The first social web found a powerful economic bargain: users paid no entry fee, advertisers paid for access to attention, and platforms turned observation into prediction so the two sides could be matched at enormous scale. The next social internet faces a harder test. Can it finance creators, local businesses, safety, infrastructure and innovation without turning every ordinary action into advertising intelligence? Softa Technologies, under Sunil Kumar Singh, is trying to answer that question before the AI and deepfake era makes it unavoidable.
The invoice hidden inside free
For most people, the free internet does not feel like a marketplace when they enter it. There is no cashier at the door of a social network, no meter running while video loads, no invoice after a conversation, no subscription charge attached to each photograph, community post, search, group call or creator clip. The apparent absence of price was one of the great accelerants of the consumer internet. It allowed billions of people to communicate and publish without first deciding whether connection was worth a monthly bill. It allowed new creators to reach audiences before they had businesses, families to remain connected across borders, students to learn outside formal institutions, small merchants to become discoverable, and entire cultural conversations to move onto infrastructure that felt almost frictionless. But infrastructure is never free merely because the person using it does not receive the bill. Data centres consume capital and electricity. Video costs money to store and deliver. Fraud, safety, support, moderation, identity abuse, software development, legal compliance and disaster recovery all require institutions and people. Someone has to pay. The first great social platforms answered that problem with advertising, and then made advertising dramatically more efficient by learning more about the people who might see it. What began as a way to finance free access became an architecture in which the economic value of the service was increasingly tied to the quality of its predictions about human attention.
That history matters because any serious alternative must begin by admitting what behavioural advertising solved. It did not become dominant merely because technology companies were careless about privacy or because advertisers were irrationally fascinated by personal data. It solved a coordination problem. A publisher had attention but did not always know which advertiser valued it most. An advertiser had a product but did not know which anonymous person was most likely to buy. Behavioural systems reduced that uncertainty by observing actions over time, grouping signals, building profiles, predicting interests, testing outcomes and feeding the result back into the next campaign. The more accurately the system could estimate intent, the more confidently money could move. Advertisers could pay for narrower audiences. Platforms could charge for measurable performance. Publishers and creators could be financed indirectly. Users could keep seeing a zero-price interface. That was a formidable economic machine. The problem is not that it created no value. The problem is that its economic intelligence and its human intelligence became difficult to separate. A system built to answer “which advertisement is relevant now?” gained incentives to answer “what kind of person is this, what may this person want next, what will hold this person’s attention, what vulnerability or aspiration can be inferred, and how can the next interaction improve the model?” The commercial question widened into an institutional theory of the individual.
The distinction is becoming more important in the AI era because prediction is no longer limited by the same costs that shaped the earlier web. Machine learning can combine weak signals that human analysts would never have joined. Generative systems can produce thousands of creative variations, messages, images and offers at low marginal cost. Synthetic media can make a false endorsement look real enough to travel before correction catches it. A profile that once existed mainly to choose between advertisements can become one input into content selection, automated persuasion, dynamic pricing, fraud scoring, recommendation, political messaging, or decisions about which version of reality is most likely to keep a person engaged. None of those outcomes is inevitable, and not every personalised service is abusive. But the economic incentive has changed. When better prediction can increase both advertising efficiency and the ability to shape attention, the boundary between useful relevance and behavioural control becomes a governance question. The future of the free internet therefore cannot be discussed only in the language of privacy settings. It has to be discussed in the language of business models, because the institution will eventually optimise for whatever pays its bills.
Behavioural advertising was an economic solution before it became a rights problem
The phrase behavioural advertising can sound abstract, but the mechanism is concrete. An individual searches, watches, pauses, likes, follows, visits, purchases, moves between devices, returns at certain times, opens one type of content after another, and leaves a pattern. Some of those signals are supplied deliberately. Others are produced as a side effect of participation. The pattern is then used to infer likely interests or characteristics and to make the next commercial decision more specific. European data protection authorities describe behavioural advertising in essentially these terms: observation over time, profile creation, ad delivery based on the resulting profile, and analysis of the response. The value of the system lies in the loop. Observation improves targeting. Targeting produces outcome data. Outcome data improves the model. The loop can be commercially elegant even when no employee ever reads a person’s private life in plain language. The machine does not need a biography written in sentences. It needs enough correlated signals to make the next prediction more profitable than chance. That is why regulation is increasingly touching the economic core rather than only the privacy notice. The European Union’s Digital Services Act prohibits targeted advertising to minors and prohibits targeting based on profiling that uses sensitive categories of personal data. It also requires greater transparency about advertisements and gives users of the largest services access to non-personalised feed options. The European Data Protection Board, in its work on “consent or pay” models, has gone further in one important direction: it has said large platforms should consider an equivalent free alternative that does not use behavioural advertising, including forms of advertising that process less or no personal data. These rules do not abolish targeted advertising, and Europe is not the whole internet. They do, however, reveal a structural shift. Regulators are no longer treating the advertising layer as economically neutral plumbing. They are asking whether a person’s access to digital participation should require acceptance of broad behavioural processing, whether some forms of profiling are disproportionate, and whether the market can offer relevance without demanding the same depth of personal observation.
The difficult part begins after the principle is stated. It is easy to declare that a platform should not profile people. It is much harder to finance high-bandwidth video, messaging, storage, security, grievance systems, creator income, local support and software development after deliberately refusing one of the most efficient monetisation systems the consumer internet has built. Subscription can solve part of the problem, but not for everyone and not in every market. Transaction fees can finance commerce but not necessarily public conversation. Donations can support communities but rarely infrastructure at continental scale. Enterprise services can cross-subsidise consumer products, but they can also distort priorities. Advertising remains attractive because it can distribute the cost across businesses that already want visibility, allowing individuals to participate without a direct access fee. The serious question, then, is not advertising versus no advertising. It is whether advertising can be redesigned so that the useful commercial context remains while the permanent behavioural dossier does not become the price of admission.
The alternative needs a profit and loss account, not a slogan
Any platform that proposes advertising without behavioural tracking has to answer at least five economic questions. First, how does an advertiser decide whom to reach if the platform refuses to build an enduring profile of each person? Second, how does the platform measure whether the advertisement worked without quietly recreating the same identity graph through attribution? Third, how does it prevent fraud when it intentionally collects less behavioural history? Fourth, how are creators paid in a way that rewards value without turning every creative decision into a contest for predicted engagement? Fifth, can the system generate enough revenue per unit of infrastructure to remain solvent when competitors are willing to extract more information and sell more precise prediction? Privacy does not repeal accounting. A rights-first platform that cannot pay for its servers will eventually fail users just as surely as an extractive platform can fail them through overreach. The strongest alternative therefore starts with a different unit of relevance. Behavioural advertising asks, in simplified form, “what do we know about this person that predicts a response?” Contextual advertising asks, “what is happening here that makes this message relevant?” The difference looks small until it is translated into architecture. Context can include the subject of a public page, the category of a clip, the language of a community, a district selected for local discovery, a search term entered for a specific task, the type of event being viewed, the time sensitivity of an offer, or a merchant category relevant to the visible environment. None of those signals has to become a permanent theory of the individual. A person reading about home renovation can see an advertisement for building supplies because the page is about renovation, not because the platform has concluded that the person is a homeowner, has inferred income, has combined browsing history from other services, and expects a purchase within fourteen days. A user looking at restaurants in a selected district can see nearby dining promotions because of the chosen place and present task, not because the platform has retained months of movement history.
That does not make contextual advertising automatically harmless. Context can itself be sensitive. A page about infertility, debt, religion, domestic violence, addiction or political dissent can reveal more than an advertiser should be allowed to exploit, even if no durable profile is created. Location can become surveillance if a selected district silently turns into a permanent resident category. Language can become a proxy for ethnicity or religion in some settings. Search terms can expose intimate needs. A privacy-preserving advertising model therefore requires not just less data, but a hierarchy of permitted and prohibited contexts. Some contexts may be useful for relevance and still be inappropriate for monetisation. The platform has to decide that the fact it can classify an environment does not mean it should sell it. The economic model becomes credible only when restraint is expressed as a technical rule, a commercial rule and an auditable rule at the same time. It should also be visible to the advertiser. A merchant should not be able to pay more to cross a boundary that the platform has defined as non-commercial, because a privacy rule that disappears under auction pressure is not a rule at all.
Context is not a weaker form of intelligence
One reason behavioural advertising became so dominant is that the industry learned to equate intelligence with personalisation. If the system knows more about the person, the reasoning goes, the advertisement must be smarter. But commercial intelligence can exist at several levels. A neighbourhood pharmacy does not need a psychological model of everyone within five kilometres to know that seasonal allergies are increasing. A restaurant does not need a lifetime interest graph to advertise a lunch menu near offices at noon. A language tutor does not need to know a family’s private messages to reach people looking at education content in the relevant city. A local event organiser may need language, location, category, date and availability, not a prediction of personality. In many transactions, the merchant’s problem is not “who is secretly persuadable?” It is “who is in a relevant context where this offer is useful?” The second question can still support sophisticated optimisation, but the optimisation is attached to the environment, inventory, merchant, creative and campaign rather than to an enduring personal shadow.
This changes the shape of the data science. Instead of maximising the accuracy of an individual profile, a contextual system can maximise the quality of classification, local availability, campaign integrity, price discovery, category matching, frequency control and aggregated outcome measurement. It can learn that a certain type of public content in a certain language produces useful engagement for a certain business category without needing to remember which named or pseudonymous individual responded six months ago. It can optimise which creative works in a district without turning the respondent into a segment that follows them into unrelated contexts. It can identify fraudulent bursts at campaign level, device-risk level or transaction level without converting every ordinary user into a continuously scored commercial object. The system can still be technologically demanding. In some ways it is more demanding because it must achieve economic relevance under deliberately tighter informational constraints. Those constraints can also create better questions. If an advertiser cannot rely on an invisible profile, the advertisement itself has to become more relevant. The offer has to fit the place, language, price, timing and visible need. The merchant has to understand what value is being offered rather than assume the platform’s prediction engine will find susceptible people. Measurement has to distinguish attention from business outcome. Creative quality matters more because the targeting system is not compensating for a weak proposition with ever narrower behavioural selection. This is not a nostalgic return to static banner advertising. It is a different allocation of sophistication. The intelligence moves away from constructing the person and toward understanding the transaction, the context and the market.
The auction should price opportunity, not identity
Digital advertising is not only a targeting system. It is a market-clearing system. Inventory appears, advertisers compete, budgets are paced, prices are adjusted, campaigns are ranked, frequency is controlled, fraudulent demand is filtered and results are fed back into future bids. Behavioural data became powerful partly because it improved this auction. A platform could tell an advertiser not merely that an impression existed, but that the impression belonged to a person the system considered unusually likely to respond. The predicted person became part of the asset being priced. A contextual market has to remove that component without making the auction blind. It can still price scarcity, location, language, subject, merchant category, time, content quality, commercial intent and historical campaign performance. It can still learn that lunch inventory near an office district is more valuable at noon, that a regional-language education page has strong demand from tutoring services, or that a verified local festival produces temporary demand for transport and hospitality. The market remains intelligent, but the intelligence is attached to the opportunity rather than to a hidden estimate of the individual’s psychology. That is a profound change in what the platform treats as property. The valuable object is the context in which commerce may be useful, not the private person standing inside it.
This also changes who can compete. Behavioural advertising rewards institutions that possess enormous cross-context datasets, because a larger observation graph can improve prediction and attribution. Data accumulation therefore becomes both a monetisation tool and a moat. A smaller platform may have excellent communities and still struggle because it knows less about users than an incumbent with years of behavioural history. Contextual markets can reduce some of that advantage by making high-quality public context, local inventory, merchant verification, language competence and transparent measurement more important than historical surveillance depth. They do not eliminate scale effects. A platform with more users still has more inventory and more opportunities to learn which contexts produce value. But the competitive asset is less dependent on possessing the most complete biography of the person. That matters for new entrants, regional platforms and local businesses because it shifts competition toward service quality, market understanding and trust rather than toward who can accumulate the largest invisible dossier. It also makes interoperability less threatening: a merchant can compare outcomes across contextual markets without handing each market the same cross-service identity map.
A mature contextual auction would also need an explicit sensitive-context firewall. Some inventory should simply be unavailable for commercial bidding regardless of price. A person reading about domestic abuse, severe illness, debt distress, sexual orientation, religion, political persecution or a child’s vulnerability may be in a context that is highly predictive of commercial response. That is exactly why the platform should refuse to monetise it. The strongest privacy architecture does not ask only whether the signal is personal data. It asks whether the power created by using the signal is proportionate to the transaction. This principle would make the advertising system less efficient in the narrow sense that it deliberately discards profitable opportunities. It could make the institution more legitimate in the broader sense that users know some parts of human life are not for sale. In the long run, the ability to define and enforce non-commercial zones may become as important to digital trust as encryption itself. The test is whether those exclusions are coded into campaign eligibility and review, not left to an advertiser’s discretion or to a policy page that the auction engine never sees.
The five-kilometre economy is larger than it looks
The global internet trained businesses to admire scale that can be economically irrelevant. A local bakery can have fifty thousand followers and still fail because too few live within delivery range. A plumber can accumulate video views across a country while remaining invisible to households ten minutes away. A tuition teacher can become globally discoverable but commercially dependent on a small radius around a school district. A woman running a home-based food business may not want mass exposure at all. She may want thirty reliable customers in a carefully defined service area without making her private household part of the advertising product. For a large share of real-world commerce, the economically meaningful audience is not the largest audience. It is the right audience within a practical geography. South Asia makes this especially important because the region’s digital economy contains millions of enterprises whose economics are intensely local: restaurants, clinics, repair services, tutors, salons, retailers, tradespeople, transport providers, small manufacturers, neighbourhood professionals, home businesses, local events and district-level services. Many still rely on word of mouth, messaging groups, local signage, marketplace intermediaries or paid digital campaigns that were designed for advertisers with more data, expertise and budget. A hyperlocal advertising system could reduce that gap if it can let a merchant buy relevance in understandable units: a district, category, language, time window, nearby community or visible topic. The merchant does not need access to a person’s inferred fears, income, politics, health or relationship status. It needs a credible path to local demand. This is the economic terrain behind ZHAN, the contextual and hyperlocal advertising layer Softa Technologies is developing alongside ZKTOR rather than presenting as an excuse to import behavioural profiling into the social product. Softa’s public material describes advertising relevance in terms of location, language and context rather than predictions about individual behaviour. That proposition deserves to be judged as a commercial architecture, not merely a privacy promise. A serious ZHAN-type system would need advertiser verification, category rules, scam detection, sensitive-context exclusions, transparent reasons for why an ad appeared, local price controls, campaign-level fraud analysis and reporting that helps a merchant understand performance without giving the merchant a dossier on identifiable viewers. If it succeeds, the product is not simply “ads without tracking.” It is a market mechanism for turning local context into commercial visibility while limiting what either the platform or advertiser is entitled to learn about the person.
Measurement is where privacy promises usually meet financial pressure
Targeting attracts most public attention, but measurement is often the harder privacy problem. An advertiser does not merely want an advertisement shown. It wants to know whether the advertisement caused something valuable: a visit, enquiry, registration, purchase, booking, download, repeat order or other outcome. The more precisely platforms connect exposure to conversion, the more confidently advertisers can spend. That is why attribution systems are economically powerful. They convert uncertainty into budgets. But they can also rebuild the identity graph from the opposite direction. A platform may claim not to profile users for selecting ads while still joining exposure histories, device identifiers, purchase records and partner data in order to prove conversion. If the measurement layer can reconstruct the person, the privacy distinction collapses even if the targeting screen looks clean. A non-profiling model therefore needs measurement rules that are as strict as targeting rules. One approach is aggregation: report how many people in a campaign context took an action without exposing who they were. Another is short retention: keep the minimum campaign state required to prevent duplicates or measure outcomes, then discard it instead of converting it into a permanent audience asset. Another is thresholding: refuse reports so granular that a small group can be reverse-engineered. Conversion can be recorded at merchant, campaign, district, creative or time-window level rather than as a lifetime user attribute. Some measurement can occur on-device or through privacy-preserving cryptographic techniques where scale justifies the complexity. The objective is not zero data. Commerce cannot function with zero evidence. The objective is to prevent temporary evidence of a transaction from becoming a durable commercial identity. This is also where honesty matters more than elegant terminology. A contextual advertising platform may still need frequency control so the same person is not shown the same promotion fifty times. It may need anti-fraud signals to distinguish a real local enquiry from automated activity. It may need payment records, merchant records, security logs and limited technical identifiers. Those functions should not be hidden behind claims that the system “knows nothing.” The stronger standard is purpose limitation. A technical signal collected to prevent ad fraud should not quietly become a lifestyle segment. A payment record required for accounting should not train the social feed. A grievance report should not become an advertiser category. A district selected for one search should not silently become a permanent home location. Privacy is not the absence of processing. It is the discipline to stop useful processing from expanding into unrelated institutional power.
Creators need a revenue system that does not make manipulation their job
Advertising does not finance only platforms. It finances culture. The creator economy has allowed individuals and small teams to build businesses around video, education, entertainment, commentary, craft, music, local expertise and community leadership. Any proposal to reduce behavioural advertising has to explain what happens to those creators. A platform that protects users beautifully but leaves creators economically stranded will struggle to retain the people who make public participation valuable. The creator problem is therefore not peripheral to privacy. It is one of the tests of whether a less extractive internet can sustain the same breadth of expression. The first-generation attention economy often bound creator income to the same optimisation systems that profile audiences. Creators learn quickly what the machine rewards because reach determines revenue. That does not mean every viral creator is manipulated or every recommendation system is harmful. It means economic incentives can narrow creative behaviour. Thumbnails become more extreme, emotional pacing becomes more aggressive, controversy becomes a discovery strategy, and creators can feel compelled to publish for an opaque ranking system rather than for the people they intend to serve. If a platform’s advertising engine depends on maximising time spent and predicting response, creator economics can inherit the same objective. The creator is not merely producing culture. The creator is helping produce behavioural inventory.
ZKTOR’s creator proposition is interesting precisely because Softa links monetisation to a platform that says it does not use behavioural tracking or engagement-driven virality as its organising principle. Public reporting around the project has described a 70 percent creator-share proposition in applicable monetised contexts. That headline is attractive, but the percentage is not the real standard. The denominator is. Seventy percent of what revenue, after which deductions, under what eligibility rules, attributable through which measurement method, adjusted for which invalid activity, paid on what schedule, and visible through what records? A high share can still be opaque if the creator cannot understand the base. A lower share can sometimes be more valuable if attribution, rights and payment are clear. The mature version of creator sovereignty is therefore not a number on a poster. It is an auditable chain from original work to eligible revenue to calculation to payment, without requiring the creator to surrender audience privacy in exchange for proof. A contextual advertising model can alter that chain in a useful way. A creator can be valuable because the content attracts a meaningful context, not because the platform has built intimate profiles of the viewers. A regional-language education channel can attract education advertising because of subject and audience context. A district food creator can attract restaurant campaigns because of locality and category. A technology reviewer can attract relevant technology campaigns because of content, not because the platform has inferred the household income, family composition and private browsing history of every viewer. The creator still needs reach, but reach does not have to mean behavioural extraction. Audience sovereignty in this model does not mean that the creator owns followers as data. It means the creator’s relationship with an audience, original work and monetisation is not entirely mediated by a hidden prediction engine.
AI makes the advertising integrity problem more dangerous, not less
Generative AI lowers the cost of legitimate advertising creativity. A small merchant can translate copy, generate variations, resize media, draft product descriptions and produce campaign concepts that once required a large agency. That can democratise commercial communication. The same tools also lower the cost of deception. A scammer can produce hundreds of plausible promotions, synthetic testimonials, cloned voices, fabricated endorsements and local-language variants with speed that manual review systems were not designed to handle. Advertising integrity in the AI era is therefore not merely a privacy problem. It is an authenticity problem. A platform that reduces profiling but allows synthetic fraud to flood the commercial layer has not created a safer market. This is where Softa’s broader architecture becomes relevant. HOLA AI is described by the company as a bounded intelligence layer for harm prevention, language and knowledge functions rather than an engine for behavioural targeting. ZKTOR’s safety model uses automated detection in narrow harm categories with human review for consequential decisions. The principle matters beyond moderation. A future contextual advertising network could use AI to inspect the advertisement rather than psychoanalyse the audience: detect impersonation, suspicious claims, manipulated media, mismatched merchant identity, prohibited categories, repeated scam patterns or signs that a public figure’s face or voice is being used without authority. The direction of intelligence is different. Instead of asking AI to become better at predicting the person, the system asks AI to become better at verifying the commercial object and reducing harm. No automated system can prove truth from pixels alone. Deepfakes can evade detectors. Authentic media can be mislabelled. Fraudsters adapt. A credible advertising system therefore needs provenance, merchant verification, complaint routes, human escalation and clear liability as well as models. No URL Media architecture, which ZKTOR uses to reduce ordinary external extraction and redistribution of user media, also matters indirectly because creators’ original faces, voices and footage become harder to harvest through the easiest public-link workflows. It does not make copying impossible, and screen capture or other attack paths remain. But the principle is significant: the social platform should not unnecessarily increase the supply of material that can later be weaponised in synthetic endorsements. Economic integrity begins before the advertisement is purchased.
Women and children reveal whether the economics are actually different
A platform can claim that its advertising is less invasive and still reproduce harm if participation itself requires exposure. Women, children and vulnerable users make this visible because the cost of unwanted visibility is not evenly distributed. A woman who runs a home-based business may need customers without wanting strangers to infer her household, routine or family relationships. A young creator may need discovery without becoming an easy source of downloadable media for impersonation. A teenager may participate in a community without being classified into a commercial identity that follows into adulthood. If the business model treats every increase in visibility as an opportunity to collect more signals, economic participation and personal safety can come into conflict. ZKTOR’s women-first safety emphasis, its Women and Children grievance focus, creation-time visibility controls, human review pathways and No URL Media approach are therefore not separate from advertising economics. They define who can safely enter the market. The right to be online with privacy, data safety and dignity has an economic dimension, as does the related right to visibility without profiling. A person who cannot participate without accepting surveillance, harassment risk or permanent profiling is not participating on equal terms. A small business owner who must disclose more of her private life to become discoverable is paying a different price from a large corporation buying reach. A child whose interests are turned into a permanent advertising profile is effectively financing free access with a future self they are too young to negotiate. This is one reason regulatory restrictions on targeted advertising to minors matter beyond child protection. They test whether digital markets can function when a commercially valuable category of behavioural prediction is deliberately forbidden. If the answer is yes, the lesson is larger than childhood. It suggests the industry can build monetisation around context, content, declared needs, subscription, transactions and other mechanisms when law or design removes a profiling shortcut. A rights-first platform can take the same discipline further voluntarily, not because all advertising is suspect, but because some kinds of knowledge create power that is disproportionate to the transaction being financed.
ZKTOR is the reference architecture, not the whole economic system
Softa’s most consequential claim is easy to misunderstand if ZKTOR is treated as just another social media application. The platform itself combines clips, long-form channels, feeds, pages, clubs, buddy connections, search, local discovery, secure messaging and calling. Its company-described architecture is framed around Privacy and Data Safety by Design, zero behavioural tracking, no profiling, a no-pushed-algorithm direction, user-controlled visibility, No URL Media, Zero Knowledge Server Architecture, device-to-device and end-to-end protection for private communication, OTP-based access, rotating-key concepts, region-bound data handling, country-oriented infrastructure and disaster-recovery separation, and safety without surveillance. Softa also describes a dedicated Women and Children grievance focus and a HOLA AI visual-detection layer intended for narrow harm prevention with human review for consequential action. Taken together, those choices create a reference environment in which a different commercial model can be tested. They do not by themselves solve the economics, and each security or sovereignty claim still needs evidence appropriate to its technical scope. ZKTOR still needs revenue, creators still need income, advertisers still need results and infrastructure still has to be financed. The more ambitious idea sits in the separation between functions. Private communication should not become advertising data. Safety evidence should not become a commercial segment. Search should not silently feed a social profile. Commerce should not automatically export purchase identity into content ranking. Advertising should not require access to encrypted conversations. This can be understood as a Purpose Firewall: each system receives the information required for its task but does not gain a general licence to reuse that information everywhere else. The doctrine is only meaningful if the technical architecture, internal permissions, logs and business incentives enforce it. But if enforced, it changes the economics of integration. The value of an ecosystem comes from interoperability and shared utility, not from collapsing every activity into one behavioural database. That is why Softa’s adjacent projects matter even when they are not current ZKTOR features. ZHAN is being developed as the contextual advertising layer. EZOWM is positioned as a hyperlocal commerce layer for local merchants. KHOZU is being developed around evidence-first search and source-backed answers. SNDESH extends the privacy proposition into private messaging with message-level and rotating-key concepts. HOLA AI is framed as bounded intelligence for safety, language and knowledge rather than behavioural extraction. ZWATO is intended for business administration. Subkuz operates as an editorially independent hyperlocal and diaspora news system. The significance is not that every product is finished or that the ecosystem has already proved commercial scale. The significance is architectural: social identity, private communication, advertising, commerce, search, AI, business operations and journalism are being conceived as connected but purpose-separated layers rather than as excuses to make one central profile progressively more omniscient.
Sunil Kumar Singh’s bet is really about timing
Under founder and chief architect Sunil Kumar Singh, Softa Technologies has increasingly described itself not as a company building isolated apps but as an institution preparing for a different phase of the internet. That claim is easy to dismiss if read as branding. It becomes more interesting when examined as timing. The first generation of social platforms was built for a world in which the central technical problem was connection at scale. The emerging world has additional constraints: generative AI can reproduce identity, deepfakes can weaponise public media, regulators are limiting certain forms of profiling, people expect stronger data rights, cross-border infrastructure raises sovereignty questions, and local economies still need digital access that does not require becoming globally viral. A system designed now has to solve connection and restraint at the same time. Singh’s recurring phrase, the right to be online with privacy, data safety and dignity, is therefore more than a privacy slogan if the economic layer follows the same logic. Rights survive only when the revenue system does not constantly create pressure to reverse them. A platform can promise not to profile users today and still drift toward profiling tomorrow if growth, investor expectations or advertiser demand make personal prediction the easiest route to higher revenue. Softa’s no-VC history and long-horizon capital posture are relevant here only to the extent that they affect incentives. Independence is not proof of virtue, and self-funding does not prevent bad decisions. But capital structure can determine how much time a company has to test a lower-extraction business model before being forced to maximise familiar metrics.
The stronger interpretation of Singh’s project is not that Softa has already built the final internet the world will use. That would be impossible to substantiate. It is that the company has assembled, earlier than most small technology institutions, a coherent set of design responses to problems that are becoming harder to ignore: behavioural profiling, synthetic identity abuse, unsafe media redistribution, opaque algorithmic reach, centralised cross-border data pooling, local-business discoverability, creator value distribution and the need for AI that operates under defined authority. ZKTOR is the most visible reference architecture because it puts many of those constraints into one live social environment. The wider Softa system is the attempt to show how the same principles might extend into the economic, informational and commercial layers of the internet. The claim should be tested, not celebrated by default. But the underlying question is serious: what if the next internet is distinguished less by what it can know than by what it deliberately refuses to know?
A non-profiling market needs a different theory of optimisation
The first generation of digital optimisation often treated more data as a default path to better decisions. The next generation may need to prove that optimisation can be bounded. Campaigns can still be improved, but the objective function has to be narrower. A local merchant may optimise cost per verified enquiry by district, creative, time and category. A creator campaign may optimise attributable revenue by content context and language. A public-service campaign may optimise reach within a declared geography. None of those tasks requires the platform to infer political temperament, relationship status, long-term purchasing personality or a persistent vulnerability score. The platform can become better at the task without becoming more comprehensive about the person. This is a crucial distinction because AI makes over-collection deceptively attractive. If a model can ingest another signal at almost zero marginal analytical cost, institutions need a rule that says technical usefulness alone is not sufficient authority. Otherwise AI turns every available datum into a standing invitation for future reuse, and data minimisation becomes economically irrational inside the organisation even when it remains attractive in public language.
That rule has commercial consequences. Product teams will sometimes have to reject features that could increase yield. Advertisers may ask for audience combinations the system should not create. Measurement teams may want longer retention because more history improves attribution. Fraud teams may want broad device graphs because broad graphs improve detection. AI teams may want unrestricted training data because scale improves models. A rights-first business model needs a governance mechanism capable of saying no to each department for reasons that survive internal financial pressure. This is why the revenue engine requires constitutional design. Data minimisation cannot sit only in a privacy policy written after the product is built. It has to exist in schemas, APIs, access controls, retention schedules, model inputs, merchant dashboards and audit tests. The strongest evidence that a platform believes in non-profiling is not the sentence it publishes. It is the profitable feature it is willing not to build. That discipline is especially important once advertising becomes material to revenue, because privacy promises made before monetisation are much easier to keep than boundaries tested by a growing sales organisation.
The same discipline can improve commercial clarity. If a campaign is bought on context, the advertiser should know which contextual dimensions mattered. If locality is used, the user should be able to understand whether the location came from a selected district, a current coarse region or another permitted source. If language affects relevance, that fact should be explainable. If a creator’s content context produced revenue, the creator should be able to see the attributable category without receiving private viewer data. Explanation becomes part of the market design. The advertiser no longer buys access to an opaque model that claims to know the right people. It buys a defined commercial environment with defined rules. That may initially feel less magical. It is also more auditable, more contestable and potentially more understandable to the small businesses that constitute much of South Asia’s economy. It also creates a basis for price comparison: merchants can decide whether district, language or category inventory is worth the cost rather than trusting a proprietary audience label they cannot independently interpret.
The commercial objection is the strongest objection
The most serious criticism of non-profiling advertising is not ideological. It is arithmetic. If behavioural advertising generates more revenue per user, then a platform that refuses it may have less money for infrastructure, creator payouts, safety teams, product development and customer support. Advertisers may prefer systems that promise more precise conversion. Large brands may demand sophisticated attribution. Small businesses may choose whichever service brings a customer at the lowest cost, regardless of architectural philosophy. If contextual advertising produces materially weaker returns, good intentions can become a competitive disadvantage. A privacy-first network could then face an uncomfortable choice between charging users, reducing service quality, accepting lower margins, cross-subsidising from other businesses or gradually collecting more data. That objection should not be minimised because it identifies the exact experiment ZHAN and similar models must pass. Contextual advertising does not need to beat behavioural advertising at every possible campaign. It needs to be sufficiently effective across enough economically important use cases to fund the system without violating its own constraints. Hyperlocal commerce may be one such use case because physical distance, language, category and time are unusually predictive of relevance. Search-like intent may be another because the person is declaring a present need rather than being inferred from past behaviour. Creator context may be another because the subject of the content itself can align with commercial categories. Public information, events, education, jobs and local services may offer further context-rich markets. The business model can be a portfolio rather than a single targeting trick. The platform also has costs that a surveillance-heavy model may not value correctly. Deep profiling creates compliance risk, breach risk, governance complexity, internal-access risk, user distrust and regulatory exposure. Data that improves monetisation can also increase the damage of compromise. A model that collects less may accept lower gross advertising efficiency while reducing other categories of institutional risk. That does not guarantee superior economics, but it means the comparison cannot be reduced to click-through rate. The relevant unit is durable value after the cost of data governance, security, legal obligations, safety, reputation and user retention. A system that earns slightly less per impression but creates higher trust, safer participation or better local conversion can still be economically rational.
The fraud objection is harder than the privacy slogan admits
Advertising markets attract fraud because money follows measurable events. Bots can manufacture impressions, clicks, installs and leads. Merchants can misrepresent products. Creators can purchase artificial engagement. Agencies can hide mark-ups. Scammers can impersonate brands. A non-profiling system cannot simply collect less and hope malicious actors become cooperative. In some cases, behavioural history helps distinguish ordinary activity from abuse. Removing it can reduce one source of detection intelligence. The answer has to be selective intelligence rather than ignorance. Fraud detection can examine campaign anomalies, impossible timing, repeated device or network patterns, payment risk, merchant history, creative reuse, suspicious conversion clusters and other security signals that are relevant to the fraudulent act. The boundary is purpose. A device-risk indicator used to stop automated click fraud need not become an interest profile. A merchant’s dispute history can be retained for marketplace integrity without being joined to unrelated social activity. A creator payout review can examine invalid traffic without giving advertisers access to follower identities. The system will still make mistakes. What matters is that the evidence collected for integrity remains inside the integrity function unless a clearly justified rule permits otherwise. This is also where human review and appeal become economic infrastructure. If automated fraud detection can freeze a creator’s income or reject a small merchant’s campaign, due process is not merely a rights issue. It is a business requirement. ZKTOR’s existing emphasis on structured review, documented reasoning and appeal for consequential content actions points toward the same standard for monetisation. A merchant should know why an advertisement was rejected. A creator should be able to challenge an invalid-traffic decision. A user should be able to report a deceptive promotion. A platform should publish enough aggregate information for outsiders to understand how often these systems fail. Trust is not the absence of enforcement. It is enforcement that can survive explanation.
Local advertising can create work, but employment claims need discipline
A district-first advertising economy has implications beyond campaign delivery. Local merchants need onboarding, creative support, language assistance, verification, billing, dispute resolution and campaign interpretation. Community organisations need help translating public information into local digital distribution. Creators may need production, licensing and commercial support. Regional grievance systems need cultural and linguistic competence. If these functions are genuinely distributed, they can create professional work outside a small number of metropolitan technology centres. Softa has repeatedly linked its hyperlocal model to district-level operations and youth opportunity. That possibility is credible as a direction but should not be converted into inflated job counts before the work exists. An application, training programme, independent agency, temporary campaign, contractor role and salaried job are different outcomes. The strongest employment claim is therefore not “technology will create thousands of jobs” but that the architecture creates categories of work that can be performed near the communities being served. Whether those categories become durable employment depends on advertiser demand, merchant retention, creator revenue, local operating costs and the platform’s ability to standardise tools without centralising every decision. There is a deeper economic point here. Behavioural advertising tends to centralise value in systems that are excellent at prediction, data aggregation and automated auctioning. Hyperlocal contextual advertising can shift some value toward market knowledge, language, merchant relationships and local operations. That does not make it inherently more equitable. Local intermediaries can also become extractive. But it changes the skill map. The economically valuable knowledge is no longer only “what can the central model infer about the user?” It also becomes “what does this district need, which merchant is credible, which language is appropriate, which offer is actually available, which community context makes the message useful, and how can the result be measured without compromising the person?” That is a different digital labour market.
The ecosystem only works if the walls between products are real
Integrated technology companies are often rewarded for combining data because combination creates convenience and predictive power. A search can improve recommendations. A purchase can improve advertising. A message can reveal intent. A location can improve commerce. An AI assistant can summarise everything. From a pure optimisation perspective, the temptation to connect all signals is enormous. That is exactly why Softa’s broader future-internet thesis will be judged less by the number of products it launches than by the strength of the boundaries between them. If KHOZU becomes an evidence-first search service, a private or sensitive query should not automatically become a ZKTOR advertising segment. If SNDESH provides protected private communication, the content and relationship graph should not become input for ZHAN. If EZOWM processes a local purchase, the transaction should not quietly re-rank a person’s social feed. If HOLA AI assists with language, safety or research, access should be limited to the task and the user-authorised context rather than becoming a general right to ingest the person’s digital life. If Subkuz operates with editorial independence, reader behaviour should not be traded for favourable coverage or fused into a political influence system. If ZWATO handles business administration, enterprise records should remain purpose-bound. The value of the ecosystem depends on cooperation without collapse. That architecture is harder to build than a single central data lake. It requires separate permissions, retention rules, key domains, access roles, audit logs, legal responsibilities and product-specific business models. It can create duplicated engineering and reduce some forms of personalisation. It may make cross-product analytics less convenient. But the friction is not necessarily waste. In rights-first infrastructure, some friction is a safety feature. The same way financial systems separate duties so one employee cannot authorise and conceal the same transaction, digital systems can separate data purposes so one commercial objective cannot silently appropriate every human action. The internet of the AI era may need more internal constitutional design, not merely more intelligence.
The one-million-user test changes the burden
Softa’s latest company reporting places ZKTOR above the one-million-user mark in its South Asian public beta, with participation spanning India, Nepal, Bangladesh and Sri Lanka. That figure should not be confused with one million daily active users, one million paying customers or one million deduplicated cross-product users. It is a scale marker, not a verdict. But scale changes the quality of the questions that can be asked. An architecture that works for a laboratory group may fail under multilingual use, older devices, weak networks, abuse campaigns, cross-border communities, payment disputes, merchant fraud, creator monetisation and the simple unpredictability of public behaviour. A million users generate operational evidence that a prototype cannot. For the advertising thesis, the next meaningful milestones should therefore be economic rather than promotional. How many local merchants can run a campaign without needing specialist agencies? What share return to buy again? Which contextual signals create useful relevance without becoming durable profiles? How are sensitive contexts excluded? What is the fraud rate? How much applicable income reaches creators? How quickly are disputed campaigns resolved? What information can advertisers see about audiences? Can a district-level merchant understand why an advertisement worked? Can the platform demonstrate that private messaging, grievance evidence and protected media never enter the ad system? These metrics would say more about the future of non-profiling monetisation than another download milestone. The same applies to independent audit. Softa has placed privacy and security assessment on its roadmap. The commercial architecture belongs inside that assessment. Auditors should not stop at encryption and server location. They should inspect whether prohibited data can enter advertising pipelines, whether contextual categories are generated from permitted inputs, whether sensitive contexts are monetised, whether conversion measurement can reconstruct identity, whether merchant dashboards expose identifiable users, whether deleted or expired campaign data persists, and whether cross-product joins violate the stated Purpose Firewall. A platform financed by advertising cannot call itself privacy-first if its audit treats the revenue engine as somebody else’s problem.
The revenue engine needs its own constitution
The most dangerous moment for any privacy-first platform may come after success. Early architecture is often built under the discipline of a small team and a clear founding idea. Scale introduces departments, targets, partners, advertisers, creator programmes, agencies, fraud pressures, new countries and quarterly expectations. Each request can appear reasonable in isolation. One team asks for slightly longer retention to improve measurement. Another wants cross-product identity to reduce fraud. Another wants richer advertiser segments. Another wants AI models trained on more interaction data. Another wants merchant exports that make reporting easier. The system rarely abandons its principles in one dramatic decision. It can erode through a sequence of optimisations that each seem too small to threaten the whole. A sustainable non-profiling model therefore needs explicit red lines before revenue pressure becomes strong enough to negotiate them away. Those red lines should have owners, change controls and escalation rules, because principles without institutional custody become vulnerable when the people who originally designed them are no longer in every meeting.
For Softa, those red lines would have to be visible across the ecosystem. ZHAN should not gain private-message data from SNDESH or ZKTOR. KHOZU search intent should not become a durable advertising identity unless the user has made a separate, meaningful choice that survives legal and architectural scrutiny. HOLA AI should not acquire a general right to reuse protected user content because it can improve model performance. EZOWM commerce records should not become hidden determinants of social visibility. Grievance and Women and Children safety evidence should remain outside commercial segmentation. Country-bound data arrangements and the data-sovereignty logic behind them should not be bypassed by a central advertising analytics layer. Creator monetisation records should be sufficient for payment and audit without exposing audience identities. These are not technical footnotes. They are the economic constitution of a system that claims the right to be online with privacy, data safety and dignity. If even one high-value product receives an exception, it can become the bridge through which the entire purpose-separated architecture is slowly recombined.
Independent assessment should eventually test this constitution under adversarial conditions. An auditor should try to trace whether a sensitive context can leak into ad selection, whether a temporary location choice persists after its purpose ends, whether campaign attribution can be reversed into identity, whether internal staff can query protected audience data, whether merchant exports can expose small groups, whether deleted campaign data survives in model features, whether fraud tooling becomes an undeclared profile, and whether cross-product identifiers allow reconstruction of activity the public architecture says is separated. The commercial system should also publish aggregate transparency: rejected advertiser categories, scam removal, appeal outcomes, sensitive-context exclusions, retention rules, creator-payment methodology and material changes to targeting logic. A privacy audit that ignores monetisation is incomplete because revenue is where incentives become architecture. The audit should be repeated after major monetisation changes, not treated as a one-time certificate, because the highest-risk data pathway may be a feature that did not exist when the original assessment was performed.
Who pays, then?
The answer may be more plural than the first social internet trained us to expect. Advertisers can still pay, but they can pay for context, place, category, language, verified intent and public relevance rather than for a permanent behavioural profile. Merchants can pay for local discovery. Commerce can generate transaction revenue where a transaction actually occurs. Creators can receive a transparent share of applicable income. Businesses can pay for administration, professional tools or enterprise services. Some users may choose premium features without making payment a precondition for basic participation. Search can monetise declared commercial intent under strict boundaries. Media can sell advertising around journalism without turning reader identity into the product. AI can be sold as a bounded service rather than as an excuse to absorb every available signal. The portfolio is harder to design than one universal surveillance engine, but diversification can also reduce the institutional pressure to make one data source do everything. This is where the Softa system, if it develops as described, becomes more than a collection of brands. ZKTOR provides the social and communication reference architecture. ZHAN is the proposed non-profiling advertising layer. EZOWM addresses local commerce. KHOZU addresses search and evidence. SNDESH isolates private communication. HOLA AI provides bounded intelligence. ZWATO addresses business operations. Subkuz keeps journalism institutionally distinct. The architecture says that the future internet may be integrated at the level of utility while separated at the level of authority. A person can move through social life, private communication, search, commerce and information without automatically granting one central commercial system the right to build the most complete possible model of them. That is also the deeper meaning of Sunil Kumar Singh’s claim that Softa is building for an internet the world will need rather than merely another social network. The claim is not proven because several products have names, websites or roadmaps. It will be proven only if the working systems can sustain growth, creators, merchants, safety, local relevance and financial durability while keeping the promised boundaries intact. But the direction is now intellectually coherent. AI and deepfakes increase the value of trustworthy identity. Profiling pressure increases the value of data minimisation. Regulatory pressure increases the value of contextual alternatives. Local economies increase the value of district-level relevance. Cross-border digital life increases the value of jurisdiction-aware infrastructure. The right to be online with privacy, data safety and dignity requires all of these layers to meet somewhere in a functioning economy.
The price of the next free internet
The first social web taught users to expect that public digital life should cost no money at the point of entry. That expectation is unlikely to disappear, and in many societies it should not. Direct fees can exclude the people who benefit most from open access. Advertising will therefore remain an important part of the internet’s economic architecture. The question is what kind of advertising, under what limits, with what evidence and with what distribution of value. A free service financed by commerce is not inherently exploitative. It becomes extractive when the institution claims a broader right to understand, predict and influence the person than the transaction requires. The next model may be less obsessed with perfect prediction and more disciplined about sufficient relevance. It may accept that an advertisement can be useful without being uncannily personal. It may measure enough to prove value without remembering enough to reconstruct a life. It may treat a district as a market without turning movement into a permanent map. It may pay creators without forcing them to become technicians of outrage. It may use AI to verify advertisements and protect identity rather than to intensify persuasion. It may keep messaging private even when the same company sells ads elsewhere. It may build local commercial intelligence without converting local life into surveillance. None of this is technically simple. That is precisely why it deserves to be treated as infrastructure rather than branding.
Behavioural advertising was not an accident. It was a rational answer to the economics of a particular internet. The mistake would be to assume it is the only rational answer available forever. The AI era changes the cost of inference, the danger of identity misuse and the social consequences of concentrating predictive power. The deepfake era changes the value of protected originals and authenticated commerce. Data-protection law is narrowing some targeting practices. Creators are demanding clearer economics. Local businesses still need affordable visibility. Users still want free participation. Those pressures are converging on a new design problem. Softa Technologies has chosen to confront that problem from an unusually broad angle. Under Sunil Kumar Singh, it is not merely asking whether a social network can operate without behavioural tracking. It is attempting to connect privacy, data safety, dignity, contextual advertising, creator economics, hyperlocal commerce, protected media, bounded AI, evidence-first search, private messaging, regional data architecture and accountable governance into one future-facing internet arrangement. Some components are live, some are developing, some remain propositions, and all will require independent evidence as they scale. That distinction must remain clear. But if the architecture can survive commercial reality, the result would matter far beyond one Indian platform. The deepest question is not whether people are willing to see advertisements. They always have. It is whether seeing an advertisement must require an institution to know the person behind the screen more deeply than the commercial moment justifies. A free internet will always have a price. The next generation has the opportunity to decide that the price should be money moving between advertisers, creators, merchants and infrastructure, not a permanent claim over the inner life of the people who came only to participate.