ONTACE · THE CUSTOMER ONTOLOGY
One truth for every customer.
Cited facts.
ONTACE turns scattered customer data into one living world per customer — and decisions that show their evidence.
In production for a leading Indian NBFC — 200,000+ delivery reports a day on live customer traffic.
One customer's world
worked example · as the model sees it
- World
- 20 facts
- 5 beliefs
- Decision
world › 5 object types · 5 link types · 2 text columns read
One truth per customer
Five systems, five answers.
ACE keeps one truth.
Every system has its own version of the customer. ACE merges them into one.
What five systems say
- Loan system
- days_past_due 12 · bucket 1–30
- CRM
- segment B · language English
- Collections notes
- “salary late, will pay after the 5th”
- SMS vendor
- DELIVERED · no reply in 6 months
- WhatsApp vendor
- read 19:04 · replied twice
The campaign tool reads one of them. Reminder #4 goes out by SMS at 9am, in English.
The one world ACE builds
- Customerpreferred_languagethe person, once
- Loan ×2borrowed_bywhat is owed, and how late
- Payment ×11repayshow they actually pay
- PromiseToPay ×3promised_onwhat they said they would do — and did
- Communication ×14sent_towhere they last replied, not just received
- collections_noteread by extraction“salary late this month, will pay after the 5th”
Every decision that follows can be traced back to one of these lines.
ONTACE · the ontology
Describe your business once.
Every customer becomes a world.
Pick a goal. ACE maps it onto your tables and builds a world for every customer.
Recover overdue instalments: who will pay on their own, who needs a reminder, who needs a call or a plan.
corebfsiemi_collection
Object types · 5
Customercustomers
preferred_language · contactable_channels · customer_active_status
Communicationcommunications
channel · sent_at · status · direction
Loanloans
overdue_amount · days_past_due · arrears_bucket · emi_amount
Paymentpayments
amount · paid_at · method
PromiseToPaypromises
promised_amount · promised_date · status · made_on
Link types · 5
- Communicationsent_toCustomer
- Loanborrowed_byCustomer
- PaymentrepaysLoan
- PromiseToPaypromised_onLoan
- CustomerdisputesLoanread from text
Text columns read by extraction
- loans.collections_note
- communications.reply_text · dated by sent_at
Questions
20 facts · 5 beliefs
Beliefs it forms
willingness_to_payability_to_payreachabilityescalation_riskbest_next_step
Counts as converted
payment received against an overdue instalment
within 7 days
Seven goals, ready to switch on.
one ontology · every goal shares one Customer
EMI collection
Lending
Subscription renewal
InsuranceTelecomSaaSAI subscriptionsE-commerce
Sales conversion
BFSIInsuranceTelecomE-commerce
Upsell & cross-sell
BFSIInsuranceTelecomE-commerceSecured-loan leadsUnsecured-loan leads
Retention
BFSITelecomE-commerce
Engagement
BFSITelecomE-commerceHealthcare (non-clinical)
Appointment booking
Automobile visitsInsuranceHealthcare (non-clinical)
Your tables, as they are. AI drafts the mapping; a person approves it
One Customer, every goal. Collections and renewals share one truth
Notes read in time. “Tomorrow” means the day it was written
ONTACE · the knowledge chain
Every decision carries
its receipts.
Ask why any customer got any message — and get the evidence, not a shrug. Pick a stage.
Worked example · EMI collection pack
01 · World · C-3f2a91
One graph per customer, from the pack
Records joined, notes read, refreshed on schedule — every 30 minutes if you want.
- Loan → Customerborrowed_byjoined on loan.customer_id
- PromiseToPay → Loanpromised_onjoined on promise.loan_id
- Communication → Customersent_tojoined on communication.customer_id
- collections_note“salary late this month, will pay after the 5th”citedread by extraction · dated 2 Oct
An independent judge
A second model checks every answer. No answer, no pass.
People before go-live
30 human reviews per question. A person flips the switch.
The model never sees a name
Names and numbers are pseudonymised before any prompt.
Ask it
Plain-language questions over every world
Wire in your agents
MCP, with the console's permissions
See the graph
Every customer's world, visual
Take it with you
Every stage exports to CSV
Built for where you are
Same engine.
Your starting point.
Live in days. No data team.
30 days free
DPIIT-recognised startups and Shopify stores, up to 10,000 users
Try ONTACE free for a monthUpload and go
CSV, Excel or a Google Sheet
Goals ready to switch on
Seven packs, pick one
Your own AI key
Model costs stay on your bill
Start small
₹0.40 per user, or from ₹50K a month
One truth, two halves
ONTACE knows. COMACE acts.
Both write to the same truth.
Start with ONTACE. Add COMACE when you want the decisions sent for you.
ONTACEthe ontology layer
One graph per customer from the data you already have — then a decision, with its evidence.
COMACEthe goal engine
Give it a goal. It works every channel until the number moves.
ONTACE alone
You keep your entire communications stack. ONTACE reads your data, decides, and hands the decision back for your own systems to act on.
What runs
- A customer graph built from your own records, however they are keyed
- Up to twenty cited facts and five beliefs per customer, per goal
- One decision, with the evidence attached
- Predictions that return insufficient_evidence rather than guess
ONTACE · deciding with evidence
The system knows what it doesn't know.
And it tells you.
ACE shows the evidence behind every number — or says it doesn't know. Pick a situation.
The situation
Six months in, payments connected. The real outcome has evidence.
Worked examples · the behaviour is the product
Metric ladder
most outcome-proximate first
- Payment rate (objective's conversion metric)4,812 matured outcomesused×1.0
- Reply rate12,004 sends×0.5
- Read rate9,340 WhatsApp sends×0.3
- Delivery rate12,004 sends×0.1
Rung 1 · payment rate
WhatsApp, Tuesday 18:00–21:00
Chosen on the outcome itself, not a proxy.
Confidence 0.87 — undiscounted
No uplift column exists
Followed and ignored advice, side by side. No invented lift.
Confidence is sample adequacy
How much evidence there is — not a claim it can't back.
Deterministic and reproducible
Same input, same answer. Every decision can be replayed.
COMACE · the communication half
Then it acts.
Any channel. One delivery truth.
The add-on that sends what ONTACE decides — and writes back what happened.
200,000+ delivery reports a day, in production for a leading Indian NBFC
WhatsAppSMSRCSPushAI voice callsEmail via your systems
Decides on evidence, or says it can't
Picks channel and time from real outcomes — or says it doesn't know yet.
One journey across every channel
WhatsApp fails, SMS follows. A reply anywhere stops the rest.
Any provider, one delivery truth
Every vendor's report, one vocabulary. Swap vendors freely.
Goals, not campaigns
Consent, caps and quiet hours checked before every send.
Attribution you can audit
Each payment traced to the message that drove it.
PACE does the work
Three agents find, plan and send — inside your rules.
New · voice
AI voice calls, briefed by ONTACE.
The bot knows who it's calling and why — only where your goal asks for a call.
- DND checked before every call
- Missed call? A message follows on the next channel
- Every call analysed: outcome, sentiment, summary, transcript
- A promise holds messages to its date; a dispute stops contact
Call brief · C-3f2a91worked example
- Goal
- Recover overdue EMI
- Ask
- Pay after salary credit
- Why
- Kept 3 of 3 past promises
- Likely
- High
- Language
- Hindi
- Last call
- “Call me after 6pm”
After the call: promise to pay on the 6th — messages held until then.
COMACE · ACE Live
Watch an autonomous system operate.
A single opportunity, end to end: detection, strategy, policy checks, execution, events and the business outcome — all in one journey context.
ACE Live
streaming · demo dataGoal
● ActiveIncrease qualified conversions
Target
25%
Current
18.7%
▲ 4.8% this week
Opportunities
0
Actions
0
Conversions
0
Simulated demo data
Solutions by industry
The same engine. A different argument every time.
Pick yours.
Lending & collections
An EMI is twelve days overdue. The existing system fires generic reminder #4 at 9am, in English, to a borrower who has never once answered a morning SMS.
1What ACE notices
- Missed 2 of 12 EMIs — both in months the salary credit landed late
- Kept 3 of 3 past payment promises
- Reads WhatsApp within two hours; has never opened a morning SMS
2What it does
WhatsApp, in Hindi, a respectful reminder with a payment link, at 7:40pm. Timing wins the outcome here — the approved template stands unchanged.
What moves
Recovery rate ↑Contact cost ↓Audit trail per messageMeasuredAt a leading NBFC: ₹5 Cr due, ₹3.5 Cr collected in a three-day run — 70% recovered against 46% the month before. Live in seven days.
Worked examples. The lending result is measured; the others follow the same pattern.
Pricing
Three ways to buy. Priced for India.
Enterprise decision platforms cost crores. ONTACE starts at ₹50K a month.
Free for a month
Shopify stores and DPIIT-recognised startups run free for 30 days.
- Up to 10,000 monthly active users
- One goal, live, against a control
- Your own WhatsApp or SMS vendor
- No card, and no auto-renewal
Share your store URL or DPIIT number in the form.
Monthly, no lock-in. Volume moves the per-user rate down, not the plan.
ONTACE
Start hereThe ontology layer, into your own stack
From ₹50Kper month
Flat subscription by user band, not per message.
- SMEs and online stores
- From ₹50K / mo
- Mid-market
- ₹1–3 L / mo
- Enterprise
- Custom
- One graph per customer from your records
- Decisions with evidence, over API or webhook
- Your comms stack untouched
ONTACE + COMACE
The full loop — decide, execute, learn
≤ ₹0.80per monthly active user
Bundled to land, and unbundlable later if you only want one half.
- Everything in both plans
- Outcomes feed the next decision
- Drop either half later, no rebuild
COMACE
The goal engine, on your own vendors
₹0.40per monthly active user
Usage-based, degressive to ₹0.25 as volume grows.
- Goals across WhatsApp, SMS, RCS and push
- Consent and quiet hours enforced before send
- One delivery truth across every vendor
Plus a one-time deployment fee. Priced per active user, not per message — channel spend stays on your own vendor contracts.
What could your data
decide tomorrow?
>
Get connected
Tell us the outcome.
We will show you the engine.
A short form, a real reply.
What happens next
- 1A person reads itNot a router. Usually within a working day.
- 2We map your outcomeWhat ACE would target, on which channels, measured how.
- 3You see it runningA walkthrough on real data, not slides.