One site is evidence about one site

A pilot measures value at one site under conditions that will not repeat. The vendor's engineers are on site. A motivated sponsor clears obstacles. The line chosen for the pilot is often the one most likely to succeed. The result is real, and it describes that site.

A network rollout asks a different question: will the value hold at the other sites, which differ in equipment age, control systems, network readiness, maintenance capability, labor model, product mix and who holds the budget?

The finance team approving the rollout needs an answer for each site, not an average taken from the best one.

This paper sets out how to build that answer. It covers which conditions to test at each site, how to separate a benefit that travels from one that belongs to the pilot site, who owns the budget and the running cost after rollout, how to check operational technology (OT) readiness site by site, who maintains the system once the vendor leaves, how to choose the order of sites, and what evidence supports a phased approval instead of a single network commitment.

The appendix is a one-page site sheet a plant team can fill in before the rollout decision. Why pilots stall between the first site and the second is covered in Too successful to stop, too unproven to scale. This paper is about building the case that gets past that point.

Most rollout cases start from the pilot result and multiply it by the number of sites. That arithmetic assumes the sites are alike. In an industrial network they rarely are.

Three things make the pilot site unusual.

Attention. During a pilot, the vendor's engineers, the sponsor and often a central digital team watch the system every day. Problems are fixed in hours. At the tenth site, the same problem waits for the next maintenance window.

Selection. Pilot sites are chosen because they are likely to succeed: newer equipment, a supportive plant manager, clean data. This is sensible for a pilot. It also means the pilot site is the least representative site in the network.

Measurement. The pilot baseline was usually measured once, by the team that wanted the pilot to work. Other sites may measure the same indicator differently. Overall equipment effectiveness (OEE) is a good example. ISO 22400-2 defines it as a composite of availability, effectiveness and quality, with specific definitions for each element [1], and plants that use the same name often calculate it in different ways. A 5-point gain at the pilot site can disappear when the second site calculates its baseline another way.

None of this means the pilot was wrong. It means the pilot answered "can this work here?" and the rollout needs "will this work there, at what cost, and who will pay for it?"

The scale of the gap is visible across the industry. In Deloitte's 2025 smart manufacturing survey of 600 executives at large manufacturers with US headquarters or operations, 92% said smart manufacturing would be their main driver of competitiveness over the next three years, and 29% were using AI or machine learning at the facility or network level [2]. The World Economic Forum's Global Lighthouse Network, which recognizes advanced manufacturing sites and value chains, had 238 sites as of June 2026 [3]. The recognition is given site by site.

A rollout case holds when it has been tested against the conditions that vary across the network. Six matter in most industrial rollouts.

Equipment and control systems. The age and make of the machines, the programmable logic controllers (PLCs), and the control architecture decide how much integration work each site needs. A system that connected to modern controllers at the pilot site may need gateways, custom drivers or manual data collection at an older one.

Data and network readiness. Some sites have a plant network that can carry the data the system needs. Others have isolated cells, limited bandwidth or no connection between the plant floor and the site's IT systems. The cost of closing that gap belongs in the case, per site.

Maintenance capability. The system has to be maintained by the people at each site after the vendor leaves. That depends on the skills and headcount of the site's maintenance team, which vary widely and are under pressure across the industry. Deloitte and The Manufacturing Institute estimate that US manufacturing may need as many as 3.8 million additional employees between 2024 and 2033, and that as many as 1.9 million of those jobs could go unfilled [4].

Labor model and shift patterns. A system that helps a day-shift operator at the pilot site may meet a different crew structure, union agreement or language at another site. Adoption depends on the people who use it every shift.

Product mix and process. A benefit measured on a high-volume line may shrink on a site that runs many short batches, or grow on one with more changeovers. The case should say which process characteristics the benefit depends on.

Local budget and authority. At some companies the rollout is funded centrally. At others, each site pays from its own budget, and each plant manager can decline. Where site leaders hold the budget, the case has to persuade each of them, not only the central sponsor.

For each benefit in the pilot result, ask one question: what at the pilot site made this benefit possible, and is that thing present at the next site?

The answers usually fall into three groups.

Benefits that travel. The mechanism does not depend on site conditions. For example, a quality inspection model that detects a defect type common to all sites running the same product. These can be multiplied across sites with a modest discount.

Benefits that depend on a condition. The mechanism works where a specific condition is met: modern controllers, a certain product mix, a maintenance team with the skills to act on alerts. These hold at sites that meet the condition and not elsewhere. The case needs to count only the sites that meet it, or include the cost of creating the condition.

Benefits that belong to the pilot. The gain came from the attention, the extra staff or the selection of the site. These do not travel, and they should be removed from the network case before it reaches finance.

Doing this sorting honestly often reduces the headline number. It also produces a case that survives the first site after the pilot, which is where many rollouts lose the finance team's confidence.

A rollout changes the budget question twice.

Capital or operating spend. Hardware and integration are usually capital spend (capex). Software subscriptions, cloud services and support are operating spend (opex). Many industrial companies approve these through different processes and different people. A pilot funded from an innovation budget avoided this question. A rollout cannot. The case should state, for each cost line, which budget pays and who approves it, and whether each plant controller will carry a subscription in the operating budget. The committee nobody mapped covers the approval path in more detail.

The two budgets are judged differently, so the case needs both views. A capital request is judged on payback, return and total cost of ownership over the life of the asset. One accounting firm's automation guide puts typical payback at 12 to 18 months without naming its data [7]. The hurdle that counts is the buyer's own. A subscription is judged on monthly cost against monthly savings, and it is judged again at every renewal by whoever holds the operating budget that year. A rollout that mixes the two (hardware bought once, software and support paid every year) has to clear both tests, often with different people.

The accounting also pulls toward capital spend. Equipment bought outright is depreciated, and depreciation sits below EBITDA; a subscription is an operating cost that lowers EBITDA every year it runs. In the US, the One Big Beautiful Bill Act of July 2025 permanently restored 100% bonus depreciation for qualifying equipment acquired after 19 January 2025, so a profitable buyer can deduct a capital purchase in its first year [10]. A finance team can prefer the capital route even where the subscription costs less in cash. A subscription that replaces a cost the plant already carries, such as labor, maintenance, spare parts or downtime, avoids that argument.

Timing against the capital cycle. In ACEEE's interviews with industrial firms about energy-related capital projects, ideas move through tiered approval in which authority rises up the chain of command, and at one food and beverage company a plant's projects go every month to a global capital board that divides the funds [8]. ACEEE also notes that capital budgets are discussed on an annual basis, though firms often plan within multiyear cycles [8]. A rollout request that misses that plan waits for the next one, or competes for whatever is left. The case should name the capital cycle each site's spending falls into and the date the request has to be in.

Central or site. If the central team funds the rollout and each site pays the running cost afterward, each plant manager inherits a recurring cost for a system they did not choose. That is a common reason for quiet non-adoption. The case should show the running cost per site and confirm that each site's leadership has agreed to carry it, or that the central budget will.

The finance team will also ask when the benefit arrives compared with when the cost does. Integration and training costs come first, and the benefit builds over months. A case that shows cost and benefit by quarter, per site, is easier to approve than one that shows only a payback period for the whole network.

This matters most for AI. Practitioner analysis of AI payback in manufacturing suggests that a model built for one asset at one site often does not pay back on savings alone once the engineering and infrastructure cost is counted, which we treat as an assumption to test plant by plant; it pays back when the same model runs across several lines or sites, and until then it is justified by the risk it reduces [9]. That is the multi-site case in one sentence. The plant controller at the pilot site may be right that the single-site numbers do not work, and the network case may still hold. The case should show the payback per site and for the network, and say which costs are shared across sites.

Connecting production systems at many sites raises security questions that did not come up at a single pilot, especially where the pilot ran on an isolated network.

NIST Special Publication 800-82 Revision 3, the US guide to OT security, covers the controls expected when production systems connect to other networks, including segmentation, access control and a cautious approach to patching, which it recommends testing before installing on OT systems [5]. The ISA/IEC 62443 series sets out security requirements for industrial automation and control systems, and many industrial buyers use it as a reference [6].

For the rollout case, the practical questions are simple to state and slow to answer:

• Does each site's network architecture allow the connection the system needs without weakening segmentation?
• Who at each site owns OT security, and have they reviewed the design?
• What does the corporate IT security team require, and has it approved the pattern for all sites or only for the pilot?
• How are updates to the system deployed and tested at each site?

A security review that approves the pilot does not approve the network. A late security objection is one of the most common reasons a rollout stops at the second or third site, and it is avoidable if the review happens before the capital request.

At the pilot site, the vendor or the central team often maintained the system. At scale, that model does not hold. Someone at each site has to respond when the system misbehaves, retrain models when conditions change, and decide when to override it.

The case should name, for each site:

• who responds to a failure, and within what time;
• whether that person has the skills today, or needs training, and how long that takes;
• how the maintenance work fits into the site's computerized maintenance management system (CMMS) and existing routines;
• what the vendor, or the distributor or integrator that sold it, provides after go-live in this region, and at what cost.

If the answer at a site is "the central team will support it remotely," the case should include the size of that team at full rollout. A support model that works for three sites often does not work for thirty.

The order of sites decides how fast the rollout learns and how soon a problem becomes visible. Two common approaches both have weaknesses.

Starting with the next-easiest sites builds momentum and delays the hard questions. The rollout looks healthy until it reaches the older plants, by which time most of the capital is committed.

Starting with the hardest sites tests the case early and risks a visible failure before the organization has any confidence in the system.

A better approach is to choose the second wave for what it teaches. Include at least one site that differs from the pilot on each condition that matters: an older control system, a different maintenance model, a site that pays from its own budget. If the case holds at those sites, the rest of the network is lower risk. If it fails at one of them, the failure shows exactly which condition the case depends on, and the finance team can decide whether to create that condition elsewhere or limit the rollout to the sites that already have it.

A single network-wide approval asks the finance team to accept the pilot result as evidence for every site. A phased approval asks for less and is easier to give.

A phased request usually has three parts:

1. The next wave, fully costed. The chosen sites, their integration and running costs, the benefits expected at each, and the owner at each site.
2. The conditions for the following wave. The results from the next wave that would justify continuing, written as numbers before the wave starts.
3. The decision date. When the results will be reviewed, and by whom.

This structure also protects the sponsor. If the second wave shows that a benefit depends on a condition only some sites meet, the rollout can narrow to those sites without looking like a failure.

For a vendor, the buyers are customers. For an internal rollout, the buyers are inside the company: plant managers who hold site budgets, the VP of operations who owns the rollout, maintenance and reliability managers whose teams will run the system, the machine operators and technicians who will use it, OT security owners, and the finance partners who will approve the spending, starting with each plant controller who checks the value case.

The same discipline applies. Each of these people is interviewed about the conditions at their site, using a consistent set of questions, and each interview is scored on the same dimensions: urgency, priority, budget, authority and openness. Each claim in the rollout case then receives a grade:

• Verified where the plant manager and plant controller at that site have confirmed it,
• Benchmarked where it rests on the pilot result, benchmarks or the central team's view,
• Assumption where it is still assumed.

The grades show the finance team which parts of the case are evidenced at which sites. The evidence standard is described on the method page, and what we test for operators lists the four tests most rollout cases need.

The interviews are more useful when they are not run by the team that sponsored the pilot. Site leaders are more open with someone who does not need a particular answer.

Complete one sheet per site before the rollout decision. Where an answer is unknown, write "unknown." That is itself a finding.

Site

• Site name, products, shifts, headcount.
• Who holds the budget for this rollout at this site? Have they agreed to fund the running cost?
• Which cost lines are capital (capex) and which are operating (opex) here, and which capital cycle does the request fall into?
• What payback does the plant controller need to sign the capital request, and will the site carry the subscription at renewal?

Equipment and control systems

• Control system makes and approximate age on the lines in scope.
• Integration work needed here that was not needed at the pilot site.

Data and network

• Can the plant network carry the data required? What must change?
• Is the baseline for each benefit measured the same way as at the pilot site?

Maintenance

• Who will respond to a failure, and within what time?
• Skills gap, if any, and time to close it.
• How the system fits into the site's CMMS and routines.
• Which integrator or distributor covers this site, and is it the one that installed the existing systems?

OT security

• Owner of OT security at the site. Has the design been reviewed?
• Any site-specific constraints from corporate IT or regulation.

People

• Shift patterns, labor agreements, languages.
• Who will use the system every shift, and have they seen it?

Benefits

• For each pilot benefit: travels, depends on a condition (which one, and is it met here), or belongs to the pilot.
• Expected benefit here, by quarter.

Grade

• For each claim above: Verified, Benchmarked or Assumption, with the source.

1. International Organization for Standardization, "ISO 22400-2:2014 Automation systems and integration: Key performance indicators (KPIs) for manufacturing operations management, Part 2: Definitions and descriptions." https://www.iso.org/standard/54497.html
2. Deloitte, "2025 Smart Manufacturing and Operations Survey," 600 executives at manufacturers with US headquarters or operations (revenue of $500 million or more, 1,000 or more employees) surveyed August to September 2024, press release via PR Newswire, 1 May 2025. https://www.prnewswire.com/news-releases/deloitte-survey-reveals-smart-manufacturing-is-driving-advantage-but-needs-focused-investment-and-implementation-302443462.html
3. World Economic Forum, "New Global Lighthouse Sites Demonstrate How AI Is Rewiring Manufacturing and Supply Chains," press release, 22 June 2026. https://www.weforum.org/press/2026/06/new-global-lighthouse-sites-demonstrate-how-ai-is-rewiring-manufacturing-and-supply-chains/
4. The Manufacturing Institute, "Manufacturers Need as Many as 3.8 Million New Employees by 2033," on the 2024 Deloitte and The Manufacturing Institute talent study, 3 April 2024. https://themanufacturinginstitute.org/manufacturers-need-as-many-as-3-8-million-new-employees-by-2033/
5. Keith Stouffer et al., "Guide to Operational Technology (OT) Security," NIST Special Publication 800-82 Revision 3, National Institute of Standards and Technology, September 2023. https://csrc.nist.gov/pubs/sp/800/82/r3/final
6. International Society of Automation, "ISA/IEC 62443 Series of Standards." https://www.isa.org/standards-and-publications/isa-standards/isa-iec-62443-series-of-standards
7. Laura Zindel, Wiss & Company, "Manufacturing Automation ROI: Financial Justification Guide," 1 May 2026. https://wiss.com/manufacturing-automation-roi-financial-justification/
8. Grace Lewallen, Reuven Sussman and Pavitra Srinivasan, "Industrial Capital Investment Decisions: Pathways to Energy Program Engagement," American Council for an Energy-Efficient Economy, research report, April 2026. https://www.aceee.org/sites/default/files/pdfs/i2601.pdf
9. The Neural Base, "Payback Periods," AI for Manufacturing. https://theneuralbase.com/ai-for-manufacturing/learn/advanced/payback-periods/
10. BDO, "One Big Beautiful Bill Act Expands 100% Depreciation Expensing Opportunities." https://www.bdo.com/insights/tax/one-big-beautiful-bill-act-expands-100-depreciation-expensing-opportunities

What is not yet sourced

• The 12 to 18 month typical payback (reference 7) is one accounting firm's practitioner guidance, which cites published deployment data from 2025 to 2026 without naming it. Each company's own hurdle rate is the evidence that counts.
• The claim that a single-site AI model often does not pay back alone (reference 9) is a practitioner analysis, not a measured study.
• The capital-approval process (reference 8) comes from ACEEE's interviews about energy-related projects at medium and large US industrial firms, not from a survey of all capital purchases. The annual capital budget is stated there, and the monthly capital board review is one company's account.
• The three reasons a pilot site is unusual, the three groups of benefits, and the observation that a late security objection often stops a rollout at the second or third site rest on Thing Company's experience, not a published study.
• The claim that plants calculate OEE in different ways is our observation. ISO 22400-2 defines one method; it does not report how many plants follow it.
• The site sheet and the phased-approval structure are Thing Company practice.

Thing Company is an independent market validation practice for industrial technology. We test whether a buyer exists at a price that works. For operators, that means testing the rollout case with the people at the next sites before the network commitment.

Taking a pilot to the network? Start with a Sprint Assessment.

Not ready to talk? Run the free rollout site sheet self-check from the toolkit.