Every customer conversation says what people wanted. Most businesses never hear it.
Unknown Ops turns those conversations into knowledge the agent uses, and revenue the business would otherwise miss. Owners keep it current with a one-line reply.
Lodging is where we start, not where we stop.
The agent fails quietly, and nobody is assigned to fix it.
Small businesses are deploying AI agents for calls, chat and SMS. Usually an agency or a vendor sets one up once, and then it is left alone.
When it doesn’t know something, it says “I don’t have that information” and the conversation is logged as fine. Nobody reads the transcripts, and there is no support team or developer to update it. Every agent deployed widens the gap.
A new kind of failure, and nobody assigned to it.
AI agents are new, and they fail in a new way: the agent deflects, the conversation is logged as a success, and nobody notices. Small businesses deployed agents faster than they built any way to keep them current. There is no support team to read transcripts and no developer to update the agent. Meanwhile every conversation is a record of what customers wanted to buy.
The channel between the owner and their agent.
Finding gaps is how a customer starts. The product is what keeps the agent right afterwards. After setup, the owner never needs to log in. Email is the interface today; text is next.
- 01
Find
TodayReads every conversation across vendors and surfaces what the agent couldn’t answer, ranked by how often customers asked.
- 02
Capture
Early accessFinds customers who wanted to book, buy, schedule or get a quote and didn’t, with the contact details they gave, so the business can follow up.
- 03
Fix
Early accessAsks the person who knows, usually the owner or a shift manager, and publishes their one-line reply to the agent.
- 04
Anticipate
NextAsks ahead of seasons, holidays and changes, so the agent has the answer before customers ask.
- 05
Inform
Today in reports, alerts nextTells the owner what customers want that the business doesn’t offer, with counts.
From what was said, to what the business knows, to what it should do.
These are not three products. Every interaction produces signals, the knowledge layer interprets them, and the revenue layer acts on them. Outcomes flow back and make the next read better.
Turn customer intent into revenue.
Missed leads with contact details, unmet demand with counts, and the follow-up each one needs.
Early accessWhat the business knows, and doesn’t.
Knowledge gaps, sorted into facts to supply, tools to build, and offers the business doesn’t have. Kept current by the owner’s replies.
TodayWhat happens across every conversation.
Voice agents and website chat today. SMS, messaging and human calls next, across every vendor a business uses.
Today: ElevenLabs, RetellBought by whoever builds the agent. Used by whoever owns the business.
The buyer is whoever builds and runs the agent: an agency, an implementer or an AI receptionist vendor. They connect it once, and they need proof the agent earns its fee.
The user is the owner or manager who knows the answers. They are rarely technical and never log in: one email a week, one-line replies. Multi-location operators with an operations lead can buy directly. Larger companies with support teams are a later market.
One vertical at a time, then through agencies.
Independent hotels and short-term rentals
The agents are already live, the questions repeat, and an owner or front-desk manager knows the answers.
Medspa and dental
Same shape: a busy front desk, repeat questions about prices, policies and hours, and an owner who knows the answers.
Home services
Plumbers, HVAC and cleaners are putting agents on their phones. The questions are about service areas, pricing and availability.
Distribution: agencies, implementers and receptionist vendors connect once and roll Unknown Ops out to every client. They pay per client agent and usually include it in their own fee. Each new vertical uses the same method with a new knowledge domain.
Why lodging first is a bet: we expect most questions guests ask a property’s own agent to be facts an owner can supply in one reply. That is a hypothesis, not a finding. The hotel gaps in our public test run were mostly booking lookups, which need a tool. Real property conversations are what test it.
Build in order. Start with conversations that already happened.
- 1Knowledge gap detection
Find the questions the agent missed.
Today - 2Customer intent
Missed leads and unmet demand, from the same read.
Early access - 3Lead qualification
Which missed leads are worth a call back first.
Next - 4Revenue activation
Follow-ups and routing into the tools a business already uses.
Later - 5Revenue optimization
Learn which actions produce revenue, and do more of them.
Later
Inbound first. Every lead starts as a customer who already reached out. Unknown Ops is not an outbound prospecting tool.
Intelligence layer first. Unknown Ops sits underneath the agents, CRMs and tools a business already has, rather than replacing them. Owning parts of the follow-up workflow comes later, selectively.
What the platforms aren’t built to do.
Across vendors
Voice, chat and SMS often come from different vendors. Each sees only its own conversations. We see all of them.
Built for owners
Enterprise tools assume a support team that drafts fixes. We assume a busy owner with a phone, and ask them one question at a time.
Compounding
Gaps repeat across businesses in the same vertical, so each new property can start with a list of likely questions. We reuse question patterns only, never another customer’s answers or transcripts. A hypothesis until we have customers to show it.
The owner’s habit
Once replying to us is how the agent gets updated, we are hard to remove. That is the bet the pass marks below test.
Three things, tested in public.
Before the first pilot, we will set a pass mark for each and publish it here, the same way every threshold in our public test run was set before the data was read. Results follow.
Owners reply.
Share of owners or managers who answer a question within 24 hours
One reply closes the gap.
Share of ENTRY gaps closed by a single reply
They keep replying.
Share of owners still answering questions in month two
No customer results yet. This section fills in as pilots run.
The engine works on data it has never seen.
We ran it on a public research corpus with every threshold locked before the data was read. Thousands of phrasings came back as a short, ranked list of gaps.
Read from 10,437 public dialogues · 143,044 turns. Full method and every gap →
What this does not show
MultiWOZ is a research corpus with a bounded database; its misses are mostly database misses, not production knowledge-base gaps. 38 of the 62 named gaps are absent-cuisine rows from an ontology that lists cuisines with no venues. This demonstrates that the engine works on data it has never seen. It does not demonstrate demand.