Tongwei releases the smart restroom market diagnosis white paper

Contents

01Our original judgment, split into four sentences

Lay the words out first, then check them sentence by sentence. A vague judgment cannot be verified, and it cannot be refuted.

The original words were these:

That sentence contains four assertions that can be tested independently. We split them apart and give each a verdict:

The next eight sections each take one piece of evidence as it stands, and state its source, its statistical basis, and what it can and cannot prove.

02Method: why we do not use the market-size figures in circulation

An industry that cannot even state its own market size has to be re-estimated from the materials underneath.

The first obstacle in this check was not that data could not be found. It was that there is too much data, and it contradicts itself. The same thing — a smart public-toilet system — differs by three orders of magnitude across sources:

Source: abstracts of multiple “smart public toilet / intelligent public toilet” market reports retrieved in public. All of the above are disclosed by the institutions themselves. Statistical basis and sample are not public.

Look at the third row. If China is 120 hundred million yuan a year and the world is only 10—12 hundred million yuan, China would be more than ten times the world. That is impossible in logic. Of these four figures, at most one can be right, and it is possible that none of them is.

This report therefore does not use these figures. It starts again from four layers of checkable material and estimates from the bottom up:

The credibility of these four layers declines in that order, but they can cross-check each other. Take “equipment degrades” as the example: a reporter's return visit gives the scene, the procurement contract explains why the equipment degrades, and the operations-and-maintenance unit price explains why nobody takes it over. Only when the three kinds of evidence corroborate each other does the conclusion stand.

03Evidence 1: the arithmetic of scale — this pool cannot fund R&D

Get the denominator out first. If the denominator is wrong, everything said after it is empty talk.

Weigh it once, from the bottom up

[Estimate] The following is our calculation. Every step can be challenged and replaced:

This figure can also be cross-checked against another, independent path, and the result is tighter than we first estimated. A newly built public toilet costs 25—100 ten thousand yuan in total. The “smart” portion is 5%—15% (early material commonly wrote 15%—25%; that share is high, and it should be counted as 5%—15%), which is 1.25—15 ten thousand yuan per toilet. Multiplied by 6,000—10,000 toilets, the market is about 0.75—15 hundred million yuan. The lower bound of 1.25 ten thousand yuan is close to 1.5 ten thousand yuan, and the upper bound of 15 ten thousand yuan is the same order of magnitude as 10 ten thousand yuan. The medians largely coincide. The estimate of 3—10 hundred million yuan therefore has a basis.

Along the way, a conclusion that is not often stated: "smart" is only 5%—15% of the cost of the whole toilet. From start to finish it was never the main course. That means two things. First, anyone who talks about it as an "industry" is talking about the place of a side dish. Second, precisely because it is only 5%—15%, it is the first thing the client cuts when the budget is cut, and cutting it hurts the least (the price curve in Section 06 is, in essence, this small piece being squeezed again and again).

Then the denominator inside the denominator

How many people are fishing in this pool?

• [Fact] Public industry material: nearly a hundred enterprises registered in China are involved in the smart public-toilet business, and the prevailing way they ship is “buy off-the-shelf generic hardware + adapt open-source software.”
• [Fact] The same material gives the concentration: CR5 is under 28%. The market is plainly fragmented.
• Besides registered enterprises, there is a large number of players who do not register in this category: sanitation service companies, construction companies, prefabricated-toilet factories, operators' city branches, and local system integrators. They are the ones who actually win the bids (see Section 10).

Run the numbers:

The middle card is worth a longer look. A leader's annual revenue does not exceed 2,500 ten thousand yuan — this is not "our estimate." It is the state of this category: nobody in this industry has ever gotten past 3,000 ten thousand yuan. Turn it around and use it to check the pool. The leader is at 2,500 ten thousand yuan. The fifth, estimated at 800—1,500 ten thousand yuan, puts the top five together at about 0.7—0.9 hundred million yuan. Working backward from a CR5 under 28%, the pool is at least 2.5—3 hundred million yuan. That agrees with the bottom-up 3—10 hundred million yuan, and it says the true value is more likely to fall in the lower half of the range.

[Inference] Now do the subtraction. A hardware R&D team of 15 people, at a full cost of 20—30 ten thousand yuan per person per year (wages, social insurance, office, equipment, travel), is 300—450 ten thousand yuan a year. Add tooling, type testing and certification, and a trial run, and a platform from project start to stable shipment takes 1,200—1,800 ten thousand yuan over three years.

On the revenue side the ceiling is only 2,500 ten thousand yuan. At a 40% gross margin that is 1,000 ten thousand yuan. Those 1,000 ten thousand yuan have to pay three things first:

A 15-person team costs 300—450 ten thousand yuan a year. The books barely break even only when sales, delivery, and management are all pressed to the bottom of their ranges and R&D is also at the minimum configuration. If any one of them approaches its upper bound, a gap appears. That is before the second and third years. Cumulative gross profit over three years is 3,000 ten thousand yuan. After deducting 1,950—2,775 ten thousand yuan of expenses, 225—1,050 ten thousand yuan is left. Three years of R&D takes 1,200—1,800 ten thousand yuan. The gap is 150—1,575 ten thousand yuan.

This is also why the industry commonly uses the "buy off-the-shelf generic hardware + adapt open-source software" model. It is not laziness or speculation. It is the choice that the arithmetic can support at the present scale. A leader usually keeps only 3—5 people for integration and adaptation, and buys all the hardware outside. It is not that they do not want to make a product. This account does not allow it.

This is what we mean by “it never grew up”: it is not that practitioners are unwilling to invest in R&D. It is that at the existing revenue scale, a rational operator can hardly recover an R&D investment.

04Evidence 2: even “who is No. 1” has to be bought

How an industry describes its own competitive landscape exposes how it actually competes.

While researching, we searched “smart public toilet competitive landscape.” The first batch of results was the same article — the same title, the same structure, the same paragraphs, even the same punctuation — with only the publishing site changed:

Source: public search results, September—October 2026. The four pieces are highly isomorphic. All of them hand out a tier ranking in the name of “industry survey data,” and the first tier has only one company.

This is not an industry report. This is distribution. Pay for the space, lay one press release across several local news sites, and make it look like a third-party conclusion.

We are not saying that company's product is bad — we have not tested it, and we have no basis to judge. What we are pointing at is something else:

Compare an industry that actually has market share. In switches, whether Centec is No. 1 in domestic switch chips can be settled by reading its annual report and its customer list. There is no need to buy an advertorial. In this industry, the only thing a public channel can turn up is the press release.

What makes the point even more clearly is that the figures inside this same batch of “reports” contradict themselves. In one article about the Chinese market, the same paragraph writes both “the global smart public toilet system market is expected to be USD 1.70 hundred million in 2026” and “the Chinese intelligent public toilet market will exceed 120 hundred million yuan in 2025.” China would be ten times the world. The person who wrote it did not notice, and the person who published it did not notice, because nobody actually reads it.

05Situation 3: the industry standard is a checklist, not indicators

The question this section discusses is the one this report weighs most heavily. It explains “why the industry ends up competing on price.”

First, the standards this industry currently has:

Sources: the 2019 release notice of the Shanghai Landscaping and City Appearance Administrative Bureau; the May 2022 standards press conference of the Chongqing Municipal Urban Administration Bureau; the group-standard information platform of the China National Institute of Standardization, October 2023.

Look at the form of the Shanghai guidelines: 61 facility configurations, classified in three tiers, “basic / recommended / advanced.” This is closer to an order sheet — however much budget there is, that many items are chosen. The Chongqing local standard is more detailed. It states which subsystems the system is composed of, and which functional modules each platform must have.

These standards all have value. They dealt with the disorder of “everyone builds their own, and none of them interoperate.” But they share one feature: what they specify is “what should be present,” not “what result must be achieved.”

Acceptance then becomes this: count the items against the list, tick a box if it is there, and pass when the set is complete. As for whether an item works after it is installed, whether it is accurate, and whether it is still online three months later — the list has no column for that, so the acceptance form has no column for that either.

This inference has a strong counter-proof: wherever there is a quantifiable indicator, product competition has actually happened.

[Fact] Tenders for transport-hub projects explicitly require “stall-identification accuracy ≥99.5%.” Expressway service areas require “automatic denser cleaning at peak, and automatic odor purification.” These requirements can be measured, so the supplier has to put resources into algorithms and sensors. By contrast, a pure municipal procurement usually requires only “configure one stall-guidance system” and “configure one environmental-monitoring system.” Since the result cannot be measured, the supplier does not have to spend extra on it.

06Situation 4: market quotes fell by two-thirds in two years; cost did not follow

This is the only set of figures in the whole paper that comes from us, and it is the project that pushed us out of the turnkey layer.

The evidence in the previous sections all comes from public material. This section uses one of our own projects as the sample, because a public award notice lists only the final transaction price, and does not state what a comparable project cost three years earlier; and that comparison is what decides whether this business can be done.

Source: this company's project records. A single case, not a statistical sample; but the direction agrees with the downward public award prices we found.

Over two years, with the same scale and the same list of items, the price fell by about two-thirds, equivalent to about −42% a year. This is not one supplier's quoting tactic. Over the past two years, when we inquired at the 2023 configuration, the quotes we got were all around 10 ten thousand yuan.

A point of basis has to be stated first: 150 stalls and 10 ten thousand yuan works out to 667 yuan per stall. At the 2023 configuration, with not one item removed, the cost is about 18.6 ten thousand yuan, 1,240 yuan per stall. The two differ by about 46%. So the curve below is not “the same thing got cheaper.” It is price first falling through the cost line, and the configuration then starting to shrink. The second half of this section measures those two things separately.

Put the two lines together and what you see is not "fierce competition." It is a more specific phenomenon: the quote line crossed the cost line in the first half of 2024, and it never came back. After that, every bid is a bid below cost. The only difference is who loses more, and who loses for longer.

The 20 ten thousand yuan that was cut: 11 ten thousand is profit to zero, 8 ten thousand is a cost gap

If the price cut came from higher efficiency, cost should have fallen in step. It did not. Over these two years, of the items that make up cost, four rose, and only one fell:

The difference of 20 ten thousand yuan splits roughly in two: one part goes to zero, and one part becomes a cost gap.

First: of that 30 ten thousand yuan in 2023, the 11 ten thousand yuan outside the hardware all went to zero. Working backward from the delivery cost of 150 stalls (for comparison, restated uniformly at 2025 purchase prices), hardware and installation on that job were about 18.6 ten thousand yuan (the next subsection has the breakdown). The remaining 11.4 ten thousand yuan was the following:

Of these five items, what remains inside the 10 ten thousand yuan of 2025 is zero. They were not “squeezed down.” The whole item is gone.

Second: even if all five of these costs are cut, the quote is still not enough. The 2025 quote of 10 ten thousand yuan is still below the “spec not lowered” delivery cost of 18.6 ten thousand yuan. The difference is still 8.6 ten thousand yuan. There is no service left to save. The only place left to cut is the configuration itself.

Only the two together make that 20 ten thousand yuan: about 11.4 ten thousand yuan is “profit to zero,” and about 8.6 ten thousand yuan is “the cost gap.” The first is a commercial act. The second is a physical constraint.

And this can go on because the checklist has not changed. The checklist only requires “configure one stall-guidance system.” It does not say whether that set includes operations and maintenance, customization, or redundancy (Section 05). As long as the checklist does not change, room to cut the price can always be squeezed out of the parts the checklist does not explicitly require.

An account in which nothing is cut

This is where our judgment at the time came from. Suppose cost is pressed to the bottom: no middleman, dealing with the client directly; installation handed to a local crew, settled by person-day, with no crew of our own; supply that does not lower any specification, and construction done to the specification. Back-calculated against 150 stalls (both sides combined, including the third restroom) and a quote of 10 ten thousand yuan:

Note 1: this table does not include a "software platform and integration." On a service-area project of this kind, the guidance screen and local management come with the equipment and do not connect to an upper platform, so this item is counted as zero.

Note 2: sensing devices are priced at the bought-in market price. If the whole sensing chain is replaced by the in-house unit in Section 15 (BOM about 87—134 yuan per stall), the sensing portion can fall from 6.1 ten thousand yuan to 1.3—2.0 ten thousand yuan, and total cost from 18.6 ten thousand yuan to about 14 ten thousand yuan — still far above 10 ten thousand yuan. Productization can turn gross margin from negative to positive, but it does not change this arithmetic.

Not yet included: R&D amortization; bidding cost (bid document + samples + travel, 0.5—1 ten thousand yuan per bid, paid even if the bid fails); interest on funds advanced (a 3—6 month payment term); and the back-and-forth time of acceptance.

The 8.6 ten thousand yuan that has to be made up is not in “services.” It is in equipment and construction cost. The five services have already been cut entirely. There is nowhere left to save. To make up this 8.6 ten thousand yuan, the only place left to cut is the configuration in the table below. After the schemes in the table save 10.5 ten thousand yuan in total, a quote of 10 ten thousand yuan has some gross margin.

Look at this account: every configuration item that is cut corresponds, three years later, to a specific problem. This is not a vague “the quality is a bit worse.” The equipment will fail at a specific point in the future. A low price is not “value for money.” It only moves cost off the quotation and onto the timeline — the party who pays is still the client, only it is the client three years later.

One more point ties the calculation in this section to the observation in the previous one: the screens, environmental monitoring, and footfall counting cut in the table above are exactly the items Guangzhou Metro cancelled on the 2026 extension (Section 08). This is not two organizations happening to think of the same thing. The same cost account produced the same result.

The other way around: supplying direct from the manufacturer is a disadvantage in this market

This is the most unexpected point, and the one most worth our attention.

It is usually said that a middleman adds 15%—25%. Remove the middleman, and the manufacturer selling direct gains a price advantage. Put the two speeds of change together, and calculate them on the scale of the 2023 project of 30 ten thousand yuan:

All of the middleman's profit that is saved is only enough to offset five to nine months of the price decline. That is, the price advantage gained by selling direct does not last a year before the market itself wipes it out. And you permanently lose the channel, and you lose pricing flexibility on the next bid.

More important, what the middleman provides is not only a channel. It also includes advancing funds, coordinating acceptance, and the ability to decide whether to lower the specification and to bear the corresponding responsibility. Once the manufacturer signs directly, all of this falls on the manufacturer. If the specification is lowered in a contract the manufacturer signed itself, the reputation that is damaged is its own product's, not someone else's.

07Situation 5: after project acceptance, there is no guardian

The operations-and-maintenance budget for public toilets is not small, but the budget lines contain no item called “system.”

First, how much operations-and-maintenance budget a public toilet actually has in a year

This is the most persuasive set of figures in the public material. Below are the ceiling unit prices in tenders for public-toilet cleaning and facility-maintenance services in various places:

Sources: tender notices and contract texts published on the China Government Procurement Network and the government-procurement platforms of Jiangsu, Sichuan, and Guangdong. The table shows ceiling unit prices. Actual transaction prices are usually a little lower.

Put these figures next to the cost of the “smart” portion, and the gap between them is obvious:

The operations-and-maintenance money is 5—10 times the hardware, and it is there every year. These are figures that would tempt anyone who makes equipment. The problem is this — the budget line for this money is called “public-toilet cleaning services” and “routine facility repair.” It is paid to sanitation service companies, and the scope of service is written as “daily cleaning, septic-tank emptying, sanitary-ware consumables, and routine facility repair.”

Inside it, there is not one item called “smart-system operations and maintenance,” and there is no corresponding unit-price standard.

Then the out-of-warranty clauses in the contracts

There is a warranty, and it is not short:

Sources: a contract notice on the Changsha government procurement website, a procurement-requirements document on the Sichuan government procurement website, and tender documents on the Shaanxi government procurement website.

These clauses have one thing in common: what is promised after warranty is “providing spare parts” or “charging at cost,” not “continuing operations and maintenance.” That is, the contract writes the end of responsibility as the day the warranty expires. This is not a question of whether the vendor is cunning. The contract terms and the tender requirements are written this way.

Then the account that does not add up

[Fact] The quoting basis the industry states in public: after warranty, the annual maintenance fee for ordinary equipment is about 5%—8% of the equipment's total price; the annual operations-and-maintenance fee for an IoT platform is usually between 1 and 5 ten thousand yuan.

Put these numbers together: a basic system is 1.5 ten thousand yuan, and the platform's annual fee is 1—5 ten thousand yuan. One year of platform operations and maintenance costs more than the entire hardware set.

[Inference] This explains why the renewal rate is extremely low. For a set of equipment that has already passed acceptance and has a book value of only 1.5 ten thousand yuan, the client would have to add 1—5 ten thousand yuan a year as a “platform service fee.” That spending has no budget line, no precedent, and no unit-price standard, and it has to go through government procurement again. The great majority of units choose not to approve the money, and the system gradually stops — acceptance has already passed, and the appraisal does not look at this item.

08Situation 6: the same client, three years later, and the project requirements are down to two items

Guangzhou Metro made two changes: the smart-restroom configuration went from nine items to two, and operations and maintenance was transferred from the manufacturer to the M&E company. And the transfer of operations and maintenance is the reason the configuration shrank.

The previous two sections were about “price” and “the operations-and-maintenance budget.” This section looks at a specific project: the same city, the same kind of station, the same owner, and over three years the procurement requirements changed, and the change is more thorough than the change in price.

First, how the list changed

[Fact] The smart-restroom configuration on Guangzhou Metro's existing lines three years ago, set against the configuration on the 2026 extension:

Source: this company's participation in the projects and on-site comparison records (existing lines in 2023, the extension in 2026). The list follows the owner's requirements document. No inference is made about specific brands or models.

Dark green = kept; light yellow = added; gray = moved out of the smart package (the function remains, under another trade); red = cut.

Nine items became two. This is not “the standard dropped a little.” A system was cut down to a switch.

The pattern: the boundary of the configuration is the boundary of the post

Rearrange this table by “whether there is a person behind it,” and the pattern is very clean. There is no exception:

The two items that remain each have a clear post that uses them. Occupancy sensing can show “occupied / vacant,” and cleaners and security use it to judge which stalls are in use and when they need cleaning. Smoking detection can find “someone is smoking” and hand it to smoking-control patrols and fire inspections — this work has a clear external appraisal, and failing to check it has consequences.

The two items that were “reassigned” also have a clear post that owns them. Lighting is the lighting trade's job. Flushing belongs to plumbing and sanitary ware. In our view these two are “smart” functions, but in the owner's organization they belong somewhere else. They were moved out of the smart package. The function did not shrink. The responsible person and the budget line changed. Functions that were reassigned did not disappear. What actually disappeared are the ones that were cut.

The six configurations that were cut have no clear responsible post. The five kinds of data — air quality, temperature and humidity, hydrogen sulfide, footfall, and feedback — plus a large screen, are useful only when someone uses the data to schedule work, deal with problems, and keep improving. The owner's staffing table has no such post, and the M&E company's contract does not include this kind of work. Data that no post is responsible for cannot do any work.

This is the core judgment of this section: the boundary of the configuration = the boundary of the post. Whether a function stays on the list does not depend on whether the technology is advanced, or whether the case study looks good. It depends on whether there is a specific person behind it, and whether that person's appraisal includes this item.

The second change: operations and maintenance passed from the manufacturer to the M&E company

[Fact] Three years ago, the owner would from time to time ask the manufacturer to help maintain the equipment and the system — the manufacturer sent people, and took part for free or nearly for free. In 2026, this was basically handed to the M&E company, and the manufacturer is rarely asked to take part. The result is that maintenance and repair of the old system have basically stalled.

This change is more fatal than cutting configuration items, because it explains why the configuration was cut. First look at what the party taking over can actually catch:

The result is: when the hardware breaks, someone is in charge. When the system breaks, no one is. What the site calls “maintenance and repair have basically stalled” does not mean nobody is at work. It means the existing maintenance staff can handle only part of the problems.

Section 07's conclusion also needs a correction here. More precisely, there is someone who takes over, but they can handle only the half that can be priced per piece. “There is no one to take over” is a judgment. “They can catch only half” is a fact that can be carried into the work, and it also says how the product should be designed.

Join the two changes: why it gets cut

When the owner reviews this three years later, the conclusion will not be “this system needs stronger operations-and-maintenance support.” It is more likely to be “we cannot maintain this thing; buy less of it next round.”

This is a very rational conclusion. It does not need an argument about whether the technology has value. It needs only one fact: the six items cut in the previous round produced not a single action in three years.

This also explains a phenomenon that price alone cannot explain: some places are not short of money, and they still delete configuration items. The problem is not the budget. It is that they bought it once and nobody maintained it. The price curve in Section 06 reflects the same point: the supply side is cutting price, the demand side is cutting items, and the two are two sides of the same thing.

One more point, the one that matters most to us

Three years ago the manufacturer could take part in operations and maintenance, for free and informally, on the basis of “keep the customer, and wait for the next batch of projects.” This is the only channel through which a company of this kind obtains field knowledge.

In 2026 that channel closed. After it was handed to the M&E company, what the manufacturer lost was not operations-and-maintenance revenue (there was none to begin with). It was three things: knowing what actually failed on site, knowing what the customer will change next round, and knowing what state its own product is in during the third year.

Section 13 says technology has become a liability in this market. This point is the other side of that: even the field knowledge required to make the product right was cut off with it. Product iteration without field knowledge can only be guesswork.

09Situation 7: the equipment has not broken, but it degrades and loses its function

In this market, the way users “vote with their feet” is not to complain. It is to take the equipment out.

All of the material in this section comes from public reporting, not from our inference. Three cases, and the nature of them is exactly the same.

Case 1: Laoshan, Qingdao — three standalone smart public toilets

Put into use before October 2022. In December 2022, citizens reported that the smart public toilet on Pingshan Road was shut tight and could not be used, because the access-control system had failed.

In June 2024, a reporter went back to the three public toilets (Pingshan Road, Hai'an Road, and Haijiang Road). The scene was this:

• The smart access-control system had been removed
• The smart occupancy light still showed green when the stall was occupied (green means vacant)
• The automatic-flush sensors on washbasins and squat toilets had failed
• The time, temperature, and other information on the environmental monitoring screen were all wrong
• The automatic flush units on washbasins and squat toilets had been replaced by conventional manual units

Source: Qingdao Daily, 14 June 2024, page 05, the series “Smart convenience equipment trapped as mere decoration.”

Pay particular attention to the last item. It is not that the equipment broke and nobody repaired it. The maintenance unit removed the smart parts and put manual parts back. This is the lowest-cost way to “repair”: a sensor faucet is 300 yuan, a manual faucet is 30 yuan; a sensor faucet breaks once every three months, a manual faucet does not break for three years. From the maintenance party's point of view, this is a completely rational choice.

The same report also says the park management stated that it is responsible only for landscaping and sanitation cleaning, and that maintenance of the sports facilities is the maintenance contractor's job. When the reporter called the maintenance contractor, the phone was never answered.

Case 2: Shengsi, Zhejiang — more than ten smart public toilets

• The guidance screen at the Yunlong Road public toilet was black
• Data at the Shahe Road public toilets and others had not been updated for a long time, and “today's water consumption” was always zero
• The cause: partly an unstable network, and the electronic water meters were also still being procured
• Humidity and footfall on the guidance screen were disconnected from what citizens care about, toilet-paper supply and cleanliness; citizens all walked straight in to check for a free stall, and the screen might as well not have been there

Source: republished on the Shengsi County People's Government website, “Island City Focus: Smart convenience equipment trapped as mere decoration” (2025).

“The electronic water meters were also still being procured” — this is the sharpest sentence in the whole set of materials. It shows: at acceptance, the data was fake; and acceptance passed. “Today's water consumption” showing 0 is not a fault. A meter was never connected.

Case 3: the online rate of the entire smart-sanitation industry

The industry's own account (not an outside criticism): 2020—2023 was the period of concentrated smart-sanitation construction nationwide. Projects “were generally hardware procurement and one-off project delivery, and lacked a full-life-cycle operations-and-maintenance plan.” The consequence:

• In some cities, smart sanitation terminals commissioned in 2021—2023 had an equipment online rate under 30% two or three years after going live
• A large amount of equipment is physically intact and far from its physical end of life, yet has completely lost its smart capability because of “lagging algorithm updates, the cloud platform going offline, the original manufacturer's operations and maintenance ending, the communications link breaking, and data interfaces being closed”
• Inside the industry this is called “pseudo-scrapping” — the hardware is not broken; the smart function is dead
• One cause: early projects mostly used the vendor's private closed protocols. Equipment software and hardware were deeply bound, and a third party could not connect or upgrade

Source: a public article in the industry press, “Under misaligned technology iteration, how to break the ‘pseudo-scrapping’ of smart sanitation equipment?” (2026).

[Inference] An online rate under 30% means that out of every 10 projects and devices, 7 fail within two years. This number is more honest than any customer-satisfaction survey. It is also the result of the account we ran in Section 03: no operations-and-maintenance line → it stops when warranty ends → the platform goes offline → the device becomes a brick.

10Situation 8: who is taking orders from the user side

The list of winning bidders is more accurate than any promotional material.

We sorted the award results we found by the type of winning bidder:

Source: the original texts of award-result notices on local government-procurement websites and public-resource trading platforms.

On this list, not one winning bidder has “smart public-toilet systems” as its main business. Operators, China Telecom-system integrators, construction companies, environmental-equipment vendors, and biodegradation companies are all on the list. For them, the technology is only a supporting step inside the project, not the core competence.

This is not an accident. It is the same thing as the earlier evidence, seen from another side:

• Because the scale is small, a specialist vendor cannot grow large, and can only be a component supplier
• Because the standard is a checklist, anyone can assemble one against the list; no specialist capability is required
• Because acceptance looks at the item count, engineering companies and integrators can handle it completely
• Because what the client wants is “a building that can be used,” not “a system,” construction companies and prefabrication plants are naturally closer

The Zhoushan project makes the point more clearly. The three bid prices were:

[Fact] The lowest price, 33.6 ten thousand yuan, ranked third and did not win. The winning price, 44.8 ten thousand yuan, is 33% above the lowest and 32% below the highest.

This result is more worth thinking about than “lowest price wins.” It says two things. One: evaluation is not purely lowest-price-wins; there is also a technical score and a credit score. Two: on the same project, the three quotes differed by nearly a factor of two (33.6 ten thousand yuan to 65.6 ten thousand yuan) — this is not vendors giving a discount. The tender documents did not say clearly what the words “smart public toilet” mean. Each bidder understood something completely different, and quoted three different things.

Unclear requirements → quote dispersion near 100% → evaluation can only rely on subjective scores → the weight of relationships and qualifications rises. Where this chain points, in the end, is still technical competition being eliminated.

11Counter-evidence: where it has not rotted

Without this section, this report would be emotion. And emotion cannot guide a decision.

We also looked specifically for counterexamples: among smart restrooms, are there projects that were done well? There are, and more than one. What they have in common is clear:

These three kinds of setting differ from a municipal public toilet in only one way: the owner operates it, and the owner bears the consequences.

This table is the core of this report. It narrows the sentence “the market is not healthy” to one sentence:

So the phrase “the market has rotted” in our original judgment is not precise enough. The accurate statement is: the G-end has rotted, the B-end has not, and the B-end may be the only place in this industry where product competition still exists.

This also explains why we said “needs correction” in Section 01 — it is not a wording problem. It changes the action. If the whole market has rotted, the conclusion is to exit. If the G-end has rotted and the B-end is alive, the conclusion is to change the customer, not to change the industry.

But “change the customer” has to be said more specifically. On the analysis in Section 08, what we actually have to find is not a generic “B-end.” It is an owner that has operations-and-maintenance capacity, or that is willing to keep the operations-and-maintenance function. Within rail transit, an owner with its own maintenance crew and an owner that outsources the whole of it to an M&E company are facing two different markets. Within large enterprises, handing the system to the information center or to the logistics department also produces different results. Before signing, asking clearly “who will repair this system afterwards” judges whether a project suits us better than asking only about the budget.

12The loop: why this outcome is inevitable

String the eight pieces of evidence together and a self-sustaining loop appears. Nobody designed it, but it turns by itself.

The most vicious part of this loop is step ⑧. When the old system breaks it is not repaired. A new build or a renovation is applied for directly — because “repair” has no budget line, and “build” has a special fund. So each round of “pseudo-scrapping” is not a loss. It is the source of the next round of orders.

[Inference] This means nobody in the industry has an incentive to maintain the old system. The equipment vendor hopes it breaks (easier to sell a new one). The integrator hopes it breaks (easier to take the renovation). The client hopes it breaks (easier to apply for budget). The only party that hopes it does not break is the citizen who uses it, and the citizen is not at the table.

Section 08 adds a layer of explanation to this loop: after ⑧, the next round's configuration shrinks back along the line of “whether a post can catch it.” Nine items becoming two is not an exception. It is the necessary result when this loop turns a second time — the data left by the first turn had nobody to use it, so the second turn does not buy it.

So this loop cannot be broken by “raising product quality,” and it cannot be broken by “strengthening after-sales.” There are only two places to act: the bid-evaluation rules at ⑤, and the payment structure at ⑥. And that is the whole content of the recommendations in the next section.

13Why technology becomes a liability

The previous section was the market's loop. This section is our own. Because we have been inside this loop, and paid the tuition, we came to understand one thing — here, a strong technical capability is not an advantage. It is a handle.

13.1 Say the most unpleasant part first: on the project site, nobody cares

We used to have a default premise: if the product is truly easier to use, more stable, and saves the operations side more money, the customer will recognize it sooner or later. That premise holds in other industries. It does not hold here.

This is not the complaint that “the customer cannot tell good from bad.” It is that four roles each do not need this:

This table explains a question we could not make sense of for a long time: why is our product more stable on site than other people's, and yet the customer gives no feedback? It is not that the customer did not see it. It is that nobody is appraised on this item.

13.2 An account that can be calculated

Translate “taking the product seriously” into financial language, and it looks like this (the following is our own calculation, [Estimate]):

Look at the last row. This is not “the investment has not yet reached the harvest.” It is also not only “educating the market takes time.” It is an account whose result is negative. The more seriously the product is made, the higher the cost per set. The selling price is very hard to raise, because the scoring items contain no corresponding indicator.

13.3 Why it is 0: at the moment of the deal, good and bad are the same price

The hardest point to accept is this: on the day of acceptance, a good product and a bad product look exactly the same.

Both power on, both produce numbers, both can be ticked, both can be signed for. To see the difference takes three things: two public toilets of the same scale, the same climate, and the same traffic as a control; online-rate data for more than two years continuously; and the ability to separate “an equipment problem” from “an installation problem, a network problem, and nobody looking after it.” Of these three, there is not one.

Look at where the vertical lines sit. The moment the difference starts to be obvious (about 24 months) is almost the moment the warranty expires (24—36 months). Problems inside the warranty are repaired by the vendor, so “it works” covers up “it is unstable.” The moment the warranty expires, repair responsibility is drawn away under the contract, and operations and maintenance has no corresponding budget line, so the curve falls off a cliff. Just when quality starts to speak, the people who would listen have all left.

This is why “easy to use” cannot be priced in this market. It is not a price problem, and it is not a problem of persuading the customer. It is a problem of the observation window — the window does not exist at all.

13.4 This is not “the customer cannot tell good from bad.” It is a credence good, in the economic sense

Later we found a statement that is more accurate than “the customer does not understand,” and that explains more:

• Search goods: you can judge before you buy (the cut of a piece of clothing)
• Experience goods: one use tells you (the taste of a restaurant)
• Credence goods: even after use you still cannot judge (a lawyer's service, the quality of a repair)

A smart public toilet belongs to the third kind, credence goods, and this feature is especially pronounced. A citizen can feel that “this screen is useless,” but cannot judge that “the false-alarm rate is 8 percentage points higher than a peer's.” The operations party knows “the equipment has broken again,” but cannot tell whether the cause is the product, the installation, the network, or a lack of maintenance. In this chain, no link can pass quality information back to the buyer.

In economics this is called a market for lemons: when the buyer cannot observe quality, the price mechanism systematically drives out high-quality supply — it is not that the inferior product beat the superior product. It is that the superior product cannot prove that it is superior, and so cannot recover its cost.

We used to think this was “the industry is still early, and it will slowly get better.” As it now looks, a credence-good market does not improve by itself. It needs a measurement mechanism built from outside. That is exactly the four indicators in Section 16.

13.5 Our own trap: the stronger the technical capability, the fewer product orders we can get

Tongwei has a problem of its own. Because we can do things others cannot, we often meet a conversation like this:

Every time we take one, it becomes a piece of revenue. But strung together, it is a loop:

After step ⑩ it returns to step ② — so every step in this loop is right, and strung together it turns in place.

This loop is more hidden than the market loop outside. Each deal makes sense on its own: the customer has a requirement, we can do it, there is revenue, and the customer is satisfied. But strung together, customization projects keep increasing, experience and code do not settle into a reusable product, and in the end we are still turning in place.

And it has a self-reinforcing side effect: the more strange requirements we can take, the more easily we are introduced to the next person with a strange requirement. Reputation here becomes a burden — not “this company's product is good,” but “this company can change anything.” The first brings repeat purchase. The second brings customization.

[Fact] This also explains a phenomenon mentioned earlier: our capability is described as “miscellaneous.” We can take all kinds of field protocols and strange schemes. Where a market is mature, “miscellaneous” is a barrier. In this market, “miscellaneous” is the entrance to being customized.

13.6 Then what is technology still good for: it buys the ticket, not the price

At this point it is easy to swing to the other extreme — then stop doing technology. That is also wrong. Technology is not useless in this market. Its use is not the use we thought:

Therefore, a technology investment should be managed as a “qualification cost,” and only the minimum required to meet the requirement should be spent; it should not be managed as a “product investment,” chasing a lead over the industry. The budget logic of the two kinds of spending is completely different: the first is “stop when it is enough,” the second is “add when it is not enough.” In the past we kept spending on the second logic, and received only the return that belongs to the first logic.

14A clearer conclusion on the state of the industry

The direction of our original judgment was not wrong, but three wordings can mislead a decision. After putting the report's material together and looking back at past projects, we see it more clearly.

Correction 1: not “it rotted,” but “it never grew up”

“Rotted” implies a premise — that it was once good. What we found is that there was no such stage. When Shanghai issued the guidelines in 2019, the industry's own words were “because the corresponding standards were lacking, the products provided by each builder were not the same.” 2021—2023 was the construction peak, and in the same period “an online rate under 30%” already appeared. In 2026, the industry is still defining rank by buying advertorials.

There was no period that was healthy. So the correct statement is: this market has never reached a scale that can sustain a technology investment, and so it has never entered a stage of product competition.

Why the distinction matters: saying “it rotted” makes a person think “wait until it is regulated, then enter.” Saying “it never grew up” makes a person think “it may never be regulated, and the only move is to go around it.” Only the latter is a judgment that can be acted on.

Correction 2: technical competition did happen. There was just no scoreboard

We originally said “it competed on price before it had entered healthy competition on technical proposals.” That is not accurate. The sensing layer went through three generations in six years:

Source: an industry technology survey article (2026), including indicators stated by vendors themselves. Confidence in the indicator portion: medium.

The technology really is advancing. Sensing accuracy going from “often misjudged” to “a 0.1% false alarm” is a solid product upgrade. But the client does not measure this item. What the acceptance checklist writes is “configure one stall-guidance system.” It does not write “false-alarm rate ≤0.5%.” So the cost brought by the upgrade cannot be turned into a price.

The result is: nobody is willing to be the first to upgrade. Because an upgrade adds 200 yuan of cost, and the client will not pay 300 yuan more for that 200 yuan — at acceptance he will only ask “is this item present.”

So the problem is not “there is no competition.” It is “competition happened, and it was not scored.” This means what we need is not to keep raising the technology for its own sake. It is to get the technical result into the evaluation system.

Correction 3: not “after-sales do not matter,” but “after-sales stops at the boundary of the warranty”

This distinction is critical, because it decides how we should design the business model.

So it is not that vendors are irresponsible. It is that responsibility is drawn by the contract as a precise line, and on the far side of that line nobody pays. The smart parts taken out in Section 09 all happened on the other side of that line.

This correction matters enormously to Tongwei: what we have to sell is not “better after-sales.” It is “after-sales that someone pays for.” These are two completely different things. The first depends on service attitude. The second depends on commercial design.

15What this judgment means for Tongwei

If the analysis above is right, then there are four things we can no longer do.

Four things we will not do

Then what to do: turn “acceptance” into the product

This is the conclusion drawn from the analysis above. Since, in this market, the first thing a product has to solve is “whether it can pass acceptance,” we make the ability to pass acceptance part of the product, and at the same time push price and delivery cost down as far as we can.

The concrete form: a configurable in-stall monitoring unit (including a gateway), supplied white-label to integrators, prefabrication plants, and contractors who take on M&E maintenance, with an automatically generated acceptance-document pack delivered with the goods.

A cost account

[Fact] The publicly listed channel price: a complete sensing-hardware set covering 10 stalls (door-lock sensor, indicator light, ammonia, hydrogen sulfide, temperature and humidity, footfall, water-meter and electricity-meter interfaces, gateway) totals about 6,236 yuan. The listed online price of one stall's occupied/vacant sign plus sensor is only a little over 200 yuan.

[Estimate] Our BOM breakdown (10 stalls):

[Inference] If the ex-factory price is set at 2,200—2,800 yuan (10 stalls), gross margin is about 55%—65%, and it is still 55%—65% below the market list price. An integrator has a reason to adopt our scheme: the quote can stay the same, while cost falls sharply. We can also keep a profit at a lower unit price.

This is how we step around the price war: not by staying out of it entirely, but by competing at another layer. The price war at the turnkey layer, in turn, brings us demand.

16Market proposals: change the way projects are organized, not the slogans

The four items below are proposals we make in public, for people who are willing to push for change. They are not moral appeals. They are engineering measures that can be put in place.

Proposal 1: change acceptance from “item count” to “four indicators”

Checklist acceptance is the root of the price war (Section 05). To break it, the checklist does not have to be abolished. Outside the checklist, four measurable numbers only have to be added:

These four indicators can all be generated by the system automatically, with no extra manual measurement. They also cannot be met by “installing one more sensor.” The system has to actually run, and someone has to deal with anomalies in time.

[Fact] This kind of practice already has a precedent: Indore, India, a project of 350 IoT public toilets, planned a mechanism that “notifies progressively higher managers when an anomaly is not cleared within the specified time,” and required an event to leave five timestamps: occurrence, dispatch, arrival, handling, and clearance. They wrote one summary explicitly: an alert is not an improvement.

Proposal 2: give operations-and-maintenance funds a line of their own

The present state: the 7.8—14.9 ten thousand yuan a year spent to run each public toilet is all inside the envelope called “cleaning services.” Not one yuan is designated for the system.

The proposal, outside the cleaning-service contract, is to list a separate item, “smart-system operations and maintenance,” and to give a reference unit price:

• Priced per stall: 300—500 yuan per stall per year (about 3,000—5,000 yuan a year for 10 stalls)
• Or priced on the total equipment price: 8%—12% a year (a little above the industry's current 5%—8%, because it has to cover the platform and remote diagnosis)
• The service period runs with the cleaning contract (usually 2—3 years), appraised separately, paid separately

The fiscal effect of this spending is small: a public toilet's annual operations and maintenance of 10 ten thousand yuan, plus 4,000 yuan, is only 4%. But once this money exists, “whether the system can keep running” no longer depends on luck. It is written into the contract.

Proposal 3: write open interfaces into the acceptance conditions

[Fact] The industry's own account: early projects “mostly used the vendor's private closed protocols, equipment software and hardware were deeply bound, and a third-party organization could not connect or upgrade.” This is one of the main causes of “pseudo-scrapping.”

The proposal is to write three clauses so they cannot be bent:

1. At bidding, a complete communications-protocol document and data dictionary must be submitted as part of the technical documents. They are not returned if the bid fails.
2. At acceptance, one live test of integration with a third-party platform must be completed (even if it is only pushing data to a test platform). No pass, no acceptance.
3. The contract shall stipulate: when the vendor stops service or exits the market, it shall provide the last firmware and interface documents at no charge.

These three clauses add no hardware cost, but they stop the whole asset being locked to a single supplier. For the client they are pure gain. For a poor supplier they are a disaster — and that is exactly why they work.

Proposal 4: separate warranty from operations and maintenance, and tender operations and maintenance on its own

Now the warranty is 2—3 years and operations and maintenance is 0 years, with a cliff between them. The proposal is to change it to “3 years of warranty + 3+2 years of operations and maintenance,” and the operations-and-maintenance contract is tendered separately or renewed separately, not bundled into the equipment purchase.

This has a side effect, and it is a good one: tendering operations and maintenance on its own will bring forth a group of companies that do only operations and maintenance. They do not sell equipment. They sell “keeping the system alive,” and they charge by online rate and by time to close the loop. Once this role exists, the quality of the equipment will actually be appraised — because an operations-and-maintenance company will refuse to buy something that is always broken.

1712-month actions and stop-loss lines

Whether the judgment is right has to be verified, in the end, by doing it. The following is our own plan, not advice for anyone else.

Three stop-loss lines

The ceiling on investment

[Estimate] Investment for the whole of the first stage (through 2027-06) is held to 2 people × 6 months + a one-off trial production of 3 ten thousand yuan, about 20 ten thousand yuan in total. No tooling, no certification, no platform, no brand.

The basis of this ceiling is simple: if this judgment is right, 20 ten thousand yuan is enough to verify it; if this judgment is wrong, 20 ten thousand yuan is a loss we can afford.

18Appendix: sources and confidence

The source and the degree of credibility of the key figures in this report are as follows. What can be checked is marked Fact. What is derived from facts is marked Inference. What we calculated ourselves is marked Estimate.

What this report did not do

• We did not obtain any vendor's real financial data. No listed company in the industry discloses this business separately, so “leaders' annual revenue of 2—6 × 10 million yuan” is a calculation, and it may be off by more than a factor of two.
• We did not do site visits. Every “state after completion” comes from public reporting, not from what we saw ourselves. The reporting may be selective.
• The original quotations for the 2023 order and the 2025 order were not reconciled line by line. On the price curve in Section 06, one end is our project record and the other is inquiries in multiple places. The scale is the same, but the lists were not compared item by item, so the slope can only be an order-of-magnitude judgment and cannot be used as a precise figure.
• We did not verify the “acceptance pass rate.” We cannot judge how many projects “scraped through.” We can only infer it from clauses and cases.
• The sample leans municipal. Procurement rules in counties, scenic areas, and service areas differ greatly. The conclusions of this report hold most firmly for the municipal G-end, and they explicitly do not hold for the B-end (see Section 11).