Your vendor pricing model shapes your customer experience

Companies spend a lot of time trying to improve customer service. They introduce new KPIs, run training programs, change scripts, hire consultants and build increasingly sophisticated dashboards. All of these things can help, but they tend to focus attention on how people behave rather than on the system in which that behavior takes place.

I was reminded of this recently while interviewing Eduardo Villares, Head of CX & CS at Cruzeiro do Sul. Eduardo told me about his time running a large customer service operation at Oi, where he was responsible for around 3,000 customer service positions serving millions of customers.

The operation had several problems at the time. Customer satisfaction was low, regulatory complaints were high and, according to Eduardo, more than 20% of customers were calling again within 24 hours. Looking at those numbers, the natural response might have been to focus on agent performance or service quality. Eduardo started looking at how the operation itself was set up.

The vendor model at Oi

Oi outsourced much of its customer service to external partners. The commercial model was based on PA, or Attendant Position, which essentially meant paying vendors for having agents available to handle customer interactions.

Consider what happened when a customer called with a problem. If the agent solved it during the first conversation, the vendor got paid for handling that interaction. If the problem remained unresolved and the customer called again the following day, the vendor handled another interaction and got paid again.

The second call cost Oi money and meant another frustrating interaction for a customer whose problem had not been solved. Under the existing contract, however, it was simply more volume for the outsourcing partner to handle. The vendor did not need to provide poor service deliberately for this to become a problem. It was operating according to a commercial model that did not place much economic value on avoiding the second call.

Eduardo changed the model. Oi expanded the operation from two outsourcing partners to six, creating more competition between them, and moved away from paying a fixed amount per agent position toward paying per call handled with quality.

One detail of the new arrangement is particularly interesting. According to Eduardo, if a customer called back within 24 hours, Oi would not pay the vendor for that second call. The vendor still had to handle it, which meant that a repeat contact now created additional work without generating additional revenue.

Oi also began moving traffic between vendors based on performance, incorporating post-call customer evaluations into the assessment of service quality. Eduardo says that over the following 18 months these changes helped reduce costs by R$20 million, while customer satisfaction increased from 5.5 to 8.5 and regulatory complaints fell significantly. These figures come from Eduardo’s account of the transformation rather than an independent analysis of Oi’s results.

Repeat calls and productivity

The story made me think differently about one of the most basic numbers in a contact center: call volume.

Suppose an operation reports that it handled 100,000 calls during a month. It sounds like a straightforward measure of workload and can easily become a proxy for productivity. Yet the number says nothing about how many distinct customer problems were behind those calls.

There might have been 100,000 separate problems. There might also have been 70,000, with the remaining 30,000 calls generated because previous interactions failed to resolve something. The operation looks equally busy in both cases, although the underlying performance is very different.

Manufacturing offers a useful analogy here. If a factory produces a defective component and then spends additional time fixing it, the repair increases the amount of work required to produce the same useful output. A repeat customer contact can work in a similar way. It consumes agent capacity and adds to reported volume without necessarily representing another customer need.

This matters even more when the operation is outsourced and the commercial model is linked to seats, time or interaction volume. A company may be trying to reduce unnecessary contacts while simultaneously paying a supplier according to a model in which more contacts generate more revenue.

Causes behind repeat contacts

Contact centers have been measuring repeat contact for years. Finding the causes behind it across thousands or millions of conversations has been much harder.

A repeat contact can come from very different situations. An agent may have provided incorrect information. An internal process may have prevented the issue from being resolved. Someone may have promised an action that another department never completed. Sometimes the agent simply lacks access to the system or authority required to finish the job.

On a dashboard, all of these cases can appear as the same thing: another customer contacted the company within a given period. Operationally, they require very different responses.

Traditionally, understanding those differences meant sampling calls, listening to recordings and manually categorizing what happened. That can give managers a good understanding of individual cases, but it becomes difficult to do consistently across a large operation.

This is where conversational intelligence becomes useful. Once conversations can be analyzed at scale, the repeat-contact metric can be connected with what actually happened during the interactions that preceded it. An operations team can see the most common reasons customers return, identify the processes involved and go back to the conversations where those problems occurred.

The result might show, for example, that a large share of repeat contacts has little to do with agent performance. Perhaps customers keep calling because agents promise a refund that takes another department seven days to process, or because the CRM does not expose information needed to answer a particular question. The same repeat-contact metric can point to coaching, process design, product changes or system access depending on what is happening inside the conversations.

Support and sales

Similar incentive effects appear outside outsourced contact centers, although the mechanisms differ.

A support organization that puts heavy emphasis on tickets closed per hour creates pressure to move through the queue quickly. For simple requests this may be exactly what the company wants. With more complex cases, however, spending another few minutes investigating an issue could prevent the customer from opening another ticket tomorrow.

Sales compensation creates a different problem. Imagine a salesperson receiving most of their financial upside when a contract is signed. Three months later, if the customer turns out to be a poor fit, the consequences appear in Customer Success metrics rather than the salesperson’s compensation. Over time, even a reasonable sales process can start producing tension between what makes a deal attractive at signing and what makes that customer successful after signing.

The useful thing about looking at incentives this way is that it moves the analysis beyond whether a particular KPI is sensible in isolation. Tickets closed, contracts signed and calls handled are all legitimate things to measure. The more interesting question is what people learn to optimize once those numbers become connected to targets, compensation and vendor contracts.

The 24-hour rule in practice

What I like about Eduardo’s example is how specific the intervention was. Oi did not need a new definition of customer-centricity. It identified one undesirable outcome, a customer calling again within 24 hours, and changed what that outcome meant economically for the outsourcing partner.

Today it is possible to take the analysis further. A company can separate genuinely new customer needs from contacts generated by previous failures, then use conversation data to understand where those failures originated. That creates a different set of operational questions: how much capacity is being consumed by customers coming back, which processes generate the most repeat contacts, and how much of that work could disappear if the original interaction or the process behind it were fixed?

In Oi’s case, more than one in five customers were calling again within 24 hours, according to Eduardo. Before deciding how to handle that volume more efficiently, there was a more basic question to investigate: how much of it needed to exist in the first place?