You Cannot Price an Outcome You Cannot Attribute

The BPO industry is debating whether to sell outcomes. It has not worked out how to prove who caused them.

In the fourth quarter of its 2025 financial year, Concentrix wrote off USD 1,523.3 million of goodwill.

The trigger was not a lost client, a failed integration or a collapse in demand. The company's own release attributes the charge primarily to the trading range of its share price and market capitalisation. In the same twelve months, revenue grew 2.2 percent to USD 9,825.8 million, adjusted EBITDA margin held at 15.0 percent, and cash flow from operations reached a record USD 807 million.

So the operating business worked. The market simply decided it was worth less than the accounts said it was, and the accounts followed.

This is worth sitting with, because it is the most precise measurement we have of something the industry has been discussing in the abstract for two years. When people say the market has lost confidence in BPO, this is what that sentence looks like once it has been audited and filed. Not a sentiment. A number, with a signature underneath it.

The orchestration argument is right

The prevailing response to this repricing, argued well by Mark Hillary and Peter Ryan among others, is that the industry has to stop selling seats and start selling outcomes. The provider of the future is not a labour supplier but an orchestrator of a hybrid estate, part human and part machine, sold on resolutions and retention and complaints prevented rather than on hours and headcount.

I think that is correct.

I also think the argument is incomplete in a way that carries real financial consequence, and the gap is not where most of the commentary is looking.

The favourite illustration in this debate is the operation that goes from 5,000 seats to 1,000 as automation absorbs the routine work. The usual reading is that the residual 1,000 becomes less attractive to run in-house, so outsourcing wins. Perhaps. But notice what that residual actually contains. Complaints. Vulnerable customers. Regulated advice. Escalations that have already failed once. It is the hardest fifth of the original queue, at a fifth of the volume across which to amortise overhead, carrying most of the risk.

And the industry proposes to price it on outcomes.

Three exposures, none of them priced

An outcome is a joint product. Whether a complaint is resolved, a customer retained or an escalation avoided depends on the provider's execution, and also on the client's product, systems, knowledge content, policy settings and routing architecture. Sell an outcome and you have accepted liability for a system you control perhaps half of.

That creates three exposures simultaneously. The industry has not priced any of them.

Attribution. When the outcome is not achieved, the contract has to allocate causation between two parties who each control part of the machine. Existing service level agreements cannot do this. Average handle time, occupancy, adherence and seat count are all measures of provider input, and they are attributable by construction because the provider controls them entirely. Outcome measures are not attributable by construction. Bolting outcome language onto an input-measurement framework produces contracts where the measurement exists but the causation does not. In a regulated environment, disputed causation does not resolve through commercial negotiation. It resolves through remediation programmes and lawyers.

Variance. Pricing an outcome means accepting its distribution, not its average. Global contact centre attrition runs at 30 to 45 percent annually. Australia is better, at 25 percent on the most recent industry survey, improved from 29 percent the year before. But a quarter of the frontline turning over every year means the workforce delivering an outcome in the fourth quarter is materially not the workforce that delivered it in the first. Tenure mix, rather than capability, becomes the dominant driver of quarterly variation. A provider that cannot evidence a stable outcome distribution before signature is not pricing. It is guessing, and it will discover the distribution after the ink dries.

Input cost. Pay-per-resolution pricing sits at roughly AUD 1.50 to 3.75 per autonomous resolution. That is a fixed revenue per unit against a cost of goods sold controlled by a small oligopoly of hyperscalers and model providers with considerable pricing leverage. Sign a multi-year fixed price per resolution and you are short an input cost you do not control, on volumes you cannot cap.

Individually, each of these is manageable. Together, entered without instrumentation, they are not a new commercial model. They are an uncapped liability wearing one.

The part nobody is saying out loud

Here is the structural problem underneath all three, and it is the reason this is an architecture question before it is a commercial one.

BPO replacement cost per agent sits well below the in-house equivalent. Our own benchmark for a fully loaded in-house replacement, derived in The Teams That Stay, is AUD 27,500. The outsourced figure is materially lower, and the reasons are structural rather than incidental. Loaded salary is lower, with Australian survey data placing outsourcer pay roughly AUD 10,000 below the in-house average, and every downstream cost that scales off salary falls with it. Recruitment is industrialised, so cost per hire amortises across campaigns rather than being incurred per vacancy. Time to competency is shorter because scope is narrow and work is scripted. Offshore delivery compounds all three.

That gap is not an efficiency. It is a design outcome.

The BPO operating model was deliberately engineered for workforce replaceability,

and for thirty years that was entirely rational. When the work is commodity Tier-1 voice billed by capacity, short ramp and thin knowledge dependency are exactly what you want. It is also why the industry has tolerated 30 to 45 percent attrition for two decades without treating it as an emergency. At that cost per head, churn was cheaper than retention. The maths worked.

I expect that sentence to draw fire, so let me put the objection in its strongest form. No operator has ever called 40 percent attrition acceptable. The industry has spent thirty years and a great deal of money on engagement surveys, career pathways, wellbeing programmes and retention analytics. All of that is true.

Revealed preference is what matters. A cost that genuinely threatened the model would have been engineered out, everywhere, decades ago. Instead the rate has persisted across every geography, every operator and every economic cycle, which tells you it was survivable. Retention has been funded at levels consistent with managing a known cost of doing business, not at levels consistent with removing a threat to the business.

The reason it was survivable is the billing model. Under capacity pricing, a departed agent replaced is still a billable seat. The provider absorbs the ramp inefficiency, which is real and which the good operators work hard to minimise. But the quality variance produced by a continuously reconstituting workforce is absorbed by the client, because output quality was never the thing being sold. Attrition was a margin drag. It was not a revenue risk.

Outcome pricing reverses the polarity. The variance that was previously externalised to the client lands directly on the provider's own revenue line. The attrition rate does not need to change at all for its consequences to change entirely. The same number moves from a cost line to the top line, and the retention investment that was uneconomic under capacity billing becomes unavoidable under outcome billing.

Automation has just removed the work that the maths was built on.

What remains is the complex, regulated, escalation-weighted residual described above, which carries the in-house complexity profile. The work has migrated. The workforce architecture has not. And the commercial model is simultaneously shifting from capacity to outcomes.

These three movements are not independent. Outcome pricing is the mechanism that converts the mismatch between role design and residual work complexity into a profit and loss event. You cannot price an outcome you cannot attribute. You cannot hold an outcome you cannot stabilise. And you cannot stabilise an outcome on a workforce built to be replaced.

What this means if you are buying

If you are a client considering an outcome-based agreement, or an investor assessing a provider's conversion pipeline, the disclosed contract value is the least informative number available to you. Four questions carry the actual risk.

1. Is there an attribution instrument in the contract? Not a definition of resolution. A taxonomy that separates provider-caused failure from client-caused failure, with an evidentiary standard and a dispute path, agreed before signature rather than litigated after it.

2. Is outcome variance measured before it is priced? Ask for the distribution, by tenure band, over at least four quarters. If the provider can only give you an average, they do not know their own variance.

3. Is input cost exposure capped or indexed? A fixed price per resolution against uncapped inference cost is a naked short.

4. Does the workforce carrying the account have the tenure stability to hold the outcome? Attrition on outcome-priced accounts is no longer an HR metric. It is a revenue quality control.

   
   

On current public disclosure, no major operator publishes enough to answer the fourth question. That is worth raising directly with management, because a provider converting contracts to outcome pricing without an answer to it is not executing an orchestration pivot. It is accepting outcome liability on an unstable delivery base and reporting it as commercial progress.

The destination is right

None of this is an argument against orchestration. Orchestration is where the industry has to go, and the providers who get there will be worth considerably more than the ones who do not.

It is an argument about sequence. Attribution before pricing. Variance before commitment. Architecture before commercial model. AI performs to the level of the architecture it is deployed into, and so does every contract written on top of it.

The Concentrix write-down was not the market punishing an operator for poor performance. It was the market saying it no longer believes that acquired scale in commodity delivery is worth what it cost. That belief will not be restored by a better story about orchestration.

It will be restored by the first provider who can stand in front of a client, or an analyst, and prove which outcomes it caused.

Built on Rigor. Engineered for Scale.

Figures cited are drawn from company financial disclosures and named industry benchmark surveys. Concentrix figures are from the company's fourth quarter and fiscal year 2025 results release for the year ended 30 November 2025.