An operator with tools, guardrails, and a chain of command
You edit plain-language settings — company, hours, services, service area, tone of voice, what to escalate, what to never say. Kaboota compiles that into the agent every time you hit save. You never write a prompt.
| Tool | What it does on a live call |
|---|---|
| capture_lead | The one that pays for the whole thing. Called over and over as facts come out: name, phone, address, job type, urgency, decision maker, system age and type, estimated value, source. Streams to your screen mid-call. |
| book_appointment | Checks your calendar's free/busy first, refuses slots you're already in with alternatives, writes the appointment and emails the confirmation. Diagnostic, repair, emergency, estimate or maintenance. |
| transfer_to_agent | Escalation with a handoff cap and hierarchy-first routing — L1 to L2 to L3. Frustrated callers escalate on sentiment, and the next agent gets the context, not a cold start. |
| take_message | Callback capture with urgency, category, preferred time and method. Emergencies flag the session and jump the queue. |
| search_knowledge_base | Services, pricing, service area, hours and brands you carry, so the answer on the phone matches the answer on your website. |
| send_email | Confirmations and follow-ups, sent transactionally the moment the call ends. |
Whisper to the AI mid-call
Watch a call happen in the console and type a nudge — "offer the maintenance plan," "we're booked Friday." Guidance reaches the agent within two seconds and expires when the call ends.
“One question at a time. Always repeat back names, numbers and dates.”
“Politely decline jobs outside the service area.”
“Our voice: warm, brief, no exclamation marks — at 2pm and at 2am.”
“Gas smell → tell them to hang up and call 911.”
Addresses are verified against US Census + USPS before they ever hit a work order.
Beside your people, or running on its own.
The intelligence is identical in both modes — only the autonomy changes. Silent capture, suggest-and-approve, or full execution: it's a setting, not a different product, and you can move one workflow at a time.
Beside your team
AssistiveYour people stay on the conversation. Kaboota runs underneath it — capturing, structuring, enriching and recommending — and acts only where you've said it may.
Your people still own the conversation. It rings, texts and chats exactly where it always did.
Silent capture, or suggest-and-approve. Two settings apart: notes only, or a recommended action waiting for your yes.
The job gets written down properly. Named-speaker transcript, summary, action items, sentiment, structured fields.
Every fact is sourced. Customer, site, equipment, severity, PO — each tied to what was actually said.
Straight into your system of record. A complete, evidence-linked record — no retyping, no swivel-chair.
Fully autonomous
ExecutingKaboota handles the conversation end to end — answers, qualifies, decides, books, dispatches and follows up — inside the guardrails you wrote. Nobody goes to voicemail and nobody gets put on hold.
Answers on the first ring, always. Call, chat or text — on the rooftop mid-job, at 11 PM with the family, on a Saturday. The customer never hears voicemail, and the job stays yours.
Knows an emergency when it hears one. No-heat with a newborn jumps the queue. Gas smell means hang up and call 911.
Books the slot for real. Checks live availability in Google, Calendly, Cal.com or Microsoft, then confirms it.
Pulls the customer's history mid-call. Past service, equipment and warranty from your dispatch system, while they're still talking.
Escalates like a real desk. L1 to L2 to L3 with the context carried over, and sooner if the caller's getting hot.
You can whisper mid-call. Type a nudge from the console and Sam works it in within two seconds.
Thirteen steps from hello to what it earned
Most tools stop at step three. Transcript-to-fields is where Kaboota starts — the value is in everything that happens after the words are on the page.
Identify customer and location
Matched against your records, address verified against Census and USPS.
Transcribe and understand
Named speakers, both sides, intent and sentiment as it happens.
Extract structured fields
Problem, equipment, severity, PO, decision maker — each tied to what was said.
Detect missing information
Flags the blank that will cost you a return trip instead of guessing at it.
Detect safety and exceptions
Gas smell, no heat with an infant, a warranty edge case — caught, not buried.
Enrich with CRM and financials
Service history, equipment age, warranty and margin pulled in mid-conversation.
Recommend the next-best action
The financially and operationally correct move, with the reasoning shown.
Book, route, escalate or approve
Against live availability and your chain of command — or held for your sign-off.
Write evidence-linked records
A complete record into the CRM you already run, sourced line by line.
Prepare dispatcher and tech
The brief that means nobody arrives cold or without the right part.
Capture diagnosis and margin
What was actually wrong, what it invoiced, what got paid, what it earned.
Trigger the right follow-up
Estimate cadence, maintenance renewal, review request — matched to the outcome.
Learn what actually worked
Which conversations and decisions produced the best customer and financial outcome.
Step 13 feeds step 8. The recommendations get sharper every week because the outcomes come back in.
Every extracted field is tied to the exact words the customer said, so you can check it in one glance. If the call never mentioned it, Kaboota leaves it blank instead of guessing.
“the walk-in freezer's completely down, we're losing product”
Every call lands in one bucket automatically, so the board tells the truth without anybody typing notes.
The conversation is the start, not the product
Answering is table stakes. What makes the week is everything that happens in the four days after — and that's in the box too.
Texts that chase for you
Missed-call recovery, photo intake, estimate follow-ups and reminders in one inbox, with AI-written replies you send in a tap.
Campaigns that dial
Import a list, set the retry policy, and let Kaboota work maintenance renewals and reactivations without tying up a person.
Estimate rescue
Open quotes ranked by close odds with the customer's actual objection, plus a six-step cadence from Day 0 text to Day 9 final email.
Above your CRM, not instead of it
Pulls history, equipment, warranty and margin on the call — then writes an evidence-linked record back into ServiceTitan, Housecall Pro or whatever you run.
A conversation is only handled when it reaches somebody who can end it
Every conversation becomes a record with an owner. Not a queue, not a shared inbox — a named desk with the history already attached.
Something's broken
Triaged by severity and equipment, booked against real availability or escalated with the diagnosis attached.
Something's wanted
Qualified, priced where you allow it, and handed over warm with the objection already on the record.
Something's needed
Checked against what you actually carry, with fitment and lead time rather than a promise to call back.
Something's wrong on paper
Routed to the person who can fix an invoice, never to the tech who can't.
Something's moving
Schedule changes, confirmations and cancellations handled without tying up the board.
Something's on a truck
Where-is-my-order, delivery windows, damage and returns answered from the order itself, then routed to the team that can close them.
Eight screens. Every one earns its place.
Warm cream, one green that always means revenue, and numbers big enough to read across a busy office. Built by hand — no template admin, no icon library.
What the phone actually cost you
Conversations that would have rung out, what they were worth at your own close rate, and which hours leak worst.
What got booked, and by whom
Jobs on the board from AI-handled conversations, split by service, location and who approved them.
Where the week went sideways
Repeat callers, escalations, the themes underneath them and which locations are drifting from the rest — with the actual conversation one click away.
One thing to change
A single recommendation with the reasoning shown. If the number stops moving, that's ours to fix, not yours to chase.
"Margin-based" isn't a slogan. It's in the code.
Every voice engine costs a real number per minute. Kaboota picks the cheapest one that still does your job right — and shows you the math instead of hiding it in a bundle.
| Axis | Options · cost per minute | How it's chosen |
|---|---|---|
| Speech-to-text | google-batch $0.003 · google-streaming $0.016 · assemblyai $0.0025 · deepgram $0.0077 · openai $0.003 | Batch is the listen-only default because it's roughly five times cheaper for the same transcript. Dual-channel is premium — it costs twice as much. |
| Text-to-speech | Google $0.0036 · PlayHT $0.0072 · ElevenLabs $0.027 | Better voices cost more, so they're tied to plan. If a voice provider stumbles, Google picks it up mid-call. |
| Realtime voice | OpenAI $0.24 · ElevenLabs $0.08 + model · ConversationRelay $0.07 | Three engines behind one contract. The brain behind ElevenLabs is pluggable — GPT-4o, Claude or Gemini. |
Clamping with receipts
Ask for more than your plan covers and Kaboota quietly corrects it, then tells you which knob it moved and why.
Advice both directions
Over 90% of your minutes, we say upgrade. Under 20%, we say downgrade — if it saves you more than ten bucks a month.
Bring your own keys
Already have vendor accounts? Use them. We charge a platform fee, your keys carry the usage, and they're stored AES-256-GCM encrypted.
Multi-tenant to the bone
Your whole experience is five settings deep. That's why one system runs a quiet listen-only pilot and a full AI front desk without anyone rebuilding anything.
What you see
Decides your answer mode and exactly which screens exist. Anything not listed is revoked — the safe default is listen-only.
What it costs
Voice engine options, minute caps, how many calls at once, outbound allowance.
Who answers
Voice, speech mode, tools it may use, where it sits in the chain of command.
What it knows
Profile, hours, services, service area, escalation rules, guardrails, greeting, disclosure.
Where it rings
Voice mode, dial-out target, consent line, messaging channels. We hold the numbers.
It can't answer about a part it has three prices for
Most wrong answers on a phone line aren't reasoning failures. They're data failures — the sheet said one thing, the site said another, and the counter said a third. Kaboota keeps one released version of what you sell — dimensions, finishes, stock, delivery, financing, warranty — and answers from that, on the phone and on the floor.
Take the file however it arrives
Supplier spreadsheets, PDFs, a column renamed without warning. It gets normalised instead of retyped.
You own the final answer
When the supplier's sheet and your own pricing disagree, yours is the one that ships. The source stays attached so you can see who said what.
Decide what the AI may claim
Approved language per product: what can be stated flat, what has to be qualified, what it must never promise.
One version reaches every channel
Site, quote, counter and phone read the same released record, so nobody gets three prices for one part.
Options and exclusions, not flat SKUs
Dimensions, fabrics, finishes and configurable units with real dependencies, so the AI can walk someone through valid combinations instead of guessing.
Stock where the customer is standing
Availability per location from the live feed — the answer is for their store, not the warehouse — with the alternative offered when it isn't there.
Delivery, financing and warranty, answered flat
Windows, finance terms, warranty and return policies as approved language per product, so the floor and the phone quote the same terms.
Guided selling, not a search box
It asks what the room or the job needs, narrows to the valid options, and books the showroom visit or the site survey on a real calendar.
Post-sale is the same conversation
Delivery-day questions, damage claims and warranty service land on the desk that owns them with the order and the history attached.
Publish-ready, not just answer-ready
SEO titles, meta descriptions, schema and a Google Shopping feed come out of the same released record, so the site and the phone can't drift apart.
Images and enrichment from the same file
Product copy enriched, images processed, every SKU validated before it publishes — a catalog that gets easier with each supplier update, not harder.
Ready for the shopping agent
Machine-readable, complete and consistent, so an AI agent shortlisting on a shopper's behalf includes you instead of skipping the contradiction.
Keep the systems you run. We sit above them.
Nothing here is a logo wall. Each row says what state it is actually in, and a row that says coming means it is not shipped.
| Layer | What connects | Status |
|---|---|---|
| Telephony | We hold the numbers and the carrier relationship. Star-code forwarding, SIP, SMS and MMS. | LIVE |
| Field service, CRM & DMS | ServiceTitan, Housecall Pro, Jobber, HubSpot and Salesforce-class CRMs, dealer DMS and service schedulers — read during the conversation, written back after it. | LIVE |
| Calendars | Google, Microsoft, Calendly and Cal.com, checked for real free/busy before anything is promised. | LIVE |
| Messaging & reviews | Instagram, Facebook, Messenger, WhatsApp Business, TikTok, Google Business Profile and Yelp into the same inbox as the phone. | LIVE |
| Commerce, catalog & inventory | Shopify, BigCommerce, WooCommerce, Magento, ERP, PIM, POS and supplier CSV/JSON/XML feeds — products, parts, pricing, stock per location — in whatever shape they arrive. | LIVE |
| Google ecosystem | Business Profile sync, Ads conversion tracking, GA4 events and Shopping-feed output, so what the AI booked shows up where you already measure. | LIVE |
| Your website | Embeddable chat and voice widgets, a JavaScript SDK and a headless API when you want your own UI on top. | LIVE |
| Accounting & payments | Invoice and payment status back against the job, so margin is a fact rather than a guess. | COMING |
Proven in the trades. Not limited to them.
Nothing in the pipeline is trade-specific. A showroom, a service counter and a dispatch desk are the same problem wearing different vocabulary — a conversation that has to end in the right action, booked against real availability.
Mechanical & HVAC
Where it was proven first — emergency triage, equipment history, maintenance renewals.
Plumbing & electrical
Same pipeline, own guardrails, own safety exceptions and escalation rules.
Roofing & restoration
Storm-season surges absorbed without adding a phone for six weeks.
Multi-location & franchise
Several brands, several boards, one intelligence layer with per-location hours, inventory and rules.
Furniture, mattress & design retail
Products, dimensions, stock, delivery windows, financing and warranties answered on the floor's behalf, and the visit booked.
Automotive dealerships & service
Sales enquiries and bay booking at the counter's worst hour, fitment and trade-in questions, parts-ready follow-up.
Transit & paratransit
Ride bookings, confirmations and changes handled start to finish, after hours included, where a missed call strands somebody.
Wireless & telecom stores
Plan comparisons grounded in the customer's actual usage, routine triage before a person, the store visit booked from the same conversation.
A number without a window is a decoration
Every figure carries its window
A number without a date range is a decoration. Ours say whether they're one week, one month or trailing ninety days.
Measured and projected are different words
Anything extrapolated to a year says so on its face, next to the pace it was extrapolated from.
Status is stated, not implied
Live and coming mean two different things. You should never discover from a support ticket that a capability was a roadmap slide.
The transcript is the receipt
Every claim on the board opens into the conversation that produced it. Nothing rolls up into a number you can't take apart.