Outgrace, home
OperationsFramework

A framework for ranking warehouse sites by cost-to-serve

Transport, lead time and handling cost, scored together for each candidate site. The public freight data will not supply the numbers, so most of the work falls to you.

A warehouse chosen on rent and land price often costs more to run than a dearer building in a better place. The difference shows up in trucks, in days of waiting and in labour spent moving stock twice. This framework asks you to score each candidate on three costs together, so that the cheapest building on paper does not win by default. One warning first. The public sources we reviewed do not publish the lane-level costs, lead times or handling rates you will need. You will have to build those yourself, and the rest of this piece explains how.

Build one sheet with a row for each candidate site and a column for each cost. Convert every column into the same currency and the same unit of time. Multiply each column by its agreed weight, add the results and sort. The sheet matters less than the discipline around it. Every cell needs a named source, and an empty cell should show as a gap, never as zero. A zero that nobody measured is the most common way a site wins by mistake.

Chapter one: Pillar one: transport cost per lane

Start with the lanes, not the map. Pull twelve months of shipments from your system and group them by origin, destination and mode. For each candidate site, estimate what a truck or parcel would cost on each lane, using your own carrier rates rather than a published average. Weight the lanes by volume, so that a rarely used route does not dominate the total. A site moved closer to your largest customers may cut the cost of its main lanes while lengthening a smaller group of lanes. The net figure decides, not the headline distance.

Test the ranking against two periods. Run the sheet on the lane mix from the last twelve months, then again on your forecast for the next three years. A lease runs for years, and a lane mix can shift within a single season. A site that ranks first on both views is safer than one that wins on a single snapshot. If the two rankings disagree, the gap between them is the number to take to the board.

For background on the wider freight system, the Bureau of Transportation Statistics publishes its Transportation Statistics Annual Report, an overview covering freight, the economy, reliability, safety, energy and environment. It is useful reading for a planner. It does not give a cost per lane, so treat it as context rather than as an input to the sheet.

Chapter two: Pillar two: lead time to the customer

Lead time is the time from order to delivery at the customer's door. It is a cost even when no invoice shows it. Measure it for each candidate by modelling the slowest realistic route, including dispatch cut-off times and the number of carrier transfers. A site that saves money per mile but adds a day to a fifth of your orders has taken something from your customers. Put a number of days on each customer segment, and agree a value for each day lost, before you rank.

Be wary of averages. An average lead time hides the orders that arrive a week late. For each segment, report the share of orders that miss the target, and rank on that share as well as on the mean. Two sites with the same average can differ sharply in how often they let a key account down, and the second measure is usually the one a sales director remembers.

Border exposure belongs in this pillar. The Bureau of Transportation Statistics publishes a border commodity report giving the dollar value of goods crossing U.S. land borders with Canada and Mexico, broken out by transport mode. If a meaningful share of your freight crosses a land border, a site near that crossing can change lead times more than its distance to the customer does. The report gives values rather than crossing times, so the delay figures must come from your carriers. Model the delay at the crossing as a separate line in the sheet.

Chapter three: Pillar three: handling cost inside the building

Handling cost is what it takes to receive, put away, pick, pack and dispatch once goods are inside. Two sites with identical freight costs can differ widely here, because of dock layout, yard space, the local labour pool and the number of times each pallet is touched. Ask each shortlisted operator or landlord for its handling rate per unit and its labour hours per thousand picks. Test both against your own figures from last year. Where a candidate cannot supply them, price the gap as a risk, not as zero.

Labour is local, and so is the risk. Ask how many shifts the candidate can staff from its catchment, and how long vacancies at its current operators stay open. A building with a low rent and a thin labour market may cost more in overtime and agency fees than the rent saves. Put those hours into the sheet at the rate you actually pay, not at the rate in the operator's proposal.

Automation is changing this pillar. A study by MIT's Center for Transportation and Logistics and Mecalux, according to its headline alone, finds that 57% of organisations expect AI to make over a quarter of warehouse operational decisions autonomously by 2028. We have seen only the headline, so we cannot say how that expectation was measured. It does suggest that a handling model built today may need to account for software that decides routing and slotting without a supervisor. Ask each operator which decisions its systems take alone, and what happens when they fail.

Chapter four: Pillar four: resilience and flexibility

A site that ranks first on today's data can rank last after one bad year. Score each candidate on how many alternative routes it has to the same customers, how many carriers serve it and whether it can absorb a surge in volume without a new lease. MIT's Center for Transportation and Logistics describes its work as building supply networks that are resilient, adaptive and future-ready, and says it plans for crisis response across public, private and nonprofit organisations. Its published descriptions do not give a scoring method, so the test has to be your own.

Name the single route, carrier or crossing whose loss would cost you most. Then see which candidate site survives its loss, and at what extra cost. A site with two carriers and a second road out is worth more than its rent suggests, and a site served by one carrier at one crossing is worth less than it looks on a map. Count the carriers that will quote for the site, not just the ones your network uses today.

Fix the weight of this pillar before you see any scores. Otherwise the team that wants a particular building will find reasons to make flexibility matter less once it has fallen behind. A sensible starting point is to weight it no lower than handling cost, and to revisit the weights once a year, in writing, with the chair present.

Before you rank a site, can you answer yes to these?

If any answer is no, the site's score is incomplete and should be shown as such.

  • Do you have twelve months of shipments grouped by origin, destination and mode?
  • Are lane costs taken from your own carrier rates rather than a published average?
  • Have you assigned a number of days to each customer segment's lead time?
  • Have you measured how much of your freight crosses a land border, and modelled that delay
  • Has each candidate given a handling rate per unit and labour hours per thousand picks, in
  • Have you named the single route, carrier or crossing whose loss would cost you most?
  • Are the pillar weights written down before any site is scored?
  • Does every cell in the sheet cite a source, with any gap shown as a gap?

Outgrace checklist

Where it breaks

The framework fails in three places. The first is when the inputs are guesses. The public data does not carry the lane costs, lead times or handling rates, and a ranking built on invented figures looks as tidy as one built on measured ones. The second is when the weights are chosen after the scores. The third is when the people who will run the building are left out of the model. Ranking is a finance and operations exercise. Give the lane data to finance, the handling data to operations, and let the chair review the weights rather than the scores.

The main public source on how goods move is the Census Bureau's Commodity Flow Survey. Its most recent data we reviewed covers 2022, released on June 26, 2025, with the publication following on January 20, 2026. The bureau says the same data are used to gauge demand for transport facilities and to assess energy use, safety risk and environmental concerns. It describes flows across the economy, not the cost of your lanes, so use it to understand where freight travels, not what it costs to send it.

The CFS data are used by policy makers and transportation planners in various federal, state, and local agencies

U.S. Census Bureau, Commodity Flow Survey page, U.S. Census Bureau

The limits of the research matter too. We reviewed the National Bureau of Economic Research's working papers, which cover topics such as health spending and entrepreneurship. None bears on warehouse location or freight cost, so the framework draws on no economic research from that source. Nor did the sources we reviewed report a network-design methodology or ranking criteria. The four pillars above are editorial judgement, offered as a structure, and each weight and threshold must be set by your own business.

On Monday, do three things. Pull your shipment history and split it by lane. Ask each of your two shortlisted operators for handling rates and lead-time evidence in writing. Then agree the four pillar weights with your finance lead and operations head before anyone sees a site score. A ranking built that way can be defended in the board pack. One built any other way will be argued about rather than used.

Newsletter

The Briefing

One email a week: the leaders we profiled, the ideas worth stealing, and nothing else.

Weekly, free, one click to leave. Read our privacy policy.