FR Book 30 min

Home / Client cases / HVAC

HVAC company · new owner

A new owner, an in-house ERP, and quotes stuck in the queue.

The owner had just taken over an HVAC maintenance company. Its in-house ERP was rich but full of gaps. The sprint filled it where information was lost, without replacing anything.

≈ 287quotes waiting for missing information at the start, according to the company
2 weeksof sprint, one day on site included
1 minfor a job closed on the phone to reach the ERP
The flow
  1. Job openedThe technician sees the site's equipment as it is in the database.
  2. Nameplate photoThe AI suggests brand, reference, serial number, refrigerant.
  3. ValidationThe technician confirms or corrects each value.
  4. Follow-up quoteBuilt on site, signed by the customer.
  5. ERP updatedJob and equipment written into the ERP within one minute.
One job, from the plant room to the ERP.

The context

The company runs on an in-house ERP, maintained by an external provider. The owner, recently arrived, sees that information coming back from the field is incomplete: equipment reference, serial number, photos, parts needed. The office calls technicians back, and quotes wait.

Exploring their test environment showed the cause: the job sheet captured the close-out well, but the equipment attributes were requested as free text, from a technician standing in a plant room. Photos went to a shared drive, with no link to the job sheets.

The approach

The existing job sheet was good: it was not rebuilt. The sprint fills the exact gap found in their form, block by block, hooked onto their database.

What was built

  • A field app on the phone: the technician sees the site's equipment as it is in the database, photographs the nameplate and the AI suggests brand, reference, serial number and refrigerant, from the company's own lists. The technician confirms or corrects; no serial number is written without them. Dictated comment, photos attached to the unit, follow-up quote, signature.
  • An office assistant: type a site, and the tool finds its equipment, repair history and open quotes, and suggests a likely fault with a confidence level, without inventing history.
  • A sync with the ERP: closed jobs are written into it within one minute, with nameplates added to the equipment base. An incomplete sheet is rejected, never filled with a default value.
Field app · nameplate readingFictional example

Nameplate photo

BRAND A
MODEL CH-130
No. EX-0001
R32 · 2.4 kg

Suggested by the AI · to validate

BrandBRAND A
ReferenceCH-130
Serial numberEX-0001
Refrigerant · chargeR32 · 2.4 kg
Validate the 4 fields
Fictional data. The photo stays attached to the unit, next to the values read.

What the AI does not do

A test on 48 nameplate photos taken by technicians before the project set the rules: these photos, taken for repairs, are hard to read, especially outdoors. Hence mandatory human validation, and the photo always kept next to the value read.

How

Remote kick-off, scope reset after exploring their tool, database access with their provider, then one day on site: workshops, real photos, a demo of the field app. Office assistant live after one week, field app in hand the following week, then team feedback, ERP sync and transfer to the company's own accounts.

Further reading