Somewhere in your contract archive is a business process outsourcing agreement signed in 2020 or 2021, priced on the number of people the provider assigns to your account, with an annual escalator tied to wage inflation in the delivery location. It runs for another two or three years. That contract was written on an assumption that is now visibly weakening: that the cost of the service is the cost of the labour performing it.
Your provider will get faster whether or not your contract notices. The only question is who owns the difference
Providers are not waiting for permission. Every major outsourcer has an internal automation programme, and the work most exposed — document handling, classification, first-line query response, reconciliation preparation, translation — is precisely the work that fills a back office contract. When the provider's delivery cost falls, the money goes to whoever the contract says it goes to. Under a resource-based agreement, that is the provider, in full, until renewal. This is not sharp practice; it is the arrangement both parties signed. But most buyers have never asked the question, and the first time they ask it should not be in the final quarter before expiry.
Three pricing models, three very different outcomes
Resource-based. You pay for people. Efficiency gains accrue entirely to the provider, and the incentive structure actively discourages automation on your account, because reducing headcount reduces revenue. If your contract is structured this way, expect the provider's innovation to appear in its margin rather than in your invoice. Per-transaction. You pay per invoice, ticket or reconciliation. Better aligned, because the provider keeps the gain from working faster while your cost tracks your volume. The weakness is the unit price, which was set against a cost base that is now moving and will not adjust on its own. Outcome-based. You pay for a result — a closed period, a resolved case, a compliance outcome. Best aligned in theory and genuinely hard to specify, which is why most attempts revert to transaction pricing with an outcome label attached. Most large agreements also contain a gainshare clause. Very few buyers have ever triggered one, because triggering it requires a measured baseline that nobody maintained.
Six clauses to read again this quarter
Benchmarking. Usually present, usually toothless, because the comparator set and scope definitions let any provider argue the comparison is invalid. Tighten the definition of comparability, not the right itself. Continuous improvement and technology insertion. Typically an aspiration with no number attached. This is where a productivity commitment belongs. Productivity commitments. An agreed annual reduction in unit price, in the schedule, independent of how the provider achieves it. The single most practical lever available, and the one providers concede most readily because it preserves their freedom of method. Data rights. What the provider may do with your transactions, documents and process history. Subcontracting and subprocessors. Written for people and locations, rarely for model endpoints. Audit and transparency. Whether you can establish how the work was actually performed.
The data clause is the sleeper issue
Your provider processes your invoices, your tickets and your exceptions, and that history is the raw material for exactly the models that will make the service cheaper to deliver. A provider that builds a capability on your operational history and deploys it across its client base has transferred value out of your operation, entirely legally, under terms you agreed when the clause was about confidentiality rather than about training material. This does not require a hostile position. It requires three sentences: purpose limitation on your data, aggregation and anonymisation before any cross-client use, and a statement of who owns models derived substantially from your process history. Providers will negotiate this reasonably today. They will negotiate it much less reasonably once the capability exists.
Sequence the response properly
Do not attempt to reprice on speculation. Do this instead. Establish a baseline now: volumes by transaction type, unit cost, service levels achieved, and the human effort actually applied. Without it, every later conversation is an exchange of assertions. Insert measurement and transparency rights at the next amendment, which is a smaller ask than a price change and usually passes without drama. Change the pricing structure at renewal rather than mid-term, moving from resource-based to transaction or outcome pricing. And put the productivity curve into the price schedule as a defined annual step, so that the gain is shared automatically rather than argued about at every review.
The make-or-buy question changes shape too
If automation removes a meaningful share of the labour content, the wage arbitrage that justified offshoring shrinks with it. What remains as a reason to outsource is not cost per hour. It is management capacity, absorption of volume variability, and access to a workforce you do not have to recruit and retain yourself. Those are real reasons, and they are different reasons. A make-or-buy analysis still running on a cost-per-full-time-equivalent comparison is answering a question that is becoming less relevant each year.
| Pricing basis | Review question |
|---|---|
| Resource | What baseline and productivity terms govern the labour component? |
| Transaction | How does the unit price change when volume or method changes? |
| Outcome | How are result, quality, controllability and exceptions defined? |
| All models | Who may use the data and which model endpoints process it? |
Qualitative summary of this article's source text, not a measured outcome or performance estimate.
Practical Guidance for Contract Structure Review
- Baseline volumes, unit cost and effort before any negotiation.
- Identify which pricing model each agreement actually uses.
- Add a defined productivity step to the price schedule.
- Split labour and technology components in escalator clauses.
- Limit purpose and cross-client use of your process data.
- Extend subprocessor terms to cover model providers and inference locations.
- Require disclosure of the human and machine mix in delivered work.
- Reprice at renewal, not mid-term, unless a clause already permits it.
The Regional Angle
Three considerations apply specifically to organisations buying these services from this region. The first concerns a clause almost every regional agreement contains. Contracts with South Asian and South East Asian delivery centres are typically denominated in dollars, priced per full-time equivalent, and escalated annually by reference to wage inflation in the delivery location. That escalator encodes a causal claim: that your cost rises because labour costs rise. If the provider is simultaneously reducing the labour content of the service, the claim is no longer sound, and you are indexing the whole price to the one input that is shrinking. The fix is not to fight the escalator but to split the price into a labour component and a technology and platform component, and escalate only the first. Providers generally accept this because it protects them on the input they genuinely cannot control, and it removes the automatic annual increase on the part of the service they are actively making cheaper. The second is that the region's most exposed outsourced cost line is language. Bilingual customer service, Arabic document handling, translation of statutory filings and localisation work are among the largest outsourced spend categories for regional groups, and they command a premium precisely because bilingual staff are harder to recruit. They are also the categories where machine output improved most visibly over the past two years. The predictable behaviour, and it is already appearing, is that providers substitute machine translation and generation with a light human review pass while continuing to invoice the bilingual premium that was priced for human production. That is not necessarily a bad deal, but it should be a chosen one. Require disclosure of the human and machine mix by work type, price a reviewed-output tier separately from a human-produced tier, and specify which one applies to material with legal consequence. Where the Arabic text is the binding version of a filing, a light review pass is not equivalent to human production and should not be priced as though it were. The third is that nationalisation policy belongs in the make-or-buy model and is almost never in it. Emiratisation targets now carry defined annual increases and financial consequences for companies above the employee threshold, and the Saudi framework has operated on similar logic for longer. These obligations attach to your own employee base, not to a provider's offshore staff, which historically pushed transactional work offshore for reasons that had nothing to do with efficiency. Automation changes that arithmetic in a way worth modelling deliberately: a smaller in-house team of higher-skilled roles is both cheaper to run than it used to be and considerably easier to fill with national hires who will stay, because the roles are analytical rather than clerical. Groups rebuilding a back office around a smaller supervised core should treat quota economics and retention as part of the business case, not as a compliance afterthought handled by a different department.
The objection worth taking seriously
The strongest objection is that this invites you to pick a fight with a provider you operationally depend on, over savings that are currently hypothetical. Outsourcing relationships degrade in ways that are hard to see: the best staff are quietly moved to accounts that are not squeezing margin, discretionary effort disappears, and the service meets its measured levels while getting worse in every way that is not measured. A buyer who opens a repricing conversation on the basis of a technology trend, before any demonstrated cost reduction, has spent relationship capital on speculation. That is a fair warning, and heavy-handed mid-term repricing is a reliable way to buy a worse service at a lower price. The response is that nothing recommended here is a mid-term price cut. Baselining your own volumes is an internal exercise requiring no negotiation at all. Transparency on the human and machine mix is information you are entitled to as the party accountable for the output. A productivity step in the schedule is a renewal discussion, negotiated in exchange for term or volume. None of it is adversarial, and a provider that resists all three is signalling that the economics of your account are about to change in a direction it prefers you not to see — which is, in itself, the most useful information you will get this year.
Common Questions
Should we renegotiate now or wait for renewal?
Wait, unless you have a benchmarking or change clause you can actually exercise. Use the intervening period to build the baseline, which is what makes the renewal conversation possible.
What productivity commitment is reasonable?
It depends on the process mix, and a defined annual step in the price schedule is more valuable than an aggressive one you will argue about. Specify the mechanism first; negotiate the number second.
Will providers accept transparency on automation?
Many will, at renewal, because they would rather disclose the method than surrender the pricing model. Very few will volunteer it.
What should we expect over the next twelve months?
Expect every major provider to announce AI capability this year while its commercial teams continue signing resource-based contracts, because that is where the margin is. Expect the first genuinely AI-inclusive pricing schedules to appear in renewals from next year, initially in document-heavy processes. Expect benchmarking clauses to be tested more seriously than at any point in the past decade, and to disappoint the buyers who test them. And expect the sharpest disputes to concern data and derived models rather than price, because price is negotiable at renewal and a model trained on five years of your transactions is not something you can take back.
Contract Structure Review — we baseline what you actually buy, separate labour from technology in the price, and get the data and productivity clauses fixed while they are still negotiable.