At Drinkwell, we apply science & technology to ensure safe drinking water — by innovating on three things at once: the treatment technology (the horse), the delivery method (the harness), and utilization (the habit). A hype-free account — every figure below is drawn verbatim from peer-reviewed journals, independent evaluations, and government validations.
The world's largest mass poisoning is invisible, tasteless, and dissolved in the water table.
An estimated 1.2 billion people rely on drinking water sources that are not safely managed.12 Within that, naturally occurring arsenic in groundwater affects nearly fifty countries and threatens more than 200 million people with health impairments, according to the WHO.10,12 Fluoride puts a further 300 million people at risk of skeletal and dental fluorosis.6
200M+
people threatened by arsenic in groundwater across ~50 countries — the problem this technology was invented for.10,12
South Asia is the epicenter — roughly 100 million people are affected across Bangladesh and eastern India alone.2 In Bangladesh, the World Bank reports that while 98% of the population has access to "improved" water sources, 41% of those improved sources carry E. coli and about 13% exceed even the national arsenic limit of 50 µg/L15; one study attributes 43,000 deaths a year to arsenic-related illness.11 In India, testing across Uttar Pradesh alone flagged 2,610 habitations above 50 µg/L and another 10,434 above the WHO guideline11, while 21 of 29 states face natural fluoride contamination.20
The human ledger is concrete. In Binimaypara, a West Bengal village of fewer than 1,000 people, 24 cancer-related deaths in the 1990s were attributed to arsenic in the community's wells.10 That village has now been drinking treated water from a HIX system for over a decade.
02 — The Framework
The horse, the harness & the habit.
At Drinkwell, we apply science & technology to ensure safe drinking water — by innovating on three things at once.
I · The Horse — Treatment technology
Alone in a field, it just eats grass.
A drum of HIX-Nano is genuinely powerful chemistry — but sitting in a warehouse it removes zero micrograms of arsenic. The horse is raw capability: what it can remove, how long it can run, where its waste goes.
Proven in 03–05 →
II · The Harness — Delivery method
Straps and reins don't make the horse stronger. They hook it to a job.
The same horse can pull a plow, a sleigh, or a carriage: a filter in the home, a kiosk in the village, a piped system across a city. Each harness has its own engineering and its own economics.
Proven in 06–07 →
III · The Habit — Utilization
A carriage line only matters if people keep riding.
The ride must be worth choosing every day — trusted, paid for, and safe all the way home. Riders never think about the reins; they just come back tomorrow. That is utilization: the habit.
Measured in 08 →
The harness comes in three main forms: a domestic water filter in the home, a community kiosk or Water ATM, and a piped water system run with a utility. Each layer earns its own evidence, and this report keeps that evidence separate — the deepest published record today is on kiosks, while the evidence base for the filter and for piped systems is being built now, with early results already in hand.
The horse also travels beyond Drinkwell's own harnesses, as resin supplied into government utility plants26; and in the village systems of West Bengal, Bihar, and rural Bangladesh, horse and harness work as one rig.10
Read the report accordingly: Part I is evidence about the horse. Part II covers the three harness forms, where each stands on evidence, and a map of deployments. Part III is evidence about the habit. Sections 09 and 10 close with the independent validators and with what doesn't work yet, sorted by which of the three layers is failing.
I
Part I — The Treatment Technology
The Horse.
HIX-Nano: chemistry that grabs only the toxic ions — then lets go on command, hundreds of thousands of liters later.
25,000 bed volumes / cycle≥5 regenerationsTCLP 81–135 vs 1,000 µg/L limit
03 — The Horse · Chemistry
The molecule with hands.
A polymer bead studded with billions of nanoparticle "hands" that grab only the toxic ions — and let everything else pass.
A hybrid ion exchanger (HIX) starts with a durable ion-exchange resin bead and disperses metal-oxide nanoparticles — hydrated ferric oxide for arsenic, in the 10–100 nanometer range12; zirconium oxide for fluoride, at a mean particle size of ~15 nm and a loading of 110–130 mg zirconium per gram of resin6 — throughout its pores. The polymer phase concentrates target anions toward the nanoparticles (the Donnan membrane effect, shown in the founding research to increase the volume of arsenic-laden water treated by three orders of magnitude versus the wrong host polymer11), and the nanoparticles bind them selectively.
Selectivity is the entire game. Groundwater carries sulfate, chloride, silica, bicarbonate — harmless ions that outnumber arsenic thousands to one. Media that grab indiscriminately fill up on the harmless ions and exhaust early. In controlled tests, sulfate at 500 mg/L — fifty thousand times typical arsenic levels — had practically no effect on HAIX-Zr's fluoride capture.6
The efficiency is structural, not brute-force: HAIX matches the arsenic capacity of leading granular ferric oxide media while carrying just 40% ferric oxide by mass against their 70%, because dispersed nanoparticles expose vastly more binding surface.5
The platform generalizes: the same architecture has been validated in the peer-reviewed literature for phosphate4, nitrate26, and — at Lake Isabella, California — uranium and arsenic removed concurrently below their US regulatory limits for more than 30,000 bed volumes.12
04 — The Horse · Endurance
Selectivity, measured in bed volumes.
Because the hands only grab the bad stuff, one charge of resin treats a small lake before it needs a wash.
The field literature converges on a striking number: community units in West Bengal produce on average about 1,000,000 liters — 10,000 bed volumes — of safe water per cycle before arsenic approaches the local limit.3 With HIX-Nano the figure is higher still: in the 24-month monitored Bangladesh–India study, the resin approached exhaustion only after roughly 20 months of continuous service and 25,000 bed volumes.10 A single column at Nabarun Sangha, Ashok Nagar produced 6.5 million liters of safe water between 2004 and 2013.5
Live demo — the column, and the bead inside it
Two rows, one per medium — HIX-Nano on top. In each row, raw groundwater — arsenic, iron and fluoride mixed with harmless minerals — pours into the FRP column (left), packed with millions of beads (a 1,000 GPH plant carries a 200-liter HIX bed; utility sizes run to 2,000 L28), and the dashed lens magnifies one live bead from that bed beside it. Watch the exhaustion front move down each column: alumina spends its capacity on harmless ions and breaks through, while the monitored HIX-Nano column is regenerated in place before anything breaks through — cycle after cycle.10 The magnified beads show why: HIX-Nano's "hands" grab only the toxic ions (sulfate at 500 mg/L showed practically no competition6), while alumina grabs whatever touches it.
Arsenic / fluoride ionsHarmless ions
The columnsHIX · regenerations0effluentsafeAlumina · effluentsafe
The beadsHIX · hands on toxics0 / 14regenerations0toxics missed0Alumina · wasted on harmless0 / 14toxics missed0
Behavior stylized for illustration; the selectivity contrast is the cited finding.6 In the field one HIX-Nano charge treats ~25,000 bed volumes over ~20 months before saturation.10
Then it regenerates. A simple brine — 2% sodium hydroxide with 2% sodium chloride — releases the captured ions and restores about 95% of capacity within roughly ten bed volumes of rinse.10,11 HIX-Nano has demonstrated at least five regeneration cycles in the field against one or two for activated alumina10; long-term column records show a parent bed plus three regenerations holding effluent arsenic under 10 ppb out to ~30,000 bed volumes.26
Activated alumina<600 BV
HAIX-Zr · HIX-Nano>1,400 BV
Bed volumes before effluent fluoride crossed the WHO limit of 1.5 mg/L (10 mg/L feed, identical columns). Source: Environmental Engineering Science 31(7), 2014, p.368.6 A commercial hybrid competitor reached 50% breakthrough in under 50 BV in the same test.
Drinkwell's own claim distills the comparison: 10× better arsenic-and-fluoride removal than conventional media, holding 99.9%+ removal amid competing ions.29 In the field the platform runs on three contaminants — arsenic and fluoride on HIX-Nano™, and iron on HIX™ 100+, which carries every arsenic-iron removal plant we operate. The phosphate and uranium results above are laboratory and field-pilot findings from the peer-reviewed literature412; neither has yet been treated in a Drinkwell system, and the media has not been tested against lead.
And the water itself — same stream in, very different water out
Stream width = share of the water. Blue = product delivered for drinking · amber = reject and rinse. Ion exchange passes the whole stream through the bed, losing only backwash and regeneration rinse — a few bed volumes per ~10,000 treated3 (recovery figure: Drinkwell engineering). Published RO field figures are often worse than shown here: membrane rejection up to 60%5, and 50–80% of feedwater discarded as reject.10 Drinkwell's published comparison: 90% recovery vs 40–50% for RO29; the current 1,000 LPH kiosk design specifies 95%.28
The durability extends to survival in hostile water. At pH 12, activated alumina dissolved — releasing over 40 mg/L of aluminum — while HAIX-Zr stayed practically insoluble across pH 4–12.6 In an independent laboratory study for Vestergaard, a 0.51-liter cartridge of Drinkwell's resin delivered 7,429 liters to the WHO arsenic standard on NSF-53 challenge water, and shrugged off an accidental 500 ppb arsenic spike with effluent still under 2 ppb.22 Some village units installed in the 2000s "are performing equally well even after five regenerations."7
05 — The Horse · Accountability
Where the arsenic actually goes.
The question every regulator asks first has a twenty-year, peer-reviewed answer: into a stable solid, watched by monitoring wells.
Regeneration concentrates months of captured arsenic into a few hundred liters of spent brine — the first caustic batch carries up to ~39,600 µg/L of arsenic.7 Adding ferric chloride and neutralizing to pH ~7 precipitates it: across four consecutive regenerations, dissolved arsenic in the settled supernatant fell to 0.092–0.12 mg/L, within Indian standards for open discharge.7 What remains is about 2 kg of solid residual per regeneration7 — versus disposing of roughly 100 kg of exhausted media per village if the sorbent were single-use.3
Live demo — regeneration: every hand lets go at once
The bead arrives saturated from Chapter 04. Press regenerate: a brine rinse — 2% sodium hydroxide with 2% sodium chloride10 — shifts the chemistry so every hand releases, and the captured ions wash into the spent-rinse tray. Nothing vanishes: that small concentrate is precipitated with ferric chloride and stored as a stable solid.7 Then the bead goes back to work.
Cycles0Hands filled14 / 14Ions in tray0
Released arsenic / fluorideBrine rinse
Field figures: ~95% of capacity restored within ~10 bed volumes of rinse10,11; at least five cycles demonstrated10; the tray's contents leave each regeneration as ~2 kg of stable, TCLP-passing solid.7
Residual · Sample S181 µg/L
Residual · Sample S2100 µg/L
Residual · Sample S3135 µg/L
Arsenic leached from dried treatment residuals under the US EPA TCLP protocol; bars are scaled to the 1,000 µg/L hazardous-waste threshold — the worst sample reaches 13.5% of it. Under the milder ASTM D3987-85 protocol the same samples leached only 23–36 µg/L. Source: Journal of Hazardous Materials 271 (2014), Fig. 5, p.307.7
50×
less solid arsenic waste than single-use media or RO — about 210 grams a year for a 6,000-liter-per-day system.7
The circular-economy case — Regenerate · Reuse · Recycle29
100%
Regenerate. Capacity restored on-site every 6–12 months with a simple chemical process — no replacement.
5–10 yr
Reuse. Media lifespan through repeated regeneration, vs ~1 year for single-use resins and membranes.
95%
Recycle. Less spent-media disposal over the lifecycle; jugs and bottles collected, cleaned, returned to service.
90%
Water recovery, vs 40–50% for reverse osmosis.
53%
Lower energy costs than RO systems (~6× less energy).
74%
Lower cost — ₹0.029 per liter, at 1,000 LPH scale.
The long-run evidence is rarer than the lab test: the same study monitored drinking-water wells 7–100 meters from a residual-storage sand filter for nearly four years (2005–2008) and found no upward arsenic trend — the nearest wells held at 0.003–0.023 mg/L throughout.7 Overall, regenerating rather than dumping the media cuts the volume of arsenic-laden solids "by nearly two orders of magnitude."3
Drinkwell's engineering note attached to the study translates this to operations: a 27-cubic-foot on-site concrete enclosure would take 40 years to fill.7 (These engineering figures are Drinkwell calculations appended to, not part of, the cited journal article.)
II
Part II — The Delivery Method
The Harness.
A domestic water filter. A community kiosk. A piped water system. Each with its own engineering, its own economics, and its own proof points.
<$10 / person CapEx~350 Dhaka ATM locations91.5 m³/hr at utility scale
06 — The Harness · Three forms
One horse, three harnesses.
A domestic water filter. A community kiosk. A piped water system. The horse works in all three — and we are honest about how deep the published evidence runs for each.
The three harnesses — design, evidence, and the real thing
One table, three rows: what each harness is, where its evidence stands, the engineering schematic simplified from the drawings28 — the blue vessel marked with the horseshoe is the HIX bed, the horse; everything copper is the harness — and a photograph from the field.
01 · The domestic filter Evidence: building now
A two-chamber household unit
24 L tabletop, 8–10 L/hr, four stages — sediment, HIX™ iron candle, the HIX-Nano™ cartridge, disinfectant.28 Early field results are in: 2,650 bed volumes at Gazna, arsenic to ≤0.008 mg/L, at roughly $0.0015 per liter.27 A fuller evidence program is underway.
02 · The community kiosk Evidence: 20 years, peer-reviewed
Kiosks with Water ATMs
The deepest record in this report: 88 systems analyzed in ES&T, plant lives beyond ten years, and the full economics below.10 Today’s 1,000 LPH solar design dispenses through a card-based Water ATM at 95% water recovery.28
03 · The piped water system Evidence: building now
Utility-scale plants on the grid
PHED plants in West Bengal already run HIX-Nano at 50–91.5 m³/hour26, and the same chemistry has served US municipal systems of more than 20,000 households.5 Auto iron-removal plants scale 1,000–10,000 GPH.28 Systematic performance publication is in progress.
Photos: a Drinkwell household filter in Bangladesh (field archive); the Fakirapool Water ATM booth with Dhaka WASA (field archive); the piped scheme is an illustrative render. Schematics: internal design documents.28
Built for real-world conditions — company design principles29
Locally sourced. Stable, non-hazardous metals — no import dependencies.
Simple to run. Single-stage operation by local labor; no specialized technicians.
Vertically integrated. End-to-end design passes cost savings to communities.
Rural grid-ready. ~6× lower energy than RO, built for outage-prone regions.
Where the harnesses stand — deployments across South & Southeast Asia
Over 200 community systems installed across India, Bangladesh, Cambodia, Laos and Nepal5,15, ~350 Water ATM locations in Dhaka18, and utility plants to 91.5 m³/hr in West Bengal.26 Key sites named in the cited literature:
Stylized map; positions approximate. US deployments — Sahuarita, Arizona and Lake Isabella, California12 — are off this map.
The live deployment map
Synced from Drinkwell's deployment registry. Pan, zoom, and click any site; clusters expand as you zoom in.
The kiosk record, in ledgers
The 2019 Environmental Science & Technology policy analysis examined 88 community arsenic systems across India and Bangladesh — some running up to 20 years — with three plants monitored monthly for two years.10 The headline: capital cost under $10 per person for an 800-person community, with plant life expectancy well over ten years.10 The resin itself is less than 20% of that CapEx.10
Families pay roughly 100 INR (~US$1.50) a month for up to 20 liters a day — about $2.5 per 1,000 liters, or 2% of household income. The local alternative, packaged water, costs ten times more: over $25 per 1,000 liters, a fifth of income.10 On those tariffs, all three monitored plants were operating profitably within 15 months, with combined two-year earnings of 420,000 INR on an average plant cost of 430,000 INR ($6,500).10
10×
the price gap between HIX community water ($2.5/1,000 L) and local packaged water (>$25/1,000 L) — the margin the model lives in.10
The long-run records agree. A North 24 Parganas plant grew annual revenue from 40,000 to 220,000 INR over twelve years as participation reached 500 households.10 Ballia, Uttar Pradesh — built for a minimized 250,000 INR (~$4,000) — banked 477,350 INR of revenue and 183,915 INR of profit from December 2013 to September 2017 while holding treated arsenic below 0.01 mg/L.11 Bangladesh's first community system at Betila, Manikganj (March 2015) had 112 households buying daily within 20 months, at 25 paisa/liter pickup or 50 paisa delivered, clearing 4,000 Taka a month after paying a caretaker and deliveryman.9 A Monte Carlo simulation across 2,000 scenarios put 63% above operational breakeven at 24 months.10
Regeneration is what makes the arithmetic work: regenerating a bed costs about one quarter the price of fresh activated alumina and one sixth the price of fresh HIX resin3, and centralizing it cut plant downtime from two days to two hours.3
07 — The Harness · City scale
Dhaka: the harness at city scale.
With Dhaka WASA, the kiosk grew into a city-wide Water ATM network — the harness's largest proving ground, built in partnership with a major utility.
In Dhaka the enemy is not arsenic in a tubewell; it is a pipe nobody trusts. At baseline, 65% of surveyed households reported problems with their WASA supply — dirty water, iron, smell13 — against the World Bank's finding that 41% of Bangladesh's improved sources carry E. coli.15 So with Dhaka WASA, Drinkwell installed Water ATM booths at the utility's existing pumps, treating and dispensing the utility's own piped supply. Because the challenge here is bacterial safety and trust, this chapter's results are best read as proof of the delivery system and service model: kiosk hardware, RFID payments, caretaker operations, service logistics.
2.19 billion
liters of safe water dispensed by the Dhaka ATM network, Oct 2017 – Feb 2026 — metered at the tap, card by card.31
The harness scaled: from 6 booths in 2017 to 87 by April 2019, serving nearly 27,000 users15; the Global Innovation Fund counts 88 ATM-enabled systems and 49,911 pay-as-you-go cards17; the network has since grown to roughly 350 locations across the two city corporations.18
The operational texture: 20 sites, 80 pumps, 180 booths, each booth able to dispense ~10,000 liters a day16; SIM-based location tracking that cut service response times by ~20 minutes15; more than 400 people employed across Drinkwell's sites.17
The product layer is a card and a price: an RFID card for 50 BDT, water from 0.80 BDT per liter, a 50-liter daily cap.18 At endline, 98% of users were satisfied with the service — 99% with booth locations and caretaker behavior, 95% with cost.14 (The harness experiments that failed are catalogued honestly in section 10.)
III
Part III — Utilization
The Habit.
A carriage line only matters if people keep riding. Safe water only counts if people keep choosing it, keep paying for it, and store it safely once it reaches home.
An independent evaluator followed the same 326 households for fifteen months. The habit formed — and it showed up in health, money, and trust.
The GSMA/FCDO-funded evaluation followed 350 registered users from baseline (late 2017) to endline (early 2019; n=326 after at least six months of service use).13,14 Drinkwell rose from 0% to 69% of users' drinking-water sourcing while use of unauthorized WASA lines fell by 10 points — and 98% of users came to drink Drinkwell water exclusively.14,15 That is the habit, measured. What it changed:
Baseline · 20174.5
Endline · 20190.3
Average number of water-related health issues reported per household. Households reporting no health issues rose from 21% to 90%; 97% reported a reduction in water-related disease. Source: GSMA M4D Utilities endline evaluation, Feb 2019, p.50.14
The evaluation ended in 2019 — but the machines kept measuring. Every card tap is logged, so the habit can be read directly from the transaction record, cohort by cohort:
The habit, card by card — weekly-active cohort retention
First-party evidence from the Water ATM transaction log: 182.7 million dispenses across 802,235 cards, complete through June 2026.30 Each row is a quarterly cohort of new cardholders (first-ever dispense); each cell is the share who were weekly-active — dispensed water in at least 3 distinct weeks of that month — M months after joining. The signal that matters: the month-one habit rate has held at ~64% for eleven straight quarters, through 2.5 years and a quarter-million new cardholders — growth is not diluting engagement. Under the ordinary "bought at least once that month" bar, month-one activity is 83–86%. Roughly two in five still hold the strict weekly habit six months on. And because water is consumed daily, read this as share-of-source, not app-style engagement: a cardholder who isn't weekly-active is meeting that month's drinking needs elsewhere — usually a free but untested alternative. Every green cell is the paid safe source winning a daily choice against a free competitor.
Cohort
New cards
M0
M1
M2
M3
M4
M5
M6
M7
M8
M9
M10
M11
M12
2023 Q4*
243,938
70
64
55
51
52
56
54
56
49
47
46
47
44
2024 Q1
51,377
42
58
54
48
43
38
38
37
36
32
30
29
31
2024 Q2
58,465
44
61
50
45
43
42
38
35
32
32
32
34
34
2024 Q3
42,823
48
64
56
49
43
39
38
37
39
36
36
34
35
2024 Q4
37,246
50
63
52
48
45
45
42
40
38
38
37
35
34
2025 Q1
38,256
47
64
58
51
48
45
44
42
39
36
34
33
32
2025 Q2
46,246
42
64
54
52
48
44
41
37
34
33
33
34
34
2025 Q3
46,176
51
67
57
49
43
39
36
37
36
36
–
–
–
2025 Q4
36,462
51
64
54
48
45
43
43
–
–
–
–
–
–
2026 Q1
33,109
40
64
55
53
–
–
–
–
–
–
–
–
–
2026 Q2
44,621
47
–
–
–
–
–
–
–
–
–
–
–
–
Cell = % of cohort weekly-active (≥3 distinct weeks with a dispense) in month M after first use — a deliberately strict bar; M0 is the partial joining month. "–" = beyond the observable window (data runs Oct 2023 – Jun 2026; cells are shown only where every member of the cohort could have been observed). *2023 Q4 is left-censored: it contains every cardholder already active when complete logging began, i.e. the established base — still ~40% weekly-active more than two years on. Grouping cards by registered mobile number leaves these curves unchanged (one card per registration) — and one collector often fetches for several homes16, so households served per active card exceeds one. Source: Drinkwell/Nybsys dispense log, Dhaka network.30
Three deeper cuts — households, intensity, seasons
Households: grouping the 710,257 cards by registered mobile number leaves the retention curves unchanged — registrations are one card per household, so the heatmap above already reads at household level. Intensity: those who keep the weekly habit don't taper — a weekly-active card draws a near-constant ~295 liters a month (~10 L/day) from month 1 through month 14. The habit doesn't fade; it either holds at full strength or stops. Seasons: demand breathes — winter troughs (Dec–Feb, when the piped supply is at its best) and hot-season peaks repeat in both years, so part of any cohort's "decay" is seasonal switching, not attrition. Beneath the rhythm, the network grew: 159k weekly-active cardholders in Oct 2023 → 233k in Oct 2025 (+46%).30
Liters per weekly-active card · months 1–14 since joining
flat at ≈295 L/month — no fade among the retained
Monthly volume, whole network · Oct 2023 – Oct 2025 (million liters)
O23J24AJOJ25AJO25
winter months — trough ≈42–48 ML vs peaks ≈80–88 ML
The habit paid households back: average weekly water spend fell from 156 to 91 BDT14; trust in the water supply rose from 39% to 89%14; and perception of the utility itself improved for 90% of users — a datum any municipal partner should notice.14 GSMA's published case study summarizes the health arc: respondents reporting water-related health issues fell from 79% to 10% in twelve months.15
The habit is the layer the whole rig is judged by: water people keep choosing, at tariffs they keep paying, stored safely once it reaches home.
09 — Validation
Independent eyes.
None of the load-bearing claims in this report rests on Drinkwell's own testing alone.
Validator
What they tested
What they found
CSIR-IMMT Bhubaneswar (Govt. of India lab, 2019)
HIX-Nano100 & 200 columns, 32 and 28 days continuous, no regeneration19
Arsenic 69–296 ppb in → 1–17 ppb out across 33 runs; fluoride 4.7–9.9 mg/L in → 0.02–0.04 mg/L out across 27 runs
IISc Bangalore (Prof. K. Kesava Rao, 2016)
Fixed-bed fluoride capacity vs activated alumina at 5 mg/L feed20
HAIX-Zr 8.4 mg F⁻/g vs activated alumina 6.2 mg F⁻/g; declared no financial interest
Formal endorsement "for removal of Arsenic from ground water"; notes US Patent 9,120,093 B2 licensed to WIST/Drinkwell
Vestergaard laboratory (Vietnam, 2015)
0.51 L resin cartridge on NSF-53 challenge water, 50 ppb As(III)22
7,429 L delivered to the WHO 10 ppb standard; effluent <2 ppb during an accidental 500 ppb spike
A.N. College Patna (Dr. Ashok K. Ghosh, 2013)
Village plant at Ramnagar, Bihar, running since March 201125
Certified arsenic-safe output to WHO standards for over 2.5 years of continuous operation
GSMA / FCDO (2018–2019)
350-user baseline/endline impact evaluation of the Dhaka ATM network13,14
The section 07–08 results: 98% satisfaction, health issues 4.5 → 0.3, spend 156 → 91 BDT/wk
"By a long way the best of its kind I have seen."Peter Ravenscroft — then UNICEF Bangladesh Water & Sanitation Specialist, on the West Bengal community arsenic-removal program23
10 — The Caveats
What the data does not gloss over.
A hype-free report earns its adjective here. Four findings cut against the grain — each tagged with the layer that is actually failing.
HabitClean at the tap is not clean in the home
An internal study with Danone Communities (SGS Dhaka lab testing) sampled three Dhaka ATMs and fifteen customer households. Every ATM sample was clean (<1.8 MPN/100 mL). But after 48 hours of household storage, only 43% of homes still held safe water; 67% tested positive for both total coliform and E. coli, one as high as 1,600 MPN/100 mL.24 The last meter — storage vessels and hygiene — is a behavioral problem no resin solves, and it bounds what any kiosk model can claim.
HorseThe chemistry has a gate
HIX selectivity is strongest in the near-neutral groundwater typical of the Bengal basin. At Nalhati, West Bengal, the team met the hardest water it had ever tested — pH up to 9.4 with fluoride at 45 mg/L — and while the school system met drinking standards during operation, it had treated only 575 bed volumes at evaluation.11 Extreme-pH, very-high-fluoride water is a different engineering problem, and pretending otherwise would be the fastest way to fail a pilot.
HarnessKiosks die of economics, not chemistry
The published site register is candid about failed units, attributing them to "no collection of money, lack of maintenance"10 — never to the media. Below tariffs of ~30 INR/month, nearly half of simulated systems run at a loss.8 In Dhaka, caretakers estimated 40–60% of registered users lapse into dormancy16, and mobile-money top-ups were abandoned when a 1.5% transaction fee proved prohibitive against water margins.15 The moat is keeping systems alive — which is why the payment layer, the caretaker economics, and the regeneration logistics get as much engineering as the bead.
HorseThe honest comparison with RO
Reverse osmosis works on almost any water — but discards 50–80% of the feed as reject10, needs pumps and membranes, and produces a liquid waste stream rather than a spoonful of stable solid. The HIX case is not that RO is bad; it is that where the chemistry fits, ion exchange delivers the same safety at a fraction of the water, energy, and waste burden.
The layering is the strategy. Capability lives in the horse. Defensibility lives largely in the harness: the sites that died, died of harness failures, while the sites that thrived paired the same chemistry with working economics for a decade or more.10 And the whole rig is finally judged at the habit. Readers from the AI industry will recognize the shape of the stack — model, harness, product — but water has to earn a layer software takes for granted: the habit.
Appendix — Citations
Sources.
Sarkar S., Gupta A., Biswas R.K., Deb A.K., Greenleaf J.E., SenGupta A.K. (2005). Well-head arsenic removal units in remote villages of Indian subcontinent: Field results and performance evaluation. Water Research 39(10):2196–2206. doi:10.1016/j.watres.2005.04.002
Sarkar S., Blaney L.M., Gupta A., Ghosh D., SenGupta A.K. (2008). Arsenic removal from groundwater and its safe containment in a rural environment: Validation of a sustainable approach. Environmental Science & Technology 42(12):4268–4273. doi:10.1021/es702556t
Sarkar S., Greenleaf J.E., Gupta A., Ghosh D., Blaney L.M., Bandyopadhyay P., Biswas R.K., Dutta A.K., SenGupta A.K. (2010). Evolution of community-based arsenic removal systems in remote villages in West Bengal, India: Assessment of decade-long operation. Water Research 44:5813–5822. doi:10.1016/j.watres.2010.07.072
Blaney L.M., Cinar S., SenGupta A.K. (2007). Hybrid anion exchanger for trace phosphate removal from water and wastewater. Water Research 41(7):1603–1613. doi:10.1016/j.watres.2007.01.008
German M., Seingheng H., SenGupta A.K. (2014). Mitigating arsenic crisis in the developing world: Role of robust, reusable and selective hybrid anion exchanger (HAIX). Science of the Total Environment 488–489:547–553. doi:10.1016/j.scitotenv.2013.10.092
Padungthon S., Li J., German M., SenGupta A.K. (2014). Hybrid anion exchanger with dispersed zirconium oxide nanoparticles: A durable and reusable fluoride-selective sorbent. Environmental Engineering Science 31(7):360–372. doi:10.1089/ees.2013.0412
Ghosh D., Sarkar S., SenGupta A.K., Gupta A. (2014). Investigation on the long-term storage and fate of arsenic obtained as a treatment residual: A case study. Journal of Hazardous Materials 271:302–310. doi:10.1016/j.jhazmat.2014.02.015
German M.S., SenGupta A.K., Watkins T.A. (2016). From lab and SocEnt startup to impact on 200,000 lives: A sustainable microenterprise model for village-scale arsenic removal from drinking water. VentureWell OPEN 2016 conference paper.
SenGupta A.K., German M., Chatterjee P., Shaw A., Sarkar S., Watkins T.A., Rahman M., Chowdhury M. (2017). Breakthrough technology or breakthrough solution: What are we really after? Environmental Science & Technology 51(5):2529–2530 (Viewpoint). doi:10.1021/acs.est.6b05540
German M.S., Watkins T.A., Chowdhury M., Chatterjee P., Rahman M., Seingheng H., SenGupta A.K. (2019). Evidence of economically sustainable village-scale microenterprises for arsenic remediation in developing countries. Environmental Science & Technology 53(3):1078–1086. doi:10.1021/acs.est.8b02523
German M.S., SenGupta A.K. (2019). Transforming the global arsenic and fluoride crisis into an economic enterprise: Role of HAIX-Nano in Ballia, Uttar Pradesh and Nalhati, West Bengal. Ch. 13 in Advances in Water Purification Techniques, Elsevier, pp. 327–350. doi:10.1016/B978-0-12-814790-0.00013-2
SenGupta A.K., German M., Chen H. (2021). Transforming a global water crisis into an economic opportunity: Unmet needs and lessons learned during the last two decades. Journal of Environmental Engineering (ASCE) 147(12):02521002. doi:10.1061/(ASCE)EE.1943-7870.0001934
GSMA M4D Utilities (2019). Drinkwell endline findings — M&E evaluation, Bangladesh, February 2019 (n=326, ≥6 months of service use).
GSMA (2019). Drinkwell: Building a network of purified water ATMs in Dhaka. Mobile for Development Utilities case study.
Quantum Consumer Solutions for GSMA (2019). Project Water ATM: Findings from a qualitative study evaluating Water ATM services in Dhaka, March 2019.
Global Innovation Fund (2019). Impact Report 2019, Drinkwell profile, pp. 37–38.
Ahmed J.U. (2025). Water ATM booths: Is access to drinking water unequal? Sage Business Cases. doi:10.4135/9781071979785
CSIR-IMMT Bhubaneswar (2019). Laboratory scale studies on arsenic and fluoride removal efficiency individually on supplied HIX-Nano100 & HIX-Nano200. Report No. 1129/E&S/SSP-347/June/2019.
Rao K.K. (2016). Validation letter, Dept. of Chemical Engineering, Indian Institute of Science, Bangalore, 12 September 2016.
Nath K.J. (2018). Letter of the Arsenic Task Force, Government of West Bengal, 24 February 2018 (re: US Patent 9,120,093 B2).
Vestergaard Water Laboratory (2015). Performance of arsenic removal ion exchange resin from Drinkwell. Study LSF.13.1003.8, Vietnam, 24 March 2015.
Ravenscroft P. (c. 2013). Endorsement letter (UNICEF Bangladesh Water & Sanitation Specialist, 2011–2013).
Drinkwell / Danone Communities (2025, internal). Water quality study at the end user level, Bangladesh. SGS Dhaka laboratory testing; 3 ATMs, 15 households, 33 samples at 24 h and 48 h storage.
Ghosh A.K. (2013). Validation letter, A.N. College Patna, 26 November 2013 (Ramnagar Village plant, Maner Block, Patna).
Chatterjee P.K. (2018). Mitigation of groundwater arsenic crisis: Role of HIX-Nano media. Presentation to WSSO, Public Health Engineering Dept., Govt. of West Bengal, 27 November 2018.
Haldar A. (2026). A sustainable solution for domestic arsenic and iron removal using hybrid cation and anion exchange (HIX) nano media. National WASH Conference 2026 (SPM-NIWAS), paper NWC2026-P16, pp. 19–21.
WIST Water Solutions (2026, internal design documents): Tripura PHED Auto Iron Removal Plant DPR (1,000–10,000 GPH; HIX bed sizing 200–2,000 L; EBCT ~3.2 min; media good for 10–12 regeneration cycles); CMF Rajasthan 1,000 LPH HIX-Nano fluoride plant estimate (95% water recovery, solar-powered, Water ATM package); 24 L tabletop iron/arsenic/fluoride filter specification.
Drinkwell (2026). HIX-Nano™ technology page, drinkwell.com/hix — company figures at 1,000 LPH scale: 90% water recovery vs 40–50% for RO; 53% lower energy cost (~6× lower energy use); 74% lower cost (₹0.029/L); 5–10 year media life via 6–12-month regenerations; 95% reduction in spent-media disposal; 10× selectivity vs conventional media; 99.9%+ removal; 700+ systems, 3M+ people served, 2.5B+ liters delivered since 2015.
Drinkwell / Nybsys (2026, internal). Card-level Water ATM dispense log, Dhaka network — 182.7 million dispense transactions across 802,235 cards, complete coverage Oct 2023–Jun 2026 (earlier fragments excluded). Cohort analysis July 2026; weekly-active = dispensed in ≥3 distinct weeks of a calendar month.
Drinkwell / Nybsys (2026, internal). Pump-level monthly dispense log, Dhaka Water ATM network — 2,194,646,185 liters dispensed across all pumps, Oct 2017 – Feb 2026.
