guide · getting started
Getting started with AI in field service: the first 30 days
Follow a few of your jobs from the first email to the paid invoice, find where the admin hours go, and start with the simplest area that costs real time. Connect the systems it leans on, run it in shadow mode on real jobs, and only switch it on once its work matches your team's. Most Centrus customers are integrated with their job system and running their first agents within the first month.
Updated · 11 min read · Centrus AI
On this page
- The first 30 days at a glance
- What AI actually does in a field service business
- Before you start: the readiness checklist
- Week 1: follow your jobs through
- Week 2: connect what it leans on
- Week 3: run it in shadow mode
- Week 4: switch it on, then widen
- What teams already live got back
- Five mistakes to avoid in the first month
- What it costs
- Frequently asked questions
The first 30 days at a glance
Each week has one job and one thing to show for it. By the end you have one area of admin running for real, your own before and after in hours, and a list of what to take on next.
| Week | What you do | What you have at the end |
|---|---|---|
| 1 | Follow your jobs through and pick one area | A map of where the admin goes, and a number for the first area |
| 2 | Connect the systems that area leans on | The data in, and the right people able to ask it |
| 3 | Run the first agent in shadow mode | What it would have done on real jobs, checked against what your team did |
| 4 | Switch it on with a small team, then widen | Your own before and after, and the next area |
What AI actually does in a field service business
For an electrical contractor, a facilities management company or a fire and security business, AI does three useful things. It answers questions from your own records, it does the admin that follows a rule, and it tells the right person when something needs them.
- Answers. An engineer asks on WhatsApp for the access notes, the last visit or a fault code, and gets the answer from the job system and your own manuals without ringing the office.
- Admin. A work order lands in the inbox and becomes a job. A supplier invoice becomes a draft bill. The monthly client pack is filled in from the job system.
- Alerts. A job nears its quoted cost, a subcontractor's insurance is about to lapse or a certificate is missing a reading, and the person who owns it hears first.
None of this needs a new job system. BigChange, Joblogic or Simpro stays the record. The AI reads it, writes back when you allow it, and does the follow-up across the inbox, the files and finance.
Before you start: the readiness checklist
You do not need a data project or an IT team. You need the business ready: someone who owns it, a team who know why, and a few decisions made before the first call. If you can tick most of these, you are ready to start.
ai readiness checklist
Is the business ready for its first agent?
Eight things to have in place before the first call. Most are decisions rather than work, and you can settle them in a week.
- A sponsor who owns itOne senior person who owns the problem, clears the team's time and decides when the first area goes live.
- The team told why, up frontSay plainly that the admin is going, not the people, and who to tell when an answer looks wrong.
- A small core team, with timeTwo to four people who do the job every day, given an hour a week to check the shadow results.
- One area, picked and countedThe simplest area that passes the four tests, counted: how often each week, and how long each one takes.
- The rules written downThe rate card, the PO limit, which client's P2 means same day. If a person follows a rule, an agent can too.
- The systems, and admin accessThe job system, finance, folders and inbox that area uses, and someone with admin access to approve each one.
- Who sees what, decidedProcedures and manuals for everyone, pricing and HR for the office. Decide it before anyone asks.
- What good looks like by week 4The week 1 count to beat, and an honest bar: most runs right first time, with a person catching the rest.
Week 1: follow your jobs through
Before you choose anything to automate, follow the work through the business. Take three or four job types you do every week, such as a reactive callout, a planned maintenance visit, a quoted job and the month-end report for a client, and trace each one from the first email to the paid invoice.
The easiest way is to imagine explaining each one to a new starter on their first day. Not the trade: they cannot change a boiler or test a board yet. The admin around it. Where the work arrives, what gets typed where, who is told, which spreadsheet is updated, what gets downloaded, renamed and saved. For every step, note who does it, which system it leans on and roughly how long it takes.
Every step you would have to explain to the new starter is a step someone could write down. A step that can be written down is a step an agent can follow. By the end of the week you have a map of the admin, area by area, with the systems each one leans on.
Questions that find the hours
- Where does the most time go, and where is the most value? They are often different answers, and both are worth writing down.
- Is finance downloading PDF invoices, renaming them and saving them to SharePoint? It is a few minutes each time, and it adds up. A supplier invoice can just as well become a draft bill in Xero, matched to the PO on the job.
- Is the same spreadsheet built by hand every month? A client report, or the sales team's commission sheet, can take a day a month to put together, and it is easy to get wrong. It can be filled in from the job system.
- What do the engineers ring the office for? Access codes, the last visit, a fault code. Every call is someone in the office stopping what they were doing. Engineers can ask on WhatsApp instead.
- What came up at the last board meeting? A problem the directors are already discussing has an owner and a reason to fix it. Ask which part of it is admin that follows a rule.
Then pick one area
Choose one area from the map and nail it before you touch the rest. Score each candidate against four tests: it happens every day or every week, it follows a rule someone could write down, it lives in systems you already run, and a mistake would be caught before a customer saw it. Start with the simplest one that passes, not the biggest. A quick, visible result buys the time and goodwill for the harder ones.
Then put a number on it. For one week, tally how often it happens and roughly how long each one takes. PFL Electrical's figure works the same way: about 100 messages a day from engineers, at 10 minutes saved each, is 83 hours a week back (read the PFL Electrical case study).
Week 2: connect what it leans on
Your map says which systems the first area leans on. Connect those: the job system and the finance system once, Train pointed at the folders where the documents live, and the inbox the work arrives in. Job, finance and inbox data is read live when someone asks, not copied across, so the answer is the one in BigChange, Joblogic or Simpro at that moment.
Then decide who sees what. Tags on each document keep pricing and HR with the office and procedures open to everyone, so an engineer in the van never sees director rates. Folders re-sync every 1, 2, 4, 8, 24 or 48 hours, whichever you choose, so a new rate card is what Chat answers from after the next sync.
Give access to the small team who own the first area before anyone else. The questions they are asked most often are the best test of what it knows, and they will spot a wrong answer in seconds. See how Train and Chat work.
Week 3: run it in shadow mode
Build the first agent around the area you picked, then run it alongside the person who does the job today, without letting it act. That is shadow mode. The agent reads the real emails and records and does the work, but nothing is created, sent or changed. Your team carries on as normal, and at the end of each day you compare what it would have done with what they did.
There are two ways to do it. If your tool has no test mode, build a read-only version first: give it read access only, and have it write each job it would have raised to a spreadsheet, one row per email, for someone to review. Agent Builder has the test mode built in. A dry run reads your connected systems and runs the AI for real, simulates anything that would send, create or update, and shows exactly what each step would have written. A dry run costs 1 credit, however many steps the agent has (test an agent before it goes live).
Tweak it until the shadow results match. When a draft is wrong, correct it once: say "from now on" in chat, or give the answer a thumbs down, and Centrus drafts a lesson for one person, a tag or the whole company, approved before it applies. Expect most fixes to be rules nobody had written down, such as which client's P2 means same day.
“Team members are more productive. Manual tasks completely automated. The ROI is incredible.”
Week 4: switch it on, then widen
Switch it on for real with the same small team, and keep an approval step in front of anything sensitive for the first weeks. A named person approves each job before it is saved, and every run is logged, so you can see what it read and what it wrote. You decide when routine work no longer needs the check.
Then measure it. Count the same job over the same length of time and compare it with your week 1 number. That is your result, in your own hours, and it is the one to take to the directors. Iron out what the first live week turns up before you add anything.
Only then widen, one step at a time. Take the next link in the same chain, such as the worksheet check after the job is raised, or the next area on your map. Start with the simplest each time, and keep the core team small so feedback comes back in hours, not weeks. See what teams automate first.
Set expectations early. An agent will not be right ten times out of ten in its first week, and that is what the shadow run and the approval step are for. Ask for help when you get stuck. Agent Builder suggests the steps from a plain-English description, and Centrus customers are set up with the UK team, but the fastest fixes usually come from a ten-minute conversation.
What teams already live got back
Three UK field service companies run Centrus on the job systems they already had. All three started with answers for the team, from the job system and their own files. BSML then added agents for the back-office work, one area at a time.
83
hours back every week
Electrical · BigChange
Read the case study
40
hours back every week
Electrical, fire and facilities · Joblogic
Read the case study
100+
monthly hours saved
Facilities Management · BigChange
Read the case study
Five mistakes to avoid in the first month
- Rolling it out too wide, too fast. When something half works in front of fifty people, they lose heart and stop using it. Start with a small core team and widen once it works.
- Starting with the hardest job. The biggest problem is rarely the best first one. Start with the simplest area that costs real time, get a result, then take on the harder ones.
- Skipping the shadow run. An agent that acts on day one has not earned trust yet. Run it in shadow, check its work, then switch it on.
- Automating a job with no rules. If two people in the office would do it differently, agree the rule first, then build the agent.
- Skipping the count. Without a week 1 number, the week 4 number is a feeling.
What it costs
Centrus is from £299.99 a month on Small Team, for up to 5 users, and £499.99 a month on Large Team, for up to 20. Both include Train, Chat and Agent Builder with UK-based onboarding and support. Plans run month to month, and annual billing saves 20%. See pricing.
Frequently asked questions
Do we need to replace our job system to use AI?
No. BigChange, Joblogic or Simpro stays the record. Centrus connects to it, reads jobs, sites, quotes and invoices live when someone asks, and writes back only when you allow it, such as raising a job from an emailed work order or a purchase order from the van.
How long does it take to get started with AI in field service?
The team can ask questions as soon as the job system and folders are connected. Most Centrus customers are integrated with their job system and running their first agents within the first month, which is why this guide plans the start in four weeks: follow the work, connect, run in shadow, switch on.
How do we test an AI agent before it goes live?
Run it in shadow mode on real jobs. In Agent Builder a dry run reads your systems and runs the AI for real, but anything that would send, create or update is only simulated, and each step shows what it would have done. Compare that with what your team did, fix the rules, then switch it on.
What happens if the AI gets something wrong?
Answers from your trained documents show their sources, so anyone can check them. Agents can pause on an approval step before anything sensitive is saved or sent, and every run is logged. When something is wrong, a lesson corrects it once, for a person, a tag or the whole company.
Can we just give the team ChatGPT?
ChatGPT can read files and email once someone connects them, but it has no official connection to BigChange, Joblogic or Simpro, it is not on WhatsApp in the UK, and someone in the business has to build and look after every agent. Centrus comes connected and is set up with the UK team.
Is our data used to train AI models?
No. Your data is not used to train AI models, and training is switched off on our AI provider account. Access is controlled per person through tags and roles, and your data is encrypted in transit and at rest.
