AI support products are easy to demonstrate badly.
Put a chat window on a landing page, ask it a question that appears in the FAQ, show the answer arriving quickly, and call the thing intelligent. It looks neat enough in a sales call. It also avoids the harder questions that decide whether the product is ready for real customers.
What happens when the website has overlapping guidance? What if the same business runs a public marketing site, a help centre and several product pages that use slightly different language? What if the assistant should answer confidently in one case, but send the visitor to the team in another? What if the brand, tone and contact route need to change from one product to the next?
Those are not demo problems. They are product problems.
That is why the latest Solviro in-use page is interesting from a BPS Designs perspective. Solviro is not only described as a managed AI support assistant. It is now live across SwimClub Manager, PlanMan and RiskMate, three products operated by the same team behind Solviro, each with its own audience, content structure and support journey.
The important part is the disclosure. These are genuine live deployments, but they are not independent customer testimonials. BPS is not claiming measured support savings, faster response times or commercial uplift from them yet. That restraint matters. It turns the page from a marketing boast into a useful product signal: before asking customers to trust an AI support workflow, run it in anger on your own products.
Dogfooding finds the awkward edges
Internal product use is not a substitute for customer evidence. It is the layer before it.
When a team dogfoods a product properly, it exposes the rough edges that a clean demo normally hides. With AI support, those edges tend to sit around knowledge quality, setup assumptions and operational ownership rather than model capability.
For Solviro, the three current deployments are usefully different:
- SwimClub Manager has mature SaaS content, feature pages, public FAQs, customer stories and a separate help centre.
- PlanMan has specialist project-management content for architects and town planners, including features, integrations, tutorials and product guidance.
- RiskMate has sector, compliance, pricing, risk-management and FAQ content for governance-focused teams.
That variety matters because support automation is heavily shaped by context. A swimming club volunteer, a planning practice and a risk team do not ask questions in the same way. They do not need the same tone. They do not expect the same fallback route when the assistant cannot answer.
If an AI support product only works against one tidy knowledge base, the team has not learned enough yet.
Live use tests the whole operating model
The strongest AI products are rarely just a model wrapped in a user interface. They are an operating model around a model.
For AI support, that operating model includes:
- choosing which sources are allowed to inform answers
- reviewing content before it goes live
- testing real customer questions before launch
- configuring the assistant’s name, brand and greeting
- deciding when the assistant should stop and use a contact route
- reviewing conversations after launch
- improving the underlying content when gaps appear
Solviro’s public positioning leans into that. The service is managed: the team creates the workspace, connects the content, reviews useful pages, tests answers and helps launch the website assistant. Its answer model is also deliberately constrained around approved content and controlled fallbacks rather than a free-roaming chatbot that tries to sound helpful at all costs.
Dogfooding is where those claims become practical.
It is one thing to say a product supports multiple knowledge sources. It is another to wire it into a real SaaS website and a separate help centre, then see whether the assistant consistently chooses useful source material. It is one thing to offer configurable branding. It is another to make the assistant feel natural as SwimClub Manager Support, PlanMan Support and RiskMate Support without blurring the products together.
The same applies to fallbacks. A good fallback is not failure. It is the product refusing to guess when the approved guidance is not enough.
The honest limit is part of the product
One of the healthiest details in Solviro’s positioning is the focus on an “honest limit”: show the answer, its approved source and the point where a person should take over.
That is the right instinct for customer support. Many customer questions are straightforward. Delivery timings, joining instructions, feature availability, pricing routes, account support and booking steps can often be answered from published guidance. Personal exceptions, account-specific problems and unclear policies usually should not be guessed by a public assistant.
Dogfooding helps the team tune that boundary.
When Solviro runs on BPS-operated products, the team can see whether the assistant is being appropriately cautious. If visitors ask questions the content cannot support, that is useful information. It might mean the fallback is doing its job. It might also mean the product site is missing an answer customers reasonably expect.
That distinction is hard to see in a staged demo. It becomes visible when the assistant is exposed to the messy, repetitive, ordinary questions real visitors ask.
Conversation review turns support into product feedback
AI support should not only reduce interruptions. It should make the product easier to understand.
Conversation history is useful because it shows what visitors actually need, not what the team assumed they would ask. Repeated questions can expose:
- pages that exist but are hard to find
- pricing or trial details that are technically present but unclear
- product terminology that makes sense internally but not to buyers
- support routes that are buried too deeply
- onboarding gaps that keep creating the same confusion
That is where dogfooding becomes more than internal testing. It creates a feedback loop between support, product content and positioning.
A live assistant on SwimClub Manager can reveal different knowledge gaps from one on PlanMan. RiskMate may surface more sector-specific or compliance-related questions. Those differences help Solviro mature because the team is not only testing whether answers appear. It is testing whether the review workflow creates better product knowledge over time.
This is also where BPS Designs’ product-led direction shows. Running several SaaS products gives the team more than a portfolio. It gives them a practical lab for product operations: support journeys, content quality, onboarding, pricing communication and customer trust.
Be careful with the claims
There is a temptation with AI products to publish the boldest possible metric as soon as something works once.
That is usually a mistake.
If a team wants to claim support volume reduction, faster response time, improved conversion or better customer satisfaction, it needs a baseline and a measurement period. It needs to know what changed, what stayed the same and what other factors might have affected the result.
The Solviro in-use page avoids claiming outcomes it has not measured. That is not a weakness. It is the correct posture for an early product proof point.
There is still useful proof in saying:
- the assistant is live on three real products
- each deployment has its own brand and support context
- each deployment uses approved product content
- each deployment has a real fallback route
- the team is reviewing the experience before broader customer rollout
Those are operational claims, not inflated performance claims. They are much more credible at this stage.
Why this matters for other product teams
If you are building an AI product, especially one that touches customers directly, dogfooding should not be a vanity exercise.
It should answer practical questions:
- Can the product handle different customer contexts without bespoke engineering every time?
- Which parts of setup are still too fragile for customers to manage alone?
- Where does the product need stronger defaults?
- What does the team need to review before launch?
- What does support need to see after launch?
- Which claims are evidenced, and which are only assumptions?
That last question is the one that keeps the product honest.
AI products can look more finished than they are because the interface is conversational. A confident answer can hide weak source material, poor ownership and unclear escalation. Dogfooding helps teams slow down in the right places. It forces the product to survive contact with real content, real brand constraints and real support expectations.
For a managed product like Solviro, that is especially important. The promise is not just “the assistant can answer”. The promise is that a business can get from existing support content to a live, controlled customer-support assistant without turning the setup into a project of its own.
That promise has to be tested in use.
The product lesson
The useful lesson from Solviro’s current deployments is not that every AI support product should start with three sister-company case studies. It is that product teams should make their proof match the maturity of the product.
Early on, proof might mean live internal deployments, clear disclosure and a list of operational behaviours the team has tested. Later, it can mean independent customers, measured support impact, customer satisfaction data and stronger commercial evidence.
Skipping straight to big claims is tempting. It is also how AI products lose trust.
The better route is slower and sturdier: use the product, expose it to different contexts, document what is genuinely live, be clear about what has not yet been measured, and let the operating model improve before the marketing gets louder.
That is a very BPS Designs way to build. Not flashy for the sake of it. Product first, evidence close behind, and no pretending the demo is the whole truth.