Summary
- Dazsoft Desenvolvimento de Softwares Ltda ME is the legal entity behind the Dazsoft site and the Siptek line. Public pages also use “Dazsoft Comercio e Desenvolvimento de Softwares Ltda”; the shared CNPJ, address history, principals and network registration show a naming discrepancy within one business, not evidence of two unrelated suppliers.
- Siptek sits unusually deep in the voice workflow. Its advertised functions cover Class 4 routing, tenant separation, rating and billing, number portability, predictive dialing, IVR, call recording, reconciliation, caller-origin authentication, softphones and AI-assisted transcription. A fault or policy error can therefore travel from a route table to a customer invoice or a consumer’s handset.
- “Support for STIR/SHAKEN” is not enough proof of compliance with Brazil’s Origem Verificada system. A buyer should test certificate custody, authority over calling numbers, attestation, Rich Call Data, failure handling, traceability and delivery across participating carriers and compatible handsets.
- Dazsoft publishes ambitious scale, latency, availability and support statements but little public audit material behind them. Procurement should turn each statement into a measured acceptance test, require an exportable evidence trail and make exit readiness part of the initial design.
A call begins before anyone says hello
The decisive moment in a high-volume call is not the greeting. It is the compressed sequence before the called party hears a ring. A campaign selects a record. A dialer decides whether to place another attempt. A switch checks the origin, the destination, the account and the route. A rating function predicts or applies cost. A carrier accepts or rejects the signaling. An identity service may sign the call. A classifier watches early progress. When a person answers, the system has to connect a free service representative quickly enough to avoid silence.
Later, a call-detail record becomes a supplier charge, a customer bill, a compliance trace and perhaps the input to transcription and quality analysis.
Dazsoft’s significance lies in how many of those decisions it says Siptek can make. Its Softswitch page describes a Class 4 system with multi-tenancy, traffic routing, charging, pre-answer classification, call-detail-record reconciliation, financial automation and support for STIR/SHAKEN. The company says the switch can hold 100,000 VoIP accounts or source IPs, process millions of records a month and add less than 20 milliseconds of call-processing delay. Its automatic dialer page adds predictive, preview and power dialing; interactive voice response; voice, SMS and WhatsApp campaigns; live dashboards; call classification; and Gemini-assisted transcription and summaries. Its PABX page extends the same family into extensions, queues, recordings, browser calling, white-labelling and CRM or ERP integration.
Those are supplier statements, not independently reproduced measurements. That distinction matters because Dazsoft also says more than 6,400 companies use its products, that its systems handle more than 606 million calls a month and that more than 51,000 channels are active each day. The figures appear across its product and company pages, but the public material examined for this article does not define the counting method, period, duplication rules or assurance behind them. They establish the scale Dazsoft wants buyers to believe; they do not, by themselves, establish capacity available to any particular tenant.
Still, the product map is coherent enough to make Dazsoft strategically interesting. Siptek is not presented as a narrow app at the edge of a contact centre. It is presented as the control fabric between campaign logic, carrier interconnection, the service representative, the customer account and the financial ledger. If it works as advertised, a smaller operator can buy capabilities that would otherwise require several products and a specialist engineering team. If it fails, the same concentration can turn one incident into missed calls, incorrect charges, untraceable traffic, regulatory exposure and a difficult migration.
That is the thesis to test: Siptek’s value and its liability come from the same place.
One CNPJ, two public legal names
This article concerns the assigned entity Dazsoft Desenvolvimento de Softwares Ltda ME, not a similarly named reseller, a product brand or an inferred corporate group. The cleanest bridge is Brazil’s company identifier. A Receita Federal data presentation by Casa dos Dados lists CNPJ 06.209.516/0001-27 as active, with the legal name Dazsoft Desenvolvimento de Softwares Ltda, opening date April 8, 2004, and software development, licensing, consulting and support activities. It names Daniel Zanutti Gomes and Samuel Palca Luchetti as managing partners. Its Jundiaí address is the same address now displayed in the footer of Dazsoft’s product pages.
The company’s own privacy policy, last marked as updated in October 2020, uses a longer name: Dazsoft Comercio e Desenvolvimento de Softwares Ltda. Crucially, it assigns that name the same CNPJ, 06.209.516/0001-27. A São Paulo commercial-registry notice published in 2022 also referred to “Daz-Soft - Comercio e Desenvolvimento de Softwares Ltda” when recording a capital and corporate-purpose change. The most defensible reading is therefore a historical or document-maintenance difference around one registered company. It is not defensible to pretend the words have always been identical.
There is a second independent bridge. Public network-registration data for the company’s address block links AS274461 and 45.170.138.0/24 to “Dazsoft Desenvolvimento de Softwares Ltda ME,” the same CNPJ, the dazsoft.com.br domain and Daniel Zanutti Gomes. Dazsoft’s Softswitch page identifies Daniel Zanutti as its technical director in the presentation of Siptek 3.0. Meanwhile, Dazsoft’s own pages consistently label Siptek as its switch, dialer, PABX and softphone family; the separate Siptek Softphone site returns to Dazsoft’s telephone number, email and company description.
This evidence is sufficient to attribute the public Siptek offering to the specified entity. It also creates a practical contracting point. A buyer should require the proposal, data-processing terms, invoice, certificate enrolment and support commitment to state the same current legal name and CNPJ. The old wording in a privacy notice is not proof of wrongdoing, but identity ambiguity is avoidable friction in a system that may hold call recordings, customer records and signing authority. The contract should resolve it explicitly.
Nor should the autonomous-system registration be overread. Owning an ASN and a /24 block demonstrates a real network footprint and further confirms identity. It does not prove that Dazsoft is a licensed telephone carrier, that every Siptek workload sits on that prefix, or that the advertised cloud estate is redundant. Those are separate questions.
Siptek’s control surface, from route table to headset
Siptek’s breadth becomes clearer when the call path is separated into layers.
At the carrier-facing layer, the Softswitch is described as Class 4. In conventional terms, that is the transit and wholesale side of voice: accepting SIP traffic, applying routing policy, choosing gateways, enforcing account limits, recording call outcomes and rating traffic. Dazsoft says the system serves SCM and STFC operators, supports cloud and on-premises deployment, separates tenants, and combines routing and charging in one interface. Its operator solution page adds prepaid and postpaid rating, reseller administration, automated collections, call-record reconciliation and number portability.
At the enterprise layer, Siptek Suite PABX registers extensions, builds queues and IVR trees, records calls, supplies browser or desktop endpoints and integrates with business systems through a REST interface. Multi-tenancy and white-labelling let a carrier or reseller expose the product under its own brand. That is commercially useful, but it means a Dazsoft change can propagate into the services of customers that an end user may not recognise as Dazsoft customers.
At the outbound-work layer, the dialer decides pace. Preview mode lets a representative inspect the record before a call. Power mode accelerates a defined list. Predictive mode estimates how many calls to launch against available staff and expected answer rates. The product page also advertises pre- and post-answer classification, retry scoring, IVR, dashboards and multichannel campaigns. Those are not decorative features. They govern how many consumers are contacted, how often, whether a live answer meets a person, and what evidence remains after a complaint.
At the endpoint, the Siptek Softphone supports browser and desktop use, recordings, two lines, transfer and several codecs. Its site offers both software-as-a-service licensing and a sale option, plus a 30-day trial. Testimonials on that site are useful as leads, but they are supplier-selected evidence. One says Dazsoft support helped through adaptation problems; another says the software is used with PABX and dialing customers. A buyer should verify references directly, under permission, and seek workloads similar in call rate and topology to its own.
Finally, Dazsoft is adding analytical and financial layers. It says Checkcall IA compares Siptek call records with supplier records and flags differences in duration, cadence, rate or configuration. It says Call Analytics uses Gemini to transcribe calls, produce summaries and identify speech patterns. It connects billing to Asaas for charge creation, notifications, suspension and reactivation, while offering interfaces for other banks and ERP systems.
The stack therefore controls four kinds of truth at once: signaling truth, about what happened on the network; commercial truth, about what the call cost; operational truth, about what staff and campaigns did; and evidentiary truth, about what can later be shown to a customer, carrier or regulator. Integration can keep those truths aligned. It can also allow one bad configuration, clock drift, delayed event or access compromise to corrupt several at once.
Routing and billing are one economic machine
A softswitch does more than connect A to B. It decides which B-side supplier receives the traffic, under which tariff, with which number presentation and under what limits. A least-cost route may improve gross margin, but only if it still meets quality, fraud and regulatory constraints. The cheapest path can have lower answer-seizure performance, longer post-dial delay, codec changes, number rewriting or weaker support for authenticated identity. A procurement exercise that tests only whether a call completes will miss the economics.
Dazsoft’s public material says Siptek can choose lower-cost routes, expose SIP traces, monitor traffic and manage prepaid and postpaid charging. It says Checkcall IA can compare the switch’s records against supplier records and has found value divergences as high as 40 percent in real cases. That percentage is a company claim without a published sample, denominator or case file. The useful proposition beneath it is sound: a carrier should not accept its supplier’s invoice as the only account of traffic.
The reconciliation test should start with a controlled set of calls that includes answered, busy, rejected, cancelled, forwarded, ported, voicemail and failed-authentication cases. For each call, the buyer should compare the Siptek record, upstream record, downstream record where available, campaign record, recording index and invoice line. Time zones, start and answer timestamps, billable duration, rounding, cause code, selected route, tariff version, number portability result and identity result should agree or have a documented transformation.
Brazilian regulation makes this more than housekeeping. Anatel’s current consumer billing guidance requires detailed reports to carry, where applicable, called number, origin and destination area, start time to the second, billable duration, value and other detail. Its 2026 anti-spoofing controls also allow the regulator to require call-detail records and current traffic routes from operators. A switch that cannot preserve the relationship between route, asserted origin, outcome and charge is not merely inconvenient during an invoice dispute; it weakens the operator’s regulatory evidence.
Portability adds another moving part. Under Anatel’s 2025 general services regulation, Brazil’s central reference base updates each operator’s operational portability base, which is then used for correct routing. “Portability support” should therefore be tested as a timed operational process: update frequency, failed downloads, rollback, stale-data alarms, routing before and after a port, and audit evidence showing which version answered a lookup. A vendor screen that displays the right carrier today says little about how safely tomorrow’s update will be handled.
The financial automation deserves the same discipline. Automatic suspension after a payment event can prevent credit exposure, but a false unpaid state can disconnect a customer. Automatic reactivation can restore service quickly, but only if the bank event, ERP state and switch permission converge. Buyers should test duplicate webhooks, delayed settlement, partial payment, reversal, bank outage and manual override. Every automated action should be idempotent, attributed, reversible and visible to support.
The dialer converts efficiency into conduct
Predictive dialing is a capacity optimisation that acts on people. It launches more calls than there are currently free representatives because many attempts will be busy, unanswered or otherwise unproductive. Accuracy improves utilisation. Error produces abandoned calls, dead air, repeated attempts and the very short-call patterns that Brazilian regulation has targeted.
Anatel’s current page on abusive-call controls, updated in May 2026, explains the enforcement path. The regulator has required blocking of users that generate at least 100,000 calls in a day when 85 percent or more are short. Since June 2024, a short call includes a completed call lasting up to six seconds, whether ended at origin or destination; voicemail-directed calls also enter the expanded treatment. In May 2026, Anatel extended the relevant measure to October 2028. Its broader consumer explanation says more than 248 billion calls had been prevented, more than 1,200 company blocks made and nearly R$40 million in fines applied by the time of its July 2026 update.
This changes how Siptek’s advertised classifier and retry score should be evaluated. A classifier that discards an answering machine or invalid number before involving staff may save channels. But “unproductive” must have a precise definition. Does the system mistake a slow greeting, an accessibility device, a noisy mobile line or an older caller for a machine? Does a post-call score increase repeat attempts to people who briefly answered and rejected the call? Does a campaign stop when short-call ratios rise, or merely report the damage?
A serious proof should use consenting test recipients and synthetic lines that emulate busy, no answer, carrier announcements, voicemail, fax, delayed greeting, silence, speech in different Brazilian accents and rapid hang-up. The buyer should record classification precision, false rejection, time to connect a live representative, abandoned-call rate and short-call share by campaign and hour. It should deliberately remove representatives mid-run and confirm that pacing falls before consumers hear silence. It should test list suppression, time windows, retry caps and immediate opt-out propagation.
Verified identity does not legitimise an abusive campaign. Anatel says authentication complements, rather than replaces, opt-out and conduct rules. The regulator’s public-calling guidance still requires reasonable calling volume, commercial hours, complaint handling and respect for consumers who have opted out. A verified company can still call too often; authentication can make responsibility easier to assign.
That is why the dialer, switch and compliance archive must be evaluated together. The campaign system should be able to reconstruct who supplied the list, which legal basis and suppression version were used, which policy allowed an attempt, which number was presented, which route carried it, whether it was authenticated, how it ended and whether another attempt was scheduled. Efficiency without that lineage is a liability accelerator.
Origem Verificada is a chain of custody, not a badge
Dazsoft has built a dedicated marketing page at origemverificada.com saying its Softswitch is ready to authenticate and carry calls using Origem Verificada and STIR/SHAKEN. The page says the product can help operators display a seal, company name, logo and call reason, and describes the capability as a route to compliance and premium revenue. The domain is a Dazsoft commercial site: it links back to Dazsoft, carries the same Jundiaí address, telephone number and email, and promotes other Siptek products.
That site should not be confused with the national service’s portal at origemverificada.com.br. Anatel’s authentication and identification guide says the specific Origem Verificada service is implemented by the Autoridade de Identificação e Autenticação, or AIA, through ABR Telecom. The AIA manages certificates, validates registrations, applies security and governance policies and can revoke certificates for misuse. Companies enrol through participating telephone providers and submit credentials and display data for validation.
The distinction exposes what “support” has to mean technically. RFC 8224 defines how a SIP Identity header carries a signed PASSporT assertion, allowing a verification service to check that an originating identity was authorised. RFC 8588 adds the SHAKEN attestation and origination identifier. Attestation can be full, partial or gateway-level; the origination identifier supports traceback. The 2025 RFC 9795 defines Rich Call Data for information beyond the number and recommends authoritative vetting for names, icons and other content. A logo arriving on a screen is the last visible result of enrolment, number authority, signing, certificate distribution, SIP preservation, terminating verification, handset support and display policy.
Brazil’s implementation also has distinct authentication and identification outcomes. Anatel says compatible 4G or 5G phones need VoLTE and a current operating system. Display varies: newer Samsung devices may show name, number, seal, logo and reason; older supported Android devices may show fewer fields; on iPhones, the seal may appear only in call history. Older 2G or 3G devices receive the call without enriched display. A failed logo display is therefore not automatically a failed signature, and a displayed name is not sufficient proof that the underlying assertion was accepted.
The regulation has moved beyond voluntary marketing. Anatel’s October 2025 implementation decision requires users making more than 500,000 calls per month per provider to authenticate all calls. The count aggregates access codes tied to the same CPF or to a company’s head office and branches. Providers must notify non-compliant large callers and block origination after the remediation period; the regulator may also consider volumes spread across providers or related companies. Separately, the 2025 regulation sets a progressive path toward broad authentication by 2028.
A Siptek proof should consequently answer at least six questions with packet captures and AIA or carrier evidence, not screenshots alone.
First, where does signing occur? It may be in Siptek, at an adjacent session border controller, at the contracted carrier or at another approved service. Responsibility for private keys and certificate renewal follows that answer.
Second, what authority does the signer have over each presented number? Full attestation should not be asserted simply because a customer typed a number into a portal. The operator needs an auditable enrolment and number-assignment check.
Third, what survives interconnection? The test must cross each important originating and terminating carrier, include forwarding and ported numbers, and show the Identity header or approved equivalent reaching verification without damaging legitimate number transformations.
Fourth, what happens on failure? Expired or revoked certificates, unavailable certificate endpoints, clock skew, malformed assertions and unsupported destinations need defined fail-open or fail-closed behaviour. The answer may differ between authentication, call completion and enriched display.
Fifth, how is Rich Call Data governed? Names, logos and reasons can be useful, but the IETF’s RCD specification warns that accuracy and legitimacy need an authoritative vetting process. The buyer needs approval, change control, content integrity and revocation, not merely an upload form.
Sixth, can the operator prove the result later? The call record should retain the number-authority basis, attestation, certificate reference, origination identifier, verification outcome and display payload version without exposing private key material. That trace must connect to the campaign and route record.
Dazsoft’s public pages do not answer these questions in enough detail. That does not show the capability is absent. It means a procurement team should treat the statement “supports STIR/SHAKEN” as the beginning of an acceptance plan, not the end of diligence.
Reliability is topology plus recovery
Dazsoft says its hosted systems use leading data centres, Tier III facilities, several layers of redundancy and infrastructure associated with Google Cloud, Equinix and AWS. It says Softswitch processing adds less than 20 milliseconds and that specialist support has an average response below two minutes. These are attractive statements, but they mix different units of reliability.
Processing delay inside a switch is not end-to-end post-dial delay. A laboratory path with warm caches is not a congested carrier route. An average first response is not restoration time. A facility certification does not prove that a particular customer is deployed across independent fault domains. “High availability” says nothing about the duration, exclusions or service credits of a contractual commitment.
The public network footprint offers one useful clue and several cautions. Registration records show Dazsoft controls AS274461 and a /24 IPv4 range. A public route snapshot examined for this article showed the prefix reachable through Equinix’s AS15830. That is consistent with a real hosted network presence and with one of the names shown on Dazsoft’s product page. It does not reveal every upstream, private interconnection, cloud region, customer-specific path or disaster-recovery site. Nor does it prove that the voice plane, management plane, recordings and analytical workloads share the same topology.
The buyer should ask for a service-specific diagram that separates signaling, media, administration, databases, recordings, analytics, certificate services and external dependencies. It should identify regions or facilities, availability zones, upstream carriers, session border controls, load distribution, state replication, backup location and the exact boundary between Dazsoft and the customer. Cloud and on-premises versions need separate diagrams.
Then the buyer should break the diagram. Remove one upstream. Stop one signaling node. Deny access to the portability feed. Isolate the rating store. Expire a certificate. Lose a recording volume. Delay the bank webhook. Saturate calls per second while keeping simultaneous calls below the nominal limit. Force the primary site away and back. Each exercise should measure call completion, post-dial delay, media quality, duplicate charging, record loss, recovery point, recovery time and the clarity of alarms.
Public-sector procurement shows how specific the market can be. A 2025 procurement notice combined cloud PABX, SIP channels, portability, training, assisted operation and at least 365 days of recording storage. The notice does not set a universal standard and does not evaluate Dazsoft. It demonstrates that “cloud telephony” can be translated into measurable continuity, retention and transition obligations.
A credible Siptek contract should do the same. It should define availability per service, exclude only narrow scheduled windows, state measurement location, cover both control and media planes, assign severity, establish response and restoration targets, require incident chronology and root-cause reporting, and give the customer telemetry to reproduce the calculation. The two-minute response claim should be tested at 3 a.m. during a simulated critical incident, not accepted from a brochure.
Security begins at the SIP edge and ends in the tenant
Voice infrastructure has an awkward security shape. It is internet-facing, stateful, latency-sensitive and financially metered. It handles credentials, telephone numbers, recordings and call metadata. It must accept traffic from known partners while resisting scanning, credential attacks, toll fraud, signaling floods, malformed messages and unauthorised route changes. Multi-tenancy adds the requirement that one customer’s administrators, traces, recordings and rate tables never cross into another’s.
Dazsoft’s PABX page says calls can use TLS and SRTP. Its Softswitch page describes authentication by source IP or username and password, as well as redundant infrastructure. Its privacy notice mentions HTTPS, end-to-end encryption and firewalls in general terms. These are relevant controls, but public wording does not establish where encryption terminates, whether it is mandatory, how keys are managed, whether management access has multi-factor authentication, or how tenant isolation is tested.
The management plane is as important as the media plane. An administrator who can change a route can redirect traffic or increase cost. Someone who can change a presented number can create spoofing exposure. Someone who can edit a tariff can alter invoices. Someone who can export a SIP trace can see sensitive metadata. Privilege should therefore be granular: route administration, rating, campaign operation, recording access, identity enrolment, support impersonation and financial suspension should be separate rights.
High-impact changes should require stronger authentication, approval where appropriate, immutable audit entries and rapid rollback.
Session border protection should be made explicit. Buyers should ask how Siptek normalises SIP, enforces rate and call-per-second limits, detects registration attacks, controls media pinholes, blocks malformed headers, handles topology hiding and resists denial of service. They should request evidence for secure development, dependency maintenance, code signing, penetration testing, vulnerability intake, remediation targets and customer notification. Anatel’s cybersecurity regulation, updated in 2026, sets security expectations across telecom networks and allows the regulator to impose technical measures; its precise application depends on the regulated provider, but a supplier in the critical path will be part of the provider’s risk evidence.
Data incidents add another clock. The ANPD incident regulation requires a controller to notify the authority and affected people when an incident may cause relevant risk or damage, and to retain incident records for at least five years. A Dazsoft customer may be the controller while Dazsoft acts as its processor; in other contexts, roles may differ. The contract must ensure that Dazsoft’s detection and notice reach the controller early enough for assessment and lawful communication under the current applicable timetable.
The frozen public evidence set used for this article did not reveal a Dazsoft public status history, named security certification, independent assurance report, vulnerability-disclosure page, published penetration-test summary, detailed subprocessor list or incident archive. Absence from this evidence set is not proof that none exists privately. It is a list of documents to request. The right response is not to infer insecurity; it is to refuse unverifiable security adjectives when the system can change routes, bills and caller identity.
AI analytics redraws the data boundary
Dazsoft says Siptek’s Call Analytics uses Gemini to transcribe calls, generate customised summaries, identify keywords and recurring behaviour, and support near-real-time monitoring. That can reduce manual sampling and make more conversations searchable. It also converts audio into a larger and more portable body of personal data.
A recording already contains voice, identity clues, account information and whatever the caller discloses. A transcript makes that material easier to search, copy and correlate. A summary adds an automated interpretation. Keyword and behaviour labels can affect staff assessment, customer treatment or escalation. Each derivative needs a defined purpose, access policy, retention period, correction route and deletion process. Accuracy should be measured separately for noisy lines, overlapping speech, regional accents, code-switching, names, account numbers and sensitive contexts. A confident summary can still be wrong.
The company’s privacy notice is too old and broad to answer the product-specific questions. It was marked updated in October 2020, years before the current Call Analytics presentation. It allows sharing with cloud, analytics and other service providers and permits international transfer under stated safeguards, but it does not name the Gemini route, processing location, audio flow, transcript retention or optional features. The notice also switches between Dazsoft as controller and processor, correctly signalling that role depends on context but leaving the customer to specify it.
“Uses Gemini” is not enough to determine data treatment. Google’s terms differ by service and commercial tier. For example, Google’s Vertex AI data-governance documentation says customer data is not used for training without permission, while also describing limited retention cases and the steps needed for zero retention. The Gemini API paid-service terms say paid-service inputs and outputs are not used to improve Google products and are handled under a processing addendum. Those protections cannot be assumed until Dazsoft identifies the exact service, account tier, configuration and contractual chain it uses.
A buyer should require a call-level data-flow diagram: where recording begins; whether audio is streamed or stored before analysis; which Dazsoft and third-party systems receive it; where each copy is processed; how long audio, transcript, summary and diagnostic data remain; whether content is used for service improvement; and how deletion propagates. It should be possible to disable analytics by tenant, campaign or queue without disabling core telephony.
The customer should be able to export the transcript with timestamps, confidence information, correction history and linkage to the original recording, then delete both under an agreed schedule.
Evaluation should use a consented reference set with human-prepared transcripts and expected outcomes. Measure word error, omission of negation, speaker attribution, summary faithfulness and alert precision. Do not let a single aggregate accuracy conceal failures in vulnerable groups or high-risk call types. If a summary influences coaching, compliance or dispute handling, the original audio and a human appeal route remain essential.
A proof of concept must rehearse the real operation
Dazsoft says its implementation method is fast and less bureaucratic. Speed is valuable only after the boundaries are known. Cloud deployment removes server ownership from the customer but adds network, tenancy and supplier dependencies. On-premises deployment gives the customer more infrastructure control but creates a shared-responsibility problem around operating systems, capacity, backups and access. Hybrid integrations can combine both failure sets.
There is one relevant but limited piece of adverse public evidence. In November 2025, a complainant on Reclame Aqui alleged that an on-premises Siptek Suite proof had passed, but the paid installation’s dialer did not work in the customer environment; the complainant also disputed setup and termination charges. Dazsoft replied that it would audit the technical and financial documents and asked for the contracting CNPJ. The page showed no final resolution in the captured evidence and labelled the company unverified on that consumer platform.
One unresolved complaint cannot establish a general failure rate, and the customer’s full environment and contract are not public. It is nevertheless a useful procurement warning because it describes the exact seam at which proofs often fail: the demonstration does not reproduce production infrastructure, data, concurrency, routes or responsibility. The lesson is not that Siptek fails. The lesson is that “proof passed” must have contractual meaning.
A robust evaluation should progress through five gates.
The first is discovery. Inventory carriers, trunks, number ranges, portability dependencies, codecs, peak simultaneous calls, calls per second, campaigns, IVRs, recordings, integrations, firewall paths, remote staff, billing rules, retention and regulatory obligations. Assign ownership for every prerequisite.
The second is a representative laboratory. Use the intended software version and deployment type. Reproduce latency, NAT, security controls, server sizing, carrier routes and identity services. Load real configuration through controlled migration, not by retyping a simplified sample.
The third is adversarial acceptance. Run peak and burst traffic; kill components; introduce packet loss and jitter; rotate credentials; expire a certificate; deliver duplicate payment events; change a tariff mid-cycle; port a test number; generate each important SIP failure; and compare records across systems. Test both successful calls and failures.
The fourth is parallel operation. Route a bounded share of real, consented traffic through Siptek while the previous path remains available. Reconcile completion, quality, cost, records, recordings, authentication and customer complaints daily. Define quantitative promotion and rollback thresholds before starting.
The fifth is exit rehearsal. Export configurations and data, restore them into an independent environment where feasible, port test numbers away, revoke credentials, retrieve recordings and verify deletion. A system is not production-ready until the customer knows how to leave it.
Acceptance should name the environment, dataset, version, date, witnesses and raw evidence. A supplier should not be able to satisfy a production obligation with a sales demonstration, and a customer should not be able to blame the supplier for an excluded dependency. Precision protects both sides.
Pricing follows the points of control
Dazsoft does not publish a full Siptek price card in the public pages examined. The sites direct buyers to a sales conversation, while the softphone offers a 30-day trial and refers to service subscription or purchase. That suggests negotiated pricing shaped by deployment and scope rather than a single self-service rate.
The likely price drivers follow the architecture: setup and migration; cloud or on-premises deployment; simultaneous channels and calls per second; accounts or extensions; switch, PABX and dialer modules; recordings and retention; analytics volume; support coverage; number-portability functions; authenticated-calling services; interfaces; white-labelling; reseller hierarchy; and custom work. Carrier minutes, telephone numbers, messaging, cloud resources, third-party AI processing and AIA or provider charges may sit outside the software quote.
Buyers should require a price workbook that separates one-time, recurring and usage charges, then applies them to normal, peak and growth cases. It should show minimum commitments, overages, indexation, taxes, support tiers, storage retrieval, data export, disaster recovery, non-production environments and termination assistance. A low per-channel rate can be overwhelmed by setup, retained recordings or bespoke integrations. Conversely, an integrated suite may be cheaper than staffing and reconciling several products even when its licence is higher.
Financial automation also changes switching cost. If Siptek creates Asaas charges, suspends service and feeds an ERP, it becomes part of cash collection. Replacing the switch then means replacing event mappings, customer states, exception procedures and audit history. Those dependencies should appear in the business case at purchase, not be discovered at renewal.
Portability means more than keeping a telephone number
Brazil gives consumers a right to number portability, and the national mechanism reduces one classic source of carrier lock-in. Software lock-in survives elsewhere.
A Siptek customer may accumulate route policies, tariff tables, customer balances, reseller hierarchies, number assignments, IVR trees, queue rules, campaign history, suppression lists, recordings, transcripts, dashboards, identity enrolments, certificate references, bank mappings, ERP interfaces, custom code and staff knowledge. Even if every telephone number moves in three days, rebuilding that operating memory can take months.
White-labelling deepens the issue. A carrier or reseller may present Siptek under its own identity and build customer processes around its portal. Migration then affects not just the direct buyer but downstream tenants, training, branded endpoints and support scripts. Multi-tenancy makes scale economical while increasing the number of parties touched by a change.
The contract should define an exit package from day one. Configuration exports need documented, machine-readable formats and stable field definitions. Call-detail records should include original precision and cause information. Recordings and transcripts need durable linkage. Tariff versions, route changes and audit entries need history, not only current state. Identity material must distinguish customer-owned information from supplier-held private keys. Interfaces and webhooks need documentation.
The buyer should receive reasonable transition support at a known rate, continued read access for an agreed period, and verified deletion after handover.
Source-code escrow may be appropriate for a deeply embedded on-premises deployment, but it is not a substitute for operational portability. Source without build instructions, dependencies, signing material, current data and skilled maintainers may be unusable. A stronger continuity package combines escrow where justified with frequent exports, configuration documentation, independent backups, interface tests and a secondary routing plan.
The build alternative is real. FreeSWITCH’s official documentation describes an open-source switching platform that can register endpoints, route calls, handle media, expose events and integrate with external systems. Open components can reduce licence dependence, but the customer inherits architecture, security patching, billing, high availability, regulatory integration, user experience and round-the-clock support. Dazsoft’s commercial proposition is precisely to package much of that work. The correct comparison is not licence fee versus zero; it is accountable service versus the full cost and risk of assembling an equivalent operation.
Competition tests whether the suite is truly integrated
Dazsoft competes in a Brazilian market with local suites, international cloud communications, carrier equipment and open components. Public competitor pages show that its headline features are valuable but not unique.
Tellestra advertises a multi-tenant Class 4 switch available as a service or on premises, with prepaid and postpaid control, least-cost routing, portability updates, dashboards, detailed call records and ERP interfaces. VoIP Group markets Class 4 and 5 switching, real-time billing, portability, dynamic routing, session-border control, STIR/SHAKEN, contact-centre functions and cloud deployment. These are competitor statements rather than independent benchmarks, but they show the comparison set a Brazilian operator can assemble.
At the other end, an enterprise may choose a broad contact-centre service and leave carrier routing or billing to telephone providers. A technically mature operator may compose an open SIP proxy, media switch, rating engine, observability stack and separate dialer. A large carrier may choose established network-equipment suppliers. Each route trades integration against flexibility.
Dazsoft’s differentiation therefore has to be demonstrated in the seams. Does the dialer understand switch capacity quickly enough to prevent abandoned calls? Does the classifier improve answer efficiency without polluting short-call ratios? Does reconciliation explain a discrepancy down to route and tariff version? Does authenticated identity survive the exact carriers the customer uses? Does one support team own failures that cross switch, PABX, dialer and billing? Can all of that be exported cleanly?
A feature list will produce a tie because many suppliers can mark “yes.” A scenario comparison is harder to game. Give finalists the same ported numbers, rate tables, carrier routes, campaign, failure injections, privacy requirements and exit task. Measure completion, quality, reconciliation, evidence, recovery and operator effort. The best product is the one that preserves truth under change, not the one with the longest menu.
The public evidence gap belongs in the purchase decision
Dazsoft’s public presence has improved enough to describe a substantial product, yet it remains brochure-heavy. The company explains what Siptek does and publishes impressive operating figures. It provides a privacy notice and identifies the legal business. It shows customer logos and selected testimonials. It does not publicly supply the depth of technical and assurance material that a carrier-core buyer would ideally use before sharing routes, records and identity functions.
The missing public evidence includes a detailed reference architecture; per-service availability commitment; public incident history; independent capacity methodology; current security assurance; tenant-isolation test summary; vulnerability channel; software support lifecycle; release and deprecation policy; backup and recovery evidence; data-location map; named subprocessor inventory; current product-specific privacy terms; exact Gemini configuration; authenticated-calling implementation guide; public interface documentation; export specification; and transparent pricing.
Some of these documents may be available under confidentiality. Some may not exist. The procurement task is to find out and to distinguish a document from an assertion. An architecture slide should match deployed resources. A penetration result should state scope and date. A recovery report should show measured objectives. A support report should include severity and restoration, not only first reply. A capacity result should define codec, signaling mix, hardware, redundancy state and acceptable latency. A customer reference should be independently reachable and comparable.
The same caution applies to customer logos. Dazsoft’s operator and contact-centre pages show Imperatriz Telecom, Agitel, Vox Telecom, Carrefour, Mega Telecom and others under companies that trust its services. The pages do not specify product, deployment, date, volume or permission context. Logos can identify reference leads; they cannot substitute for a reference call.
Evidence gaps do not erase the business. Dazsoft has operated under the same CNPJ since 2004, holds network resources, publishes a coherent product family and addresses live Brazilian regulatory needs. The gaps change the terms on which that business should be trusted. The deeper Siptek sits, the more evidence should be delivered before the customer removes its fallback.
The procurement proof list
A buyer evaluating Siptek should leave the process with answers that can be tested again after upgrades.
For identity and scope, confirm the contracting name, CNPJ, product owner, deployment entity, support entity, invoice issuer, data-processing roles and every subcontracted service. Match those names across the proposal, privacy terms, certificates and billing.
For architecture, obtain diagrams for signaling, media, administration, data, recordings, analytics, identity and external integrations. Mark every fault domain, network path, location, tenant boundary and shared component. Document cloud and on-premises responsibility separately.
For capacity, measure sustained and burst calls per second, simultaneous calls, registrations, record generation, dashboard delay and reconciliation throughput using the expected codecs and features. Run at normal, peak and degraded redundancy. Define failure thresholds before testing.
For call quality, measure answer-seizure ratio, average call duration, post-dial delay, jitter, loss, one-way audio, codec negotiation and cause-code accuracy by route. Confirm that least-cost decisions cannot silently override quality or authenticated-calling requirements.
For dialing, test predictive pacing with changing staff availability, short-call thresholds, voicemail, delayed speech and suppression. Record abandoned-call and false-classification rates. Prove retry caps and opt-out propagation.
For billing, reconcile a known call set across Siptek, carriers, campaign history and invoices. Test tariff changes, rounding, time zones, ported destinations, duplicate events, credits, prepayment exhaustion and postpayment cycles. Require versioned rates and immutable change history.
For verified origin, test number authority, each attestation level used, certificate renewal and revocation, clock skew, malformed assertions, forwarding, portability, multiple carriers and representative devices. Separate signature success from display success. Validate names, logos and call reasons through the governing process.
For security, verify multi-factor administration, role separation, privileged approvals, audit export, encryption boundaries, secret storage, session-border controls, tenant isolation, patching, secure development, penetration scope, vulnerability handling, backup protection and incident notice. Include support access and emergency elevation.
For privacy and analytics, identify the exact Gemini service and commercial tier, processing locations, retention, improvement use, subprocessors and deletion path. Test analytics off. Evaluate transcripts and summaries against a human reference set, including difficult audio and consequential errors.
For reliability, agree service-specific availability, measurement, severity, response, restoration, recovery point, maintenance and credits. Conduct component, site, carrier, certificate, database and integration failures. Require incident timelines and corrective-action tracking.
For support, open routine and critical cases across the promised schedule. Measure acknowledgement, qualified engagement, escalation and restoration. Confirm Portuguese-language expertise in SIP, carrier billing, Anatel controls, data protection and the customer’s deployment.
For lifecycle, obtain supported-version periods, upgrade cadence, security-fix policy, backward compatibility, interface deprecation, maintenance windows and rollback. Run an upgrade with traffic and prove that records, billing and identity remain consistent.
For exit, export every important configuration and record type; restore a sample independently; move test numbers and routes; retrieve recordings and transcripts; revoke access; and verify deletion. Price transition support before signature.
Finally, require the acceptance pack to survive staff turnover. Raw captures, scripts, expected results, actual results, exceptions and approvals should belong to the customer. A future incident team should not need the salesperson’s memory to understand why the system was accepted.
What to watch through 2028
The first watchpoint is Brazil’s authenticated-calling rollout. Anatel’s three-year direction will increase the value of working implementation and expose suppliers whose support is only nominal. Buyers should monitor AIA participation, carrier coverage, certificate governance, Rich Call Data display and enforcement against large callers.
The second is Dazsoft’s evidence maturity. A public status service, security channel, current privacy material, product documentation, assurance reports and precise service commitments would materially improve confidence. So would independently verifiable customer cases that disclose deployment type and scale.
The third is operational transparency. Dazsoft’s scale and latency figures could become more useful if accompanied by definitions and test conditions. Checkcall IA could become a strong differentiator if the company publishes an anonymised evaluation of reconciliation precision, false alerts and financial recovery. Call Analytics needs equally clear accuracy and data-governance evidence.
The fourth is concentration. As Siptek adds WhatsApp management, analytics, finance and identity, buyers gain a simpler operating surface but a larger blast radius. Watch whether interfaces and exports mature at the same pace as features. Integration without portability is deferred migration cost.
The fifth is incident and support evidence. No supplier is incident-free. The useful signal is whether failures are detected, disclosed, reconstructed and prevented from recurring. Dazsoft’s claimed response speed should eventually be accompanied by restoration distributions and severity definitions.
The vendor beneath the voice
Dazsoft is easy to underestimate because much of its public presentation looks like familiar call-centre software. The deeper reading is more consequential. Siptek can sit where a Brazilian operator chooses routes, establishes who may call as which number, counts money, drives contact attempts, records conversations, checks supplier bills and explains a call after the fact.
That is a valuable position. It can give a regional carrier or contact centre a coherent system without forcing it to assemble and maintain every component. It can bring Brazilian regulatory adaptation, local support and commercial automation closer to the switch. Dazsoft’s long-lived CNPJ, product continuity and network footprint make it more than a landing-page brand.
But control has to be earned at each layer. The legal name must be unambiguous. Performance must be reproduced. Availability must survive failure. Call records must reconcile. The dialer must respect people as well as utilisation. Verified origin must preserve authority and traceability across the chain. Recordings and transcripts must have a lawful, bounded data path. Support must restore, not merely reply. Exit must work before it is needed.
Siptek’s integrated story is strongest when a buyer treats it neither as a black box nor as a collection of checkboxes. It is an operating system for voice economics and accountability. The correct procurement question is therefore not simply whether Dazsoft can place a call. It is whether Dazsoft and the customer can prove, under load and under failure, why that call was placed, how it was routed, who authorised its identity, what it cost, what happened to its data and how the operation continues when any one part goes away.

