Consulting · Operations Diagnostic · The Bahamas
Your staff answer the same requests all day. We measure it, then fix it.
Billing questions, contractor onboarding, licence requests, guest cases: high-volume work that ties up good people and still runs late. Twenty Five measures where the time goes, builds a digital clerk that handles the routine part, and runs it with your team. Your people keep the judgement calls.
- Match
Finds the account or fileFrom the name, number or email on the message
- Sort
Routine, prepare, or human firstComplaints, hardship and anything sensitive go straight to a person
- Answer
Drafts from your facts and rulesChecked against your policies before anything goes out
- Record
Logs every stepAn audit trail for regulators, insurers and your own managers
Most AI projects don’t pay for themselves. Ours has to, or we tell you not to start.
Organisations buy tools, run pilots and see little change in the numbers. The ones that get results pick one workflow, measure it, and rebuild it end to end. That is the whole method.
of organisations using AI report a measurable financial impact from it.
McKinsey, The state of AI in 2026: On the road to ROI (August 2026)The return the business case must show before we recommend building. Below that, the report says don’t.
Our rule, written into every DiagnosticSystems you need to connect to start. The Diagnostic and the first clerk run beside what you already have.
Phase 0, explained belowMeasure. Build. Run.
Each step stands on its own. You decide after each one whether to go further.
Measure: the Diagnostic
Four weeks, fixed price. We count the requests, time the work, read a sample of real cases and price what it costs you today. You get a written business case. If it doesn’t show at least three times its cost in return, we tell you not to build.
Build: the first clerk
Ten to fourteen weeks. One workflow, built around your rules, tested on real past cases before it touches a live one, then switched on with your team watching every output.
Run: monthly
We operate it, keep the rules current and report the numbers every month: requests handled, time saved, cases passed to people, and anything that went wrong.
Four weeks. A number you can defend.
You get a short status note every Friday: this week, numbers so far, next week, and anything that needs a decision from you.
Week 1 · Listen
Interviews with the process owner, the people who handle the requests, and your IT lead. We tally requests at the counter and on the phone with a simple tablet form.
Week 2 · Count
We sort a sample of real requests by type from an export of your mailbox or call log. Names, phone numbers, emails and account numbers are masked before anything is stored or analysed.
Week 3 · Price and test
A cost model of today’s work, with every assumption sourced and cross-checked against your headcount. Then a shadow test: the clerk works through your past requests. No customer ever sees it.
Week 4 · Decide
A five-page report: what the work costs, what the clerk would handle, the business case, and a plain recommendation, build or don’t. You keep the model and the report either way.
Work that arrives in volume and follows rules.
Customer and billing inquiries
Why is my bill higher, when is it due, can I set up a payment plan. Answered from the account, with hardship and disputes sent to a person.
Onboarding and access
Contractors, crew and new suppliers: documents collected, checked against the list, chased when missing, and recorded for the audit.
Requests, permits and cases
Licence and service requests, claims, guest cases: sorted, prepared and tracked against their deadline so nothing sits unanswered.
The clerk already exists. We show it to you in the first meeting.
We tested it on 55 realistic billing inquiries against 40 sample accounts, using the real AI model, not a simulation. Here is what it did with them.
Test data: fictional customers and accounts. Your own figures come from the shadow test in week 3 of the Diagnostic.
The safeguards are built in, not added later.
- Facts come from your records, not from the AI. Balances, dates and amounts are looked up by code. The AI writes the sentence; it never invents the number.
- Every reply is checked before it goes out. Against your rules: no promises, no discounts, no advice it isn’t allowed to give. A reply that fails is held for a person.
- Sensitive cases go to a person first. Disputes, hardship, complaints and anyone who seems vulnerable skip the clerk entirely.
- Nothing fails silently. If a message can’t be sent or a step breaks, a named person is alerted. Nothing sits unanswered because a system hiccupped.
- Every language is acknowledged. A customer who writes in Haitian Creole or French gets a reply in their language saying a person will follow up, and the case goes to someone who can.
- Every step is recorded. Who asked, what was found, what was sent, who approved it. An audit trail for regulators, insurers and your own managers.
You own it. Your IT team can take it over.
On your server
The clerk runs on a server you own, inside your network. Records and logs stay there. We don’t host your data.
On software your team can maintain
We build on n8n, a widely used open workflow platform, so your IT staff or any other firm can maintain it. The rules, prompts and test cases are documented and handed over.
An established AI provider
The clerk uses Claude, from Anthropic, which holds SOC 2 Type II and ISO 27001 certifications. In the Diagnostic, personal details are masked before any text reaches it.
Nothing has to be connected to start.
No integration project, no access to your core systems, no waiting on IT.
What does the Diagnostic need from us?
Counts you already have: call and email volumes, a mailbox export, a week of counter tallies, and a few hours of time-and-motion with the people who do the work. No system access.
How does the first clerk run without touching our systems?
It sits beside them. Requests come in through channels we operate for you (WhatsApp Business, a web form, an email inbox, or a forwarded line for unanswered calls), and it reads a nightly export from your system of record into a database on your own server. It does not write back. It can be removed in an hour.
Where does our data live?
On a server you control, inside your own environment. We run the workflow there and keep the records there.
Does it replace staff?
It takes the repetitive part of the queue so your people can handle the cases that need judgement, faster. Every sensitive case goes to a person first.
Can it connect to our systems later?
Yes, once it has proven itself. A live integration is a separate step, scoped and priced on its own.
What if the numbers don’t justify a build?
Then the report says so. If the business case doesn’t show at least three times its cost in return, we recommend not building, and you keep the cost model and the findings to use however you like.
How much of our people’s time does the Diagnostic take?
About an hour with the process owner, twenty minutes with a few of the people who handle the requests, and forty-five minutes with your IT lead. Tallies are a few clicks per request for one or two weeks.
How much does it cost?
The Diagnostic is a fixed price, agreed before we start. The build and the monthly run are priced from what the Diagnostic finds, so you see the business case before you commit to either.
Small and local, on purpose.
Nicolas Serabian, founder
Twenty Five Consulting is based in Freeport, Grand Bahama. You deal with the person who runs the Diagnostic, presents the findings and answers for the build, not an account manager. We also build digital tools for clients in Québec, including a missed-call system that texts customers back in French within a minute.
Start with a one-hour conversation.
We’re based on Grand Bahama. Tell us which queue keeps your team busiest. We’ll show you the clerk working on a live example and what the Diagnostic would measure.