This is a role where the deliverable is a drawing and a model, not a repository — the tooling is the drafting stack, not a codebase.

Hai also offers a customer-facing Solution Generator and ROI Calculator, which makes the early-stage sizing conversation partly self-service.

Tradeoffs and pitfalls.

• Designing to the average and discovering the peak in production is the classic failure. Always ask for peak-day data.
• Omitting inbound/replenishment load from the sizing is the second classic.

Key terms. peak-to-average ratio · lines per hour · material flow · cross- traffic · bottleneck analysis · headroom · concept design · ROI.

Deployment sizing parameters

What it is. The specific parameters that get fixed in every project.

Why it matters here. Hai's own case studies size the same five variables every time. Learn to reason about them as a coupled set and you can talk about their projects fluently.

The five, from the corpus:

Fundamentals.

• They are coupled, and the coupling is directional. Workstations set the human pick rate → that sets required tote delivery rate → that sets robot count → that sets charger count. Size in that order; sizing robots first is backwards.
• 600×400 mm is the European standard tote footprint (a half-Euro-pallet module). Its dominance in the case data means most customers are on standard containers, which simplifies proposals.
• Shelving height trades density against cycle time. Beyond ~6 m you are into the telescopic models and vertical travel starts to dominate retrieval time.
• Sanity-check any proposal against real ratios. Locations per m² of storage area across the corpus runs 7.8 to 23.4, and rack height explains nearly all of the spread: Anta Chengdu 27,643 / 3,545 m² = 7.8; Sinocare 9.7; Geely 13.3; John Lewis (10 m) 22.0; TME (22 levels, 12.2 m clear) 56,166 / 2,400 m² = 23.4. Quote the range and the driver, never a single number.

How Hai applies it.

"Robot quantity & type: 53 units, A42C" — /cases/anta-group-jinjiang

"Robot quantity & type: 32 units A42 + 8 units A42T" — /cases/best-supply-chain-footwear-distribution-center

"Storage Locations within a 3,545m² Warehouse" — /cases/anta-group-chengdu

Outcome metrics Hai promises (301 datapoints across 80 case studies; most common labels: Picking Accuracy 42×, Storage Locations 41×, ROI 12×, Storage Density 10×):

"120,000+ storage locations" — /cases/anta-haipick-climb

"300% increase in storage density" — /cases/anta-haipick-climb

"2,400 totes/hour throughput efficiency" — /cases/anta-haipick-climb

Picking accuracy is quoted at 99.9%–99.99% across dozens of sites — the single most consistently promised number in the corpus. Know it.

Tradeoffs and pitfalls.

• Density percentages are always relative to the customer's prior state. "400% more storage" from a badly-run manual warehouse is not the same achievement as from an optimised one. Ask what the baseline was.
• ROI is frequently qualitative in the case data ("Fast") rather than a number.

Key terms. locations per m² · storage density uplift · picking accuracy · throughput (totes/h, cases/h) · Euro tote 600×400 · baseline.

Goods-to-person workstation design

What it is. The interface where the automated system meets the human.

Why it matters here. The workstation sets the system's rate. Everything upstream exists to keep it fed. It is also where ergonomics, safety and throughput collide, and Hai has a dedicated R&D role for exactly this.

Fundamentals.

• The core economics of GTP: in manual picking, travel is typically 50–70% of picker time. Eliminating travel is the whole value proposition.
• The human becomes the constraint. A picker sustains roughly 300–600 picks/hour depending on ergonomics, presentation quality and cognitive load — so the design goal is to remove every decision and every reach the operator does not need.
• Cognitive load is a throughput variable, not a comfort nicety. Pick-to-light, clear tote presentation and "no decision required" flows raise rate and lower error simultaneously.
• Buffering decouples the robot fleet from the human. Without a buffer, every robot delay is picker idle time. Buffer depth is a real design parameter.
• Ergonomics: presentation height, reach distance and repetition drive both injury risk and sustained rate. This is regulated in the EU.

How Hai applies it.

"HaiPick Systems bring containers directly to operators, cutting out travel time for order picking and speeding up fulfillment." — /solutions/system-features

"Handles up to 600 totes per hour, significantly boosting overall system throughput, up to 3× faster than manual picking." — /solutions/haistation

"Loads 8 cases in 3 seconds and unloads 8 cases in 5 seconds, reducing total processing time by up to 80%." — /solutions/haistation

"Operators simply follow system instructions, just pick and place without decision-making or walking" — /solutions/haistation

"Totes are positioned at a comfortable height for operators, reducing fatigue and improving picking accuracy." — /solutions/haistation

"Order totes are automatically switched as they are filled, with robots replacing full totes and positioning new ones without manual intervention. This eliminates secondary sorting and repacking steps." — /solutions/haistation

Hai's stated design philosophy is explicit: no walking, no decisions. The 600 totes/hour figure is the workstation ceiling to size against, and automatic tote switching is what removes the operator's idle gap.

Workstation types seen in the case data: on-conveyor, on-shelving, on-robot, and HaiPort loader/unloader units.

Tradeoffs and pitfalls.

• 600 totes/hour is a machine capability; sustained human rate over a shift is lower and is what you should size on.
• Multiple workstation types on one site add flexibility and operational complexity — training, procedures, spares.

Key terms. goods-to-person · pick rate · cognitive load · pick-to-light · buffer decoupling · ergonomics · HaiPort · secondary sortation.

Simulation-based design verification

What it is. Proving the design meets its numbers before anyone signs.

Why it matters here. It is the professional answer to the question that decides whether a proposal is trusted: "how do you know it will hit the number?"

Fundamentals.

• Verification hierarchy, weakest to strongest:
  1. Analytic model (spreadsheet, queueing formulas) — fast, ignores variance.
  2. Discrete-event simulation with real order data — captures variance and interaction effects.
  3. Physical pilot — captures reality, costs time and money. Use all three in sequence; each narrows what the next must explore.
• Simulate the peak hour with real order files. Averages hide the failure mode.
• Report distributions. "p95 order cycle time under 40 minutes at peak" is a commitment; "average 12 minutes" is a decoration.
• Sensitivity analysis is the mark of a good design. Which assumption, if wrong by 20%, breaks the design? Usually order profile or peak ratio. State it.
• Build in headroom — typically 15–20% above contracted peak — for growth, degradation and the assumptions you got wrong.

How Hai applies it.

"the project plan is swiftly verified, and strategies and on-site configurations are promptly adjusted" — /solutions/haiq-software

"By incorporating the actual customer scenario's demand parameters, the platform conducts a 1:1 simulation." — /solutions/haiq-software

"Engage in projects during the PoC phase." — /join-us/recruitment/senior-system-designer-workstations

"Evaluate commercial project customizations and requirements." — /join-us/recruitment/senior-system-designer-workstations

Note that simulation outputs feed on-site configuration adjustment — the model is used to tune the live system, not only to win the deal. That is a healthy sign and a good thing to ask about.

Tradeoffs and pitfalls.

• A simulation validated on one day's orders proves one day.
• Simulations routinely omit the things that actually break go-lives: operator learning curve, master-data errors, Wi-Fi dead spots, inbound shortfalls. Naming those omissions out loud is exactly what a senior person does.

Key terms. discrete-event simulation · sensitivity analysis · p95 · headroom · PoC · commissioning · order profile · design assumption register.