Offshore the helpdesk, or hire an AI workforce? A 2026 cost reality for UK FM.

It usually starts with one slide.

The COO drops it into the operations review: a quote from a Cape Town BPO, fully loaded, all-in. Per agent, per month. The number is roughly 40% of what the same role costs in London. The board nods. Someone says "let's pilot ten heads."

That is how UK FM ended up with helpdesks and accounts teams sitting in Sandton, Durban, and Kuala Lumpur. It is also how a lot of those programmes quietly missed their savings number by year two, and how some of them came home.

This article is not "offshore is bad." Offshoring works for plenty of operations. It is about the line items most FM operators do not put in the original business case, and what the picture looks like if you put an AI workforce next to it.

The headline numbers (year one)

Take a mid-size UK FM running roughly 5,000 reactive work orders per month plus a planned-maintenance book. A typical back office:

6 to 8 helpdesk agents. 2 to 3 accounts payable / sales invoicing agents. 1 to 2 chasing and coordination roles. Call it 10 FTEs.

Fully loaded, year one:

This is the slide. It is also about as far as most procurement processes go.

The line items the slide does not have

Attrition tax

Published BPO industry attrition for South African and Malaysian customer-service roles in 2024-25 ran at 30 to 50% annually. The corresponding figure for UK in-house admin teams sits closer to 15 to 25%.

Every leaver costs you 6 to 10 weeks of ramp on the replacement, plus the vendor's recruitment fee (typically baked into the rate at 10 to 15% of annual salary). For a 10-FTE team at 40% attrition, that is four full re-trainings per year. If your contracts have SLAs on first-response time, the ramp window is exactly where you breach.

Training drag

The vendor will quote you against "trained agents." What you receive on day one is six weeks of a learning curve. Job sheets misread. Wrong CAFM templates pulled. Suppliers chased twice. Reports rejected by your end customer.

A typical pattern: agents reach 60 to 70% of UK productivity by month two, 85 to 90% by month four. That tail of "almost there" is real money. On a £180k/yr South Africa contract it sums to roughly £35k to £50k of effective lost output across the first year.

Time-zone friction

South Africa runs +1 to +2 hours from the UK. That is fine. Malaysia runs +7 to +8 hours, which is not. A Tuesday-afternoon supplier email in London arrives in KL at end-of-day or after hours. A reply lands in your inbox Wednesday morning UK time. You have just turned a 90-minute conversation into a 24-hour one.

Multiply by a busy week and the WIP queue grows visibly. Customers feel it before you do.

FX risk

ZAR/GBP moved 22% in 2022 to 2023. MYR/GBP moved 14% in 2024. If your contract is GBP-denominated, the vendor reprices on renewal. If it is ZAR or MYR-denominated, you carry the risk and your CFO refuses to sign year three at quote.

This is the line item that gets removed from forecasts and reappears in actuals.

Public holiday divergence

South Africa has 12 statutory public holidays. Malaysia has 16, with state-level additions on top. The UK has 8. Without an explicit holiday-coverage clause in the SOW, you are running a thinner desk on Heritage Day, Wesak, and Deepavali than your customers expect. They notice it as silence.

Quality variance and rework

A typical pattern in unaudited offshore admin work: 5 to 12% of processed documents need rework. Bills coded to wrong cost centres, job sheets attached to wrong WOs, sales invoices issued at incorrect markup. Your own team becomes the QA layer, and that QA layer is pure overhead. It does not appear in the original quote.

South Africa vs Malaysia, at a glance

South Africa wins on real-time overlap and accent neutrality. Malaysia wins on raw cost. Neither one beats the next column.

The third option

An agentic AI workforce is not "outsourcing the helpdesk." It is moving a defined set of repeatable workflows (job-sheet chasing, document validation, bills processing, supplier email reading, CAFM updates) onto a system that runs them at machine speed, 24/7, with no ramp, no holidays, and no FX line.

What that looks like, year one, for the same 10-FTE workload:

Roughly 70 to 90% of the chase, process, extract, and file work absorbed by AI modules. Your remaining team (3 to 5 FTEs) shifts from doing the work to overseeing it: handling the small percentage of cases the AI flags, building client relationships, and working on the contracts that earn the margin.

No vendor management of 10 offshore seats. No FX swing. No 6-week ramp on every leaver, because there are no leavers.

                       OFFSHORE BPO              AI WORKFORCE
                      ─────────────             ──────────────

   monthly cost       £15k - £18k                volume-priced
   attrition          30 - 50% / yr              0%
   ramp / new agent   6 - 12 weeks               0
   public holidays    12 - 16 days off           0 days off
   FX exposure        ZAR / MYR                  GBP only
   QA layer           5 - 12% rework             flagged for review
   night / weekend    on call rota               native 24/7
   knowledge loss     every leaver               none
   scaling 2x         hire 10 more, 6 wk ramp    raise concurrency

Three-year TCO, indicative

Same 10-FTE-equivalent workload, 5,000 reactive WOs/month. Numbers are directional and based on industry-standard 2024-26 benchmarks rather than any one vendor.

The slide-deck saving of 50% versus UK in-house gets cut roughly in half by the time year three closes.

An AI workforce sits below both offshore options on three-year TCO for this workload, and the gap widens as work order volume scales (a 50% volume increase doubles the offshore programme; the AI side moves on a different curve). The exact figure depends on your module mix and volume profile, which is the conversation we have on the demo.

The honest take

There are operations where offshoring still works:

Customer-facing voice work where you genuinely need a human at the other end and you have the management muscle to run a 50+ seat programme. Markets where compliance and data residency rule out automated processing. Programmes large enough to absorb 30 to 40% attrition without breaching SLAs.

There are operations where it is actively worse than what an AI workforce delivers:

High-volume, document-heavy admin (job sheets, certs, bills, sales invoices). FM operations under £20m revenue where you cannot economically run a 10-seat managed offshore programme. Anyone who wants their team thinking about clients, not auditing the offshore team's rework.

The real question

It is not "offshore vs AI."

It is "do you want your back-office cognition to live in a managed contract on the other side of the world, with a 30 to 50% annual rebuild rate, or in a system you control, that runs the same way every day, that you can extend module by module?"

The cheap answer is the offshore quote. The right answer, for most UK FM operations under £80m, is the system that does not turn over.