No SQL, no tickets
Staff ask in their own words. Nobody writes a query, and nobody waits on a report that will be out of date by the time it arrives.
Conversational analytics for pawn lending
Ask your data a question. Get an answer, not a report request.
Your staff already know what they need to know — who is overdue, what this customer owes, what is due this week. PawnTeller answers those questions from your live system in seconds, without anyone writing SQL or waiting on a developer.
Reads your live database. No export, no second copy of your customer data.
A question, end to end
Counter staff
Who is overdue at my branch?
The assistant proposes
SELECT customer_name, due_date, due_amount FROM payment_schedule …
PawnTeller adds, before it runs
AND business_id = 7
AND branch_id IN (your branches)
Checked & scoped · read-only
Answer
Four customers are overdue, owing $2,180 in total.
| Customer | Due | Amount |
|---|---|---|
| J. Alvarez | 12 Aug | $640.00 |
| M. Okafor | 14 Aug | $512.50 |
| R. Whitfield | 19 Aug | $725.00 |
| S. Nair | 21 Aug | $302.50 |
The gap
Dashboards answer the questions someone thought of last quarter
The questions that actually come up at a counter are endless and specific, and each new one costs a developer ticket and a week. Meanwhile handing an AI direct access to a lending database is not an option either — it can invent a number, join the wrong tables, or read another branch's book.
PawnTeller is the layer in between. The model proposes a query; deterministic code decides whether it runs.
Scoped before it runs
Business and branch filters are added after the model has finished writing. It never writes them, and it cannot remove them.
Answers in seconds
Fast enough to use with a customer standing at the counter, rather than something you go back to your desk for.
Inside the screen they already use
Embeds into your existing point-of-sale page. No second system to log into and no new habit to teach.
In practice
The questions a counter actually asks
Every one of these is answered from your live data, scoped to the branches the person asking is permitted to see.
Collections
- Who is overdue today?
- What is due this week?
- Show unpaid installments over 30 days late
- Which loans have a missed payment?
- What is the total amount overdue?
Customers
- Is there a customer named Jane Smith?
- Which customer owes the most?
- Show me the loans for customer code CUST-…
- How many customers have activity at this branch?
Portfolio
- How many loans are active?
- What is the total outstanding balance?
- What is the average loan amount?
- How much have we been paid in total?
- How much was collected in fees?
Collateral
- What is held against ticket number P-1001?
- Which items are worth more than $500?
- How many items are currently on loan?
- Look up serial number SN-…
And when it doesn't know, it says so
Ask for a figure nobody has defined — a risk score, a ratio you have never specified — and PawnTeller tells you there is no configured rule for it, then offers the related facts it can show. It does not guess. A confidently wrong number is worse than no number, because nobody checks it.
How it works
The AI writes the question. Our code decides the answer is allowed.
That split is the whole product. Everything the assistant proposes passes through checks it has no way to influence.
-
Our code
Is this answerable at all?
Before a single word reaches the model, PawnTeller checks the question against the terms your configuration defines. Undefined concepts are refused here.
-
The assistant
Proposes a query
It sees your table and column names and nothing else — no credentials, no hostname, no data from any other business.
-
Our code
Checks every part of it
The query is parsed properly, not pattern-matched. Anything outside your approved tables and columns is rejected, as is anything that writes, deletes or reaches sideways into another table.
-
Our code
Adds the boundaries
Your business id and the asker's permitted branches are attached to the query here — after the model has finished. If they cannot be applied, nothing runs. There is no override and no fallback.
-
Our code
Runs it, read-only, over an encrypted connection
Against your live database, with a row cap and a timeout. Every query is recorded with who asked and what ran.
-
The assistant
Writes the sentence above the table
A short summary of the rows that came back — which are shown in full beneath it, so the answer can always be checked against the data.
Questions we get asked
Frequently asked
Can the AI change or delete my data?
No. Only read-only SELECT statements are permitted, and anything else
is rejected before it reaches your database — not by asking the model to behave, but
by a parser that refuses to pass anything else through.
Can one branch see another branch's customers?
No. The branch and business filters are added to every query after the AI has finished writing it. The model never writes them and cannot influence them. If those filters cannot be applied, the query does not run at all.
What happens if the AI does not know the answer?
It says so. PawnTeller answers only from terms your configuration defines. Ask for a figure nobody has defined and it tells you there is no configured rule for it, rather than inventing a plausible number.
Does PawnTeller need a copy of my database?
No. It reads your live database over an encrypted connection at the moment a question is asked. There is no export, no sync, and no second copy of your customer data to keep secure.
Which database systems are supported?
MySQL today, connected over TLS with a certificate check. Support for other systems is planned — write to [email protected] if yours is different.
How much does it cost?
Pricing is being finalised ahead of launch. The plans and what each includes are published now — write to [email protected] and we will tell you as soon as pricing is set.
See it against your own data
The most useful demonstration is your questions on your database. Tell us what your counter staff ask most and we will show you PawnTeller answering it.