OUR WORK / BUILT BY XPLORATECH
Three business problems. Three engineered solutions.
A call after hours, a finished job, an email to the office. For each, we engineered a working system: AI understands what people say and write, software makes the process dependable, and a person stays in charge of what matters. Then we tested the whole journey, live.
A narrated walk-through of the three Fernhill demonstrations, shown from recorded test runs.
READ THE TRANSCRIPT
At XPLORATECH, we believe there's a better way to work. Running a busy service business means dealing with constant interruptions: customer calls, engineer paperwork, emails that need attention. Each one takes time, and when the office is busy, important information can be delayed or missed. At XPLORATECH, we build practical AI and automation systems designed to help businesses manage that workload. To demonstrate what's possible, we've created three working automations for a fictional plumbing and heating company, Fernhill. First, customer calls. When someone phones outside office hours, an AI voice assistant can answer, understand the reason for the call, and guide the conversation.
But answering the telephone is only part of the process. The system also structures the information and routes it according to the type of enquiry.
For an ordinary service request, it prepares the details for office follow-up. For a potential gas safety concern, it follows a separate, safety-focused process. Here, Fernhill Operations, an illustrative system of record, shows the recorded outcomes of both tested calls. For a real business, results go to the systems it already uses, as agreed for each client. Next, engineer paperwork. After finishing a job, an engineer submits a simple completion form. The automation interprets the notes, identifies the work carried out, and updates the corresponding job record.
In this example, the engineer marks job FH-1042 as completed, and the automation records the materials used and the follow-up needed. And when a submitted job number isn't recognised, the system doesn't guess. It adds the submission to an office review list instead.
Finally, the office inbox. Not every customer email should be handled in the same way. The Fernhill automation classifies incoming messages and applies different handling rules.
A routine boiler service enquiry can have a draft prepared for staff to review.
A customer complaint is marked for careful attention, and a message suggesting a gas safety concern is left for manual handling, rather than receiving an automatically generated reply. Importantly, the system does not send these replies automatically. Staff remain responsible for reviewing and sending responses.
These are three working automations, each designed to solve a practical business problem. Here, their recorded results are brought together in one illustrative view. Together, they demonstrate how XPLORATECH combines artificial intelligence with reliable business automation to reduce administration and help teams stay organised.
At XPLORATECH, our approach is straightforward. Use AI where it adds value. Use reliable automation to control the business process, and keep people involved where judgement matters. Less repetitive administration, clearer information, more time to focus on customers.
XPLORATECH. AI automation, engineered to evolve.
BEFORE ANYTHING WAS ENGINEERED
Three ordinary moments. Each one leaves work waiting.
Every trade business knows these moments. Nobody is doing anything wrong: the work simply waits for someone to be free to pick it up. That waiting is what each solution was engineered to remove.
- 21:47
A customer calls after hours.
The office is closed and nobody is free to pick up. The issue waits until morning, the customer tries another firm, or a message has to be pieced back together before anyone can act on it.
- Waits until morning
- Customer may go elsewhere
- Message to reconstruct
- 16:30
An engineer finishes a job.
The job itself is done, and the engineer has written it up naturally, in their own words. Before the business record is useful, someone still has to pick out the work done, the parts used and what needs a return visit.
- Free-text notes
- Parts to record
- Return visit to arrange
- 09:12
A customer emails the office.
Someone has to read it, understand it, decide what needs doing and prepare a response before the work moves on. No two emails are written the same way, so none of it can be done by rule alone.
- Read it
- Decide what it needs
- Prepare a reply
DEMONSTRATION 01
BUILT & TESTEDA customer calls after hours.
From an unanswered call to a structured business outcome.
AI-ASSISTED OUT-OF-HOURS CALLS
Who is responsible for what
AI
Interprets what the caller says, in their own words: the fault, the property, what is affected and how to get in.
CODE
Decides which question comes next, every word the caller hears, when the call ends and what reaches the business.
HUMAN
A Fernhill person picks up a structured request. A safety concern goes to whoever is on call.
WHAT THE DEMONSTRATION SHOWED
Conversation in, request out
A caller described a fault in their own words, and the office received a structured service request.
A constrained safety path
A gas concern left the conversation for a fixed route: approved wording, then an alert to on-call.
Proven on live calls
Both journeys were tested end to end on a real phone line, through to what the business received.
CHOOSE A SCENARIO
- SYSTEMCustomer calls after hours
- AIWhat they say is interpreted
- CODECode chooses the next question
- CODEEvery answer checked for safety
- CODECode writes the summary
- HUMANThe office picks it up
SERVICE REQUEST
The fault, the property, what is affected and access, with a summary written by code. No visit or time is promised.
TESTED END TO END
A good conversation is only the first check
A call can sound perfect and still leave nothing for the office to act on. End-to-end testing showed exactly that case. The AI conversation succeeded; the business process did not. So every journey is verified beyond the conversation: the structured output, its validation, the routing and the handoff, through to a request the business can use.
THE CONVERSATION
- Caller answered
- Sensible questions
- A polite close
THE BUSINESS PROCESS
- Structured output
- Validated
- Routed and handed off
- A request the office can act on
Each side is verified separately. Both journeys were proven end to end on live calls.
THE LESSON
A successful AI interaction is not the same as a successful business process.
BOUNDARIES
- A private demonstration line, not a public service.
- It does not diagnose faults, book visits or promise times.
- The office and on-call are simulated receivers. Live safety proof covers gas.
DEMONSTRATION 02
BUILT & TESTEDAn engineer finishes a job.
From an engineer's notes to a structured business record.
AI-ASSISTED POST-JOB ADMINISTRATION
Who is responsible for what
HUMAN
The engineer chooses the job and its status, and writes notes the way they would say them.
AI
Interprets the notes: the work done, materials used, follow-ups and anything unclear.
CODE
Validates the reading against what the notes actually say, controls which parts of the job record change, sends unknown jobs to Review and flags unclear readings for the office.
WHAT THE DEMONSTRATION SHOWED
Ordinary notes, understood
AI read notes written the way engineers write them and found the work, the parts and the follow-ups.
Software owns the record
Every reading was validated by code before the job record changed, and only permitted fields were written.
Unknown jobs to Review
A job number that matched nothing went to the office’s Review list instead of being written anywhere.
CHOOSE A SCENARIO
- HUMANEngineer submits the notes
- CODECode finds the job
- AIAI reads the notes
- CODEReading checked against notes
- ACTIONPermitted fields updated
JOB UPDATED
Materials, follow-up and whether the office should look, recorded. The engineer’s own notes kept word for word beside them.
BUILT IN
Every quantity has to be in the notes
AI can read a note well and still add something that was never written. So code checks the reading before anything is written. A quantity the engineer did not write is refused: the structured fields stay empty, the job is flagged for the office and the original notes are kept.
THE ENGINEER WROTE
Synthetic demonstration job FH-1042
Boiler losing pressure. Found PRV passing, swapped it and repressurised system. Tested all ok. Used 1 x 3 bar PRV from van. Customer mentioned rad in spare room still cold - needs looking at separately.
THE RECORD SOFTWARE WROTE
- Job
- FH-1042, Completed (chosen by the engineer)
- Materials
- 1 x 3 bar PRV
- Follow-up
- Cold radiator in the spare room
- Office review
- No
- Engineer’s notes
- Kept word for word
THE LESSON
AI reads the notes. Code controls the record.
BOUNDARIES
- Typed notes only. Voice notes and photos are not part of this demonstration.
- It does not schedule work, message customers or invoice.
DEMONSTRATION 03
BUILT & TESTEDA customer emails the office.
From an incoming email to a reviewed draft.
AI-ASSISTED OFFICE INBOX
Who is responsible for what
AI
Understands the email and prepares a reply that answers what was actually asked.
CODE
Checks every draft (no prices, times or promises to attend), decides the route and keeps safety emails with a person.
HUMAN
Keeps authority over what is sent: reads, edits and sends. The workflow never sends anything itself.
WHAT THE DEMONSTRATION SHOWED
Understood, then prepared
AI read each email, worked out what it needed and prepared a reply, flagging any that needed extra care.
Routed as expected
Synthetic customer emails ran through the live workflow, and each took the route it should.
A person sends
Human approval is part of the design. The workflow prepares replies; it never sends one.
CHOOSE A SCENARIO
- SYSTEMAn email arrives
- AIAI drafts a reply
- CODECode checks the draft
- ACTIONLabelled Draft ready
- HUMANA person reviews and sends
DRAFT READY
A reply waits in the customer’s own conversation, answering every point raised. Nothing has been sent.
HUMAN CHECKPOINT
The draft waits in the conversation. Only a person presses send.
FOUND IN DEVELOPMENT
A draft that offered to send someone out
During development, a draft promised to “get someone out to take a look”. Nobody had agreed a visit. The code checks were widened so no draft can promise attendance, a repair or a reply time, whatever the wording.
THREE ROUTES, EACH ENDING WITH A PERSON
DRAFT READY
“Boiler service”
A reply is prepared for a person to check and send.
CHECK CAREFULLY
“Complaint - want my money back”
An acknowledgement is prepared, flagged for extra care.
REPLY MANUALLY
“smell near boiler”
No draft. Safety correspondence goes straight to a person.
Subjects from Fernhill's synthetic test emails. The workflow never sends an email itself.
THE LESSON
AI prepares the communication. A human controls what leaves the business.
BOUNDARIES
- AI cannot see bookings, job status, schedules or prices.
- Safety emails go straight to a person; this demonstration sends no separate alert.
WHAT THE THREE DEMONSTRATE
Three principles, engineered and tested.
- 01 / CALLSEngineer the whole process.A successful AI conversation matters only if the business outcome happens afterwards.
- 02 / POST-JOBTurn messy information into dependable records.AI interprets natural information. Deterministic software controls the record.
- 03 / INBOXAutomate the work without giving away the judgement.AI understands and prepares. A person keeps the authority that matters.
Fernhill is fictional, so nothing here is a result at a business. It shows how we engineer: start with the moment, and test the whole process, not only the AI.
START WITH A CONVERSATION
Your business will have its own version of these.
Start with a free Discovery Call. We will talk through where the work waits in your business and what could be engineered.