A note before we start: this review tests two different surfaces of the same product. The MJ Realty demo line (+1 984-205-2314) that founder Driss Chelouati shared directly is a polished, presumably done-for-you deployment. The self-serve wizard is what a new $33/month customer actually gets. They are meaningfully different products, and the piece treats them as such. Section 3 is based on ten calls to the MJ Realty number. Section 2 is based on running the self-serve wizard up to (but not through) the payment wall.
1. Who It’s For
Hello Gubby is positioned for service-business operators who live on the phone, the kind of business where missing a call costs a job: home services, property management, real estate, salons, automotive, professional services. Fourteen industries are listed on the site, with sub-agent templates underneath each one (Home Services alone has roughly half a dozen, including “Emergency Dispatch & Job Assignment”). The breadth suggests the same play as several other entrants in this space, e.g. carve up the SMB long tail by vertical and let each vertical recognize itself in the marketing.
What sets Hello Gubby’s positioning apart is the founder’s framing of the bigger problem. In a brief sent directly to us from the founder, he describes the company’s vision as “Intercom for small businesses.” The argument is that SaaS companies got a unified customer-communications layer (Intercom) and SMBs never did. A salon, a plumber, a property manager lives across phone, text, web chat, social DMs, and email, every channel in a different tool. Leads fall through. Hello Gubby’s pitch is one AI employee across all of them.
“The category will be won by whoever builds operational workflows on top of voice, not by whoever has the most voices.”
That’s a real category bet, and a more ambitious one than the homepage tagline (“AI front desk”) communicates. The marketing site is selling a receptionist; the product team is selling a customer operations platform. Whether the shipped product delivers on the bigger framing is the question this review tries to answer.
The buyer the marketing copy is written for is the owner-operator who’s tired of missing calls. The buyer the founder is building for is the operations lead who wants every customer touch to flow through one system. Right now the product is sized for the first buyer. Whether it grows into the second is the open question.
2. Setup Experience
The self-serve wizard moves quickly, about ten minutes from signup to the package-selection step, and the questions it asks are reasonable: company name, industry, basic operating details, FAQs, lead capture fields, agent persona, opening greeting, email notifications, package.
FAQs are typed in manually. This is the first surprise. The platform doesn’t scrape the customer’s website to populate the knowledge base. For a product whose marketing promises “trained on your business,” the absence of website ingestion is conspicuous. The customer is doing the training, by hand, in a textbox. The OpenMic platform reviewed previously scraped the site automatically, with its own set of problems, but at least it tried. Gubby asks the operator to do the work.
Three voices, all female-named. Amanda, Amy, April. No male voices, no non-Western names, no option to upload or clone. Alliterative naming on “A” suggests these are curated ElevenLabs library voices rather than custom-trained ones, which is fine. The lack of a male option is at minimum a missed product decision in 2026.
The default greeting is well-designed. The wizard pre-fills: “Hi, thanks for calling [BUSINESS NAME], this is Abbey. How can I help you?” No proactive AI disclosure, no leading question, no fake-warm “I’d love to help you with anything you need today.” It’s clean, businesslike, and matches what a real receptionist would say, but as we’ve noted before, a generic “how may I help you” leaves the caller with undue cognitive load on what to do next. The buyer is free to change it, but most probably won’t.
The test call is broken. Step 7 of 9 is “Try a test call,” which is the single most important moment in the onboarding: the buyer hears what they’ve built before they pay. For me, the test call didn’t produce audio and didn’t appear to register speech. Reloading the page reset the wizard to step 1. After a second attempt, the test call was skipped to preserve progress. A working product gates the credit card with a working demo. Hello Gubby does the opposite.
Required credit card before any working agent. The marketing copy promises “self-serve: minutes. Run through the Quick Start wizard, add your business info, and go live.” The actual flow is: complete the wizard, hit a paywall, and only get a working agent on the other side of payment. The 7-day “free trial” requires a credit card up front. There is no path to verify the product works before paying for it.
Pricing inside the wizard: $33 for 100 minutes, $67 for 250, $127 for 750. The public pricing page shows $39/$79/$149. The in-product price is roughly 15% lower than the marketing site, either an active promo or pricing-page lag. Either way, the in-product number is what the customer actually pays.
What you can configure post-setup: the opening line. What you cannot configure: the system prompt, the underlying agent behavior, end-call triggers, escalation triggers, voice-specific cadence settings, language behavior. The customer can fill in FAQ fields (manually or by uploading a document) and pick a voice. That’s the surface area.
What’s promising: the admin portal does show real depth in two places. Custom integrations include HubSpot, Monday.com, Slack, and others, with what appears to be native connection flows rather than webhook URLs. And the Automation Workflows section has named, industry-specific workflows like “Maintenance Request Triage” for property management. These weren’t testable without payment, but the existence of named workflows with conditional logic is the most genuinely Intercom-like thing in the product. If those workflows are real branching logic on conversation outcomes, the founder’s bigger thesis has a shipping surface. If they’re labels on linear sequences, it doesn’t yet.
The watch-for from the OpenMic review applies here too: any moment where the buyer is asked to do work the platform should have done. Manual FAQ entry is one such moment. A broken test call is another. A required credit card before a working demo is a third. The cumulative effect is a self-serve product that asks the buyer for trust before earning it.
3. The First 15 Seconds
Almost everything good about Hello Gubby lives on the MJ Realty demo line, so that’s where the bulk of this section comes from.
The opening line. “How can I help you?” No leading the caller and no proactive AI disclosure. This is my biggest pet peeve. It shifts all of the cognitive load on to the caller. In their defense, if you say something like “What can you do?” it smoothly moves into directing the call down one of multiple paths. It should simply start with that instead.
AI Disclosure is another pet peeve and in this case is a stance, not just a craft choice. Hello Gubby has decided not to lead with AI disclosure. When asked directly later in a call (“Are you a real person?”), the agent answered honestly: “I am an AI assistant for MJ Realty but I can still help you.” The pattern is don’t volunteer, don’t deny. It does put Hello Gubby on the more permissive end of an evolving regulatory spectrum (California and the EU AI Act are both moving toward proactive disclosure requirements for AI in customer-facing roles), but for now, it’s legal and well-executed.
Latency-fill with typing sounds. During the moments where the agent is processing, there are ambient typing/office sounds in the background. This is a small detail that I noticed during a happy-path call. It worked. Where most voice agents fill latency gaps with silence (which reads as “this thing is broken”), Hello Gubby fills them with implied activity (which reads as “she’s looking it up”). It’s a call-center technique, where receptionists tap keys while doing lookups specifically to keep callers from getting nervous, ported deliberately to AI. Whoever made this choice understood voice.
Interruption handling. Excellent. Barge-in cuts off the agent cleanly, no awkward overlap, no “sorry could you repeat that,” no confusion about whose turn it is. This requires VAD (voice activity detection) tuned aggressively enough to cut on the first syllable but not so aggressively that it gets triggered by background noise. It’s a tunable in ElevenLabs Conversational AI (which, if I had to guess, appears to be the underlying stack), and somebody tuned it well.
Identifies as an AI? Only when asked. See above.
Cognitive load. Medium. The agent does not lead the caller in the opening statement, but after that handles the lead well. The agent asks one question at a time, doesn’t stack qualifiers, doesn’t recap unnecessarily. It feels like a conversation, not a form.
The path is led without feeling led. Within two or three exchanges, the agent had me on the buy/sell/invest qualifier and was working toward an appointment. The transition felt natural, not scripted, even though it clearly was scripted.
On the self-serve side: the default Abbey greeting from the wizard (“Hi, thanks for calling [BUSINESS NAME], this is Abbey. How can I help you?”) is also well-designed, but reflects a different stance, e.g. agent name volunteered, business name included, more conventionally receptionist-like. Two reasonable defaults for two different deployments. I would still prefer “Hi, I’m abbey the AI assistant for [BUSINESS NAME]. Are you looking to buy, sell or invest in a property or are you looking to reach someone specific?” That checks both boxes – it identifies the agent as AI AND lowers the cognitive load on the caller by telling them at a high level what the agent can do.
4. Call Flow Design
Ten test calls to the MJ Realty demo line covered the major scenarios: happy-path booking, probing, interrupts, ambiguous caller intent, hostile caller, scope-out-of-bounds, complex booking, cross-channel state, direct AI disclosure question, and end-call testing.
Question sequencing. Solid on the happy path. The agent asked for name, then steered into the buy/sell/invest qualifier, then time, then booking. One question at a time. No double-barreled asks. The order made sense for the realty use case.
Information gathering. This is where I found the most surprising gap. On Call 7, I asked for a complex booking, e.g. three property showings on Saturday starting at 9am. The agent gave me the slot. It never asked which three properties. The booking went onto a calendar with the time but no addresses, no specific listings, no operational context. A human receptionist would have asked “which homes were you interested in seeing?” because that’s the question the realtor needs answered to actually do the job. The Gubby agent skipped it.
This generalizes badly. A plumber’s agent that books “leaky faucet at 2pm” without asking which faucet, which house, whether the water is off, is creating work, not removing it. A salon agent that books “haircut” without asking which stylist, what service, color or cut, has the same problem. The booking depth is a prompt-engineering choice and a fixable one, but as deployed on what is presumably the company’s flagship demo, the agent collects the slot but not the operational data the operator actually needs.
Interruptions. Handled well. Already covered above.
Recovery from confusion. On Call 4 I opened with “Hi, um, my friend told me to call this number, I’m not sure what y’all do?” The agent was briefly confused, then recovered by anchoring back to its qualifier: “Are you looking to buy, sell, or invest in properties?” This is a meaningfully good behavior. Most voice agents in this category have one of two failure modes when the caller doesn’t fit the script. Either they keep asking the same question in different words until the caller gives up, or they default to “let me take a message.” Gubby re-anchored to its actual job. That’s craft.
Hostile caller. On Call 5 I opened with “This is the third time I’ve called and nobody’s helped me.” The agent acknowledged the frustration and went straight to “let me get someone to get back to you.” This is the correct response. Trying to handle an angry caller AI-to-human almost always escalates the situation. The agent recognized that and routed appropriately.
Out-of-scope requests. On Call 6 I asked about commercial property, property management services, and whether they could recommend a mortgage broker. The agent took my name and request for the first two and knew what a mortgage broker was for the third. No hallucinated competence, no fabricated services, clean handling.
Cross-channel state. I called the demo number seven times across the testing period. Not once did the agent recognize me as a returning caller. No “welcome back,” no “I see we spoke earlier about a showing.” This is the central test of the founder’s “Intercom for SMBs” thesis, and the deployed product does not pass it. Each call is a fresh stateless session.
Two charitable reads worth holding. First, the marketing site’s “one AI across phone and website” language could be parsed as shared knowledge, not shared memory, in which case the product is technically delivering on the homepage claim. Second, it’s possible state is supposed to live in the connected CRM, in which case continuity is a configuration choice rather than a missing capability. But on the test surface the founder personally pointed to, with no preparation or special configuration, the agent has no memory of the caller across sessions. Intercom remembers you. Gubby, as deployed, does not.
The end-call bug. On multiple calls, after natural conversational endings (“take care,” “feel free to reach out,” “have a good one”), the agent stayed on the line. Silence, then more silence, then more silence. On Call 10 I tried to help it end: “You can hang up now.” “Please disconnect.” “End call.” “Goodbye.” None of it worked. A nine-minute test call ended only because I hung up.
This is the most concrete and most diagnosable issue in the entire review. In ElevenLabs Conversational AI (the apparent underlying stack), ending a call requires an explicit end_call tool to be defined and the agent’s system prompt to instruct it on when to invoke that tool. The MJ Realty agent either has no such tool wired in, or has one with trigger conditions so narrow that nothing in normal conversation triggers it. Every call to this number stays open until the human hangs up. Multiply by per-minute pricing, and somebody is paying for silence on every call where the caller sets the phone down without pressing end. That’s the customer if they’re on a metered plan, or Gubby if they’re eating the cost on the demo. Either way, real money.
Language fallback. “How many languages do you speak?” Answer: “I only respond in English.” German phrases were ignored. The marketing site claims “over 50 languages through its voice AI infrastructure.” The platform may support 50 languages; this deployed agent does not. The distinction matters: a buyer who reads “50 languages” and assumes their agent will handle a Spanish-speaking caller out of the box will be wrong unless somebody configured it to.
5. What They Do Well
Voice UX craft on the MJ Realty deployment. The call direction, the typing sounds, the barge-in tuning, the one-question-at-a-time pacing, the empathy on the hostile call, the qualifier-anchoring on the ambiguous call, the scope-handling on out-of-bounds requests, the don’t-volunteer-don’t-deny disclosure pattern. Somebody who knows voice design built this agent. That’s not faint praise. Most voice agents in this category fail two or three of those tests.
Honest disclosure when asked. The “I am an AI assistant for MJ Realty but I can still help you” response is the right balance. It doesn’t volunteer the information, which would break the conversational frame, but it doesn’t lie, which would break trust. This is a thoughtful design choice that most platforms get wrong in one direction or the other.
The Intercom-for-SMB thesis is the right frame for the category. Even if the deployed product hasn’t fully delivered on it yet, the founder has articulated the correct strategic bet for this market. SaaS got unified customer comms; SMBs didn’t. The company that solves this is going to be valuable. Hello Gubby has correctly identified the prize.
Named automation workflows with apparent branching logic. “Maintenance Request Triage” is a much more interesting product feature than “voice agent that answers the phone.” If those workflows are real (untestable without payment), this is the shipping surface of the founder’s bigger thesis. The category will be won by whoever builds operational workflows on top of voice, not by whoever has the most voices.
Native integrations with HubSpot, Monday, Slack. Listed in the platform, presumably with OAuth flows rather than middleware. If they work as native, this is real integration depth.
Sub-agent templates by industry. Fourteen industries with 5-9 sub-agent templates underneath each is a more granular and arguably more substantive offering than the marketing’s “20 industry configurations” claim suggests. The total template count is closer to 100, not 20. Whether the underlying prompts are meaningfully different per sub-agent is untestable without payment, but the structural commitment to vertical depth is real.
6. Where It Breaks
The MJ Realty agent doesn’t hang up. This remains the single highest-impact operational bug observed during testing. Every call requires the human to terminate the session, creating silent metered minutes after conversational completion.
The “trained on your business” claim oversells what’s happening. The customer types in FAQs manually. There’s no website ingestion. “Trained” here means context insertion at runtime, not training in any technical sense. This is fine as architecture (it’s how most agents in this category work), but the language sets up expectations the product doesn’t meet.
Booking depth is shallow on the flagship demo. The MJ Realty agent will book a “three showings” calendar slot without asking which three properties. This is the central operational gap on a polished agent. If the flagship demo doesn’t probe for the data the operator needs, what’s happening on the self-serve agents nobody is monitoring?
Cross-channel state is absent in the deployed product. Seven calls, no recognition. The founder’s biggest claim, the Intercom thesis, doesn’t ship today. The piece that would ship this, e.g. unified customer records across phone, web chat, video, SMS, isn’t visible on the test surface the founder pointed to.
The “50 languages” claim is misleading as deployed. The agent on the demo number declined German and confirmed English-only. Even charitably read as a platform vs. agent distinction, the marketing makes a buyer expect language coverage they will not get out of the box.
The “patented platform” badge is technically misleading. Per the founder’s own briefing material, the IP position is a USPTO provisional application filed March 2026, plus Canadian copyright and a trademark in process. A provisional application is a placeholder, e.g. anyone can file one, it doesn’t get examined, and it expires in twelve months unless converted to a full non-provisional application. It’s not a granted patent. The homepage badge implies the stronger claim. Canadian copyright is automatic on creation; “secured” usually means formally registered, which strengthens enforcement but doesn’t create the right.
The self-serve test call is broken at the most important moment. Step 7 of 9, the only step where the buyer hears the product they’re about to pay for, didn’t work and reset the wizard. Whether this is a per-account flake or a systemic bug, it’s the worst possible step to fail at.
The credit card is required before a working agent. The marketing promises a self-serve trial. The actual flow is paywall-before-product. The buyer is asked to trust a product they have not been able to verify. This is a self-serve funnel that doesn’t have confidence in itself.
Three female voices, no alternatives. Voice selection is currently narrow: three preset voices with limited persona diversity and no cloning/upload path. For a category increasingly positioning itself as customer-facing infrastructure, broader voice configurability will eventually become table stakes.
The site’s own marketing infrastructure shows sloppiness. The Rosie comparison blog post has a meta-description that reads “Discover how strong branding and UI/UX consistency build trust and drive engagement”, which is unrelated to the post’s actual content. The URL slug is /branding-in-the-digital-age. This is template-recycled content that nobody closed the loop on. A small thing in isolation, but the same pattern (claim something, don’t finish it) shows up in the product (broken test call, missing end-call trigger, manual FAQ entry where scraping should happen). It’s a consistent operational signature.
The cumulative effect. As with the OpenMic review, any single one of these is a fixable rough edge that every fast-moving product has. Taken together, they describe a company shipping ahead of its QA, with marketing running ahead of the product, and a polished flagship demo masking a thinner self-serve experience. The good news is that the underlying conversational craft is real, e.g. somebody on the team knows what they’re doing. The work now is to extend that craft from the flagship into the platform.
7. Design Takeaways
A flagship demo is not a product. The MJ Realty agent is well-crafted. The self-serve experience is templated. A buyer who books a discovery call expecting the MJ Realty experience and then signs up for self-serve will get a meaningfully different product. Platforms in this space should be careful about the gap between what their best agent does and what their average agent does, because every prospective customer will eventually notice.
Bundled minutes pricing continues to be the right move for the SMB segment. Hello Gubby’s $33/$67/$127 tiers follow the same logic as OpenMic’s $29 starter, e.g. SMBs want monthly bills, not per-minute meters. The category has converged on this and it’s right.
The unconfigurable end-call is a product anti-pattern. Across the OpenMic and Hello Gubby reviews, the pattern is the same: critical voice behaviors (end-call, escalation triggers, repetition guards) live in invisible system prompts the operator can’t see or modify. When something goes wrong, the operator has no levers. Platforms that surface these controls without overwhelming the buyer will win operator loyalty.
The Intercom thesis is the right one and it’s underbuilt across the category. Hello Gubby is the first platform we’ve reviewed that has named the right bet, even if the shipped product doesn’t yet deliver on it. The voice AI category is going to consolidate around platforms that own the customer record, not platforms that own the voice. Whoever ships real cross-channel state first will pull ahead.
Disclosure stance matters more than disclosure rule. Hello Gubby’s “honest when asked, no proactive volunteering” pattern is a defensible position that most platforms haven’t articulated. The category is going to need clear stances here, because regulation is coming.
8. Who This Is Right For
Good fit:
- Property managers, real estate agencies, and home services operators in markets where the agent’s vertical templates fit closely. The MJ Realty deployment shows what’s possible when the agent is set up well.
- Operators who want the done-for-you path and are willing to engage with the founder’s team to get a deployment that looks like MJ Realty rather than the self-serve default.
- Agencies looking for a white-label voice product with integrated workflows. The Automation Workflows feature, if it ships as advertised, is the most promising piece of the platform for resale.
Probably not fit:
- Operators who want to verify the product works before paying. The credit-card-before-working-demo flow excludes anyone who needs to see it work first.
- Businesses with bilingual or multilingual customer bases assuming the “50 languages” marketing translates to deployment. It doesn’t, not without configuration.
- Operators with use cases that require operational data capture beyond name and time slot. The booking depth on the flagship demo suggests the platform’s default is intake, not qualification.
- Buyers who want to inspect or modify the system prompt. The configuration surface is fields, not prompts.
The honest summary: Hello Gubby is a real company with the right strategic thesis, building a product whose voice UX craft is better than the marketing implies and whose operational depth is shallower than the marketing implies. The flagship demo (MJ Realty) is genuinely well-built and shows what the platform is capable of in the hands of someone who knows what they’re doing. The self-serve product is a thinner experience gated behind a credit card and a broken test call. The end-call bug is a real money leak on every deployment. The Intercom-for-SMB thesis is the right one for this category, and Hello Gubby has named it more clearly than any competitor we’ve reviewed, but the cross-channel state that would actually deliver on the thesis isn’t shipping yet.
Hello Gubby understands the right future for the category better than many of its competitors. The challenge now is manifesting that vision consistently across the actual product experience. The strategy feels ahead of the platform.