MPB Hits 48% Resolution by Revamping Its Help Center

Summary

• MPB, a used equipment platform, rebuilt its help center and brought in the AI agent Fin.
• It set response rules and organized its knowledge, and the resolution rate rose to 48%, double the early level.
• Success with generative AI depends on structured knowledge and strict rules more than on installing a tool.

When logistics back up, customer inquiries explode

At MPB, a platform that buys, sells, and takes trade-ins of used photo and video equipment, thousands of used items arrive and sell every week. According to a customer story published by the customer support platform Intercom, MPB ranks as the largest platform in this field in the world. Unlike ordinary e-commerce, every item listed here is one of a kind, with its own condition and history. The questions customers ask are bound to differ from item to item.

Even in normal times, the customer support team handles thousands of conversations a week. The problem is that at certain moments inquiries pour in out of control. What causes the spikes at MPB is not retail seasonality such as Christmas or year-end. It is logistics delays. When too much used equipment from people who want to sell flows into the logistics hubs too fast, a bottleneck forms.

When work piles up at the Circular Commerce Centres, the hubs that handle MPB’s product inspection and receiving and shipping, the operations team finds it hard to meet the processing times it promised on the website. As soon as a delay occurs, inquiries of “Where is my gear?” pour in. People want to know whether the item they sent arrived safely, when inspection will finish, when the sale payment will be deposited, and whether the order is confirmed. In busy months when logistics back up like this, customer conversations soared to as many as 40,000.

Gideon Knight, a customer operations manager, described the situation like this. “When our Circular Commerce Centres experience a backlog, we can experience a large uptick in volume. Understandably, customers just want to know what’s happening, but dealing with this often unforecasted surge in contact volume can put real pressure on the support team.”

Why it had to protect trust without simply adding people

MPB’s customers fall into two groups: people who want to buy used equipment and sellers who want to dispose of the gear they used. Buyers want the trust to buy expensive used optical equipment with peace of mind. Sellers want speed, with the item they sent inspected quickly and the payment settled promptly. By the nature of the platform, careful handling is essential for both groups of customers.

MPB was growing fast in both new and existing markets. As the business grows, conversation volume grows with it. But unlike an ordinary company, it could not keep hiring support staff without limit as conversations increased. MPB’s core challenge was to keep the trusted, people-centered quality of support it had maintained for the past 14 years without simply adding headcount.

Chris Beattie, global head of customer experience, recalled that a change in support strategy was unavoidable. “Customer experience is truly at the heart of everything we do at MPB. Over the last 14 years, we’ve focused on delivering a consistently great experience for our customers. Our CEO is really passionate about this, and it’s always been about putting the customer first.”

Beattie added: “As MPB continues to grow in new and existing markets, we realized that our support strategy needed to keep up. We knew we had to adapt the way we work. Our goal was to scale without simply adding more people to the team … all while keeping customer satisfaction a top priority.”

MPB was already using Intercom as its help desk and handling common inquiries with rule-based automation. But a method that pushed out fragmentary answers by fixed rules needed people’s hands constantly to maintain and manage. In a recommerce setting where thousands of unique items move every week, rule-based tools alone could not capture the subtle and complex flow of what customers go through.

Rebuilding the help center and putting Fin at the front

MPB decided to bring in Fin, Intercom’s AI agent. It judged that Fin could faithfully follow the company’s support policy, adapt flexibly to guidance content, and maximize the value of human agents’ help at the moments it is truly needed. MPB placed Fin at the first window of live chat and set it so that Fin greets every incoming conversation first.

But installing software alone does not solve the problem. MPB set a principle that the quality of an AI’s answers can never rise above the quality of the base knowledge it refers to. So the first thing it did was to overhaul the help center, the customer guidance window, into a structure that is clear, consistent, and easy for Fin to read and process.

As it changed the knowledge structure, it also created new roles in the organization. It formed a team of knowledge development specialists led by a dedicated content strategist to manage the help center’s knowledge base in an orderly way. At the same time, it divided responsibility clearly so that customer operations manager Gideon Knight took charge of Fin’s setup, training, and ongoing performance management. Gideon Knight said of this, “We didn’t just switch it on and hope for the best. We invested in the content, the structure, and the training Fin needed to actually be useful.”

In the real work, the knowledge development team probably controlled each help center document so that it covered just one topic. It likely put key information such as processing times and required prerequisites at the top of each article, to keep Fin from mixing several documents and giving the wrong schedule. For multilingual support, the team probably set it up so that Fin translates the English source documents in real time, and added only country-specific shipping or payment terms as separate documents.

Setting response rules to control uncertainty

To control how Fin handles conversations, MPB introduced a configurable rule system called Fin Guidance. Because of the nature of a platform whose product mix keeps changing, the AI could not lean on fixed answers alone. Through strict rules, MPB trained the AI to handle uncertain questions and to give exact policy guidance.

The first rule is how to respond to a seller’s question about equipment price. When a seller asks the price of equipment, Fin does not estimate the used market price and answer on its own. Instead it guides the customer to MPB’s quote tool, which produces an exact price based on real-time data. Market prices change frequently with the condition of the equipment, so the aim is to avoid losing the seller’s trust with an inaccurate estimate.

The second rule is checking the sale status for customers who have already shipped their gear. If the equipment the customer sent has already arrived at an MPB logistics hub, Fin confirms the specific guidance schedule: when inspection takes place, when the inspection result notice goes out, and when the sale payment is deposited.

In practice, because Fin is not connected in real time to the internal transaction system, it probably worked out the current situation through step-by-step conditional questions. The structure was likely to ask the customer whether they had shipped the item and whether they had received an arrival notice, and then find the standardized inspection and settlement timelines in the help center and tell them.

The third rule is the response to surges in inquiries from logistics backlogs. When inquiries asking about the whereabouts of gear surged because of recent operational delays, the operations team quickly gave Fin focused training to handle “Where is my gear?” inquiries. Fin told customers a clear expected processing time and the next step. As a result, repeat inquiries, where anxious customers keep asking the same question, went down, and the support team’s workload eased.

This process too probably ran in a flexible way: when a delay was detected on the logistics floor, the customer operations manager would turn on a temporary guidance rule that reflects the waiting period in Fin, and switch the rule off once logistics returned to normal.

The fourth rule covers brand tone and the criteria for handing off to a human agent. Every Fin answer was polished to be short, clear, and in keeping with the brand voice. If a customer seems frustrated during the conversation or the situation looks urgent, Fin was designed to express empathy first and then pass the conversation to a human agent.

Fin Guidance: response path by situation

A 48% resolution rate and a change in how work gets done

Thanks to this advance preparation and these clear response rules, Fin came to work as a trustworthy first line of support. Fin’s resolution rate, which sat at 25–30% in the early days of adoption, has held steadily at 40–50% and has roughly doubled to 48%. In peak months, Fin resolves as many as 10,000 inquiries on its own, without human involvement.

Change in MPB's Fin resolution rate

Quality of support also stayed steady. The CX Score, a customer experience score measured across both conversations Fin handled and conversations people handled, was 83%. This came while conversation volume ran to thousands a week in normal times and as many as 40,000 a month during logistics backlogs, and the service standard held.

Chris Beattie said of this result, “Fin’s deflecting a lot of volume. But more importantly, it’s helping us maintain service levels even when things get busy.”

The way it expands globally has changed too. MPB now supports customers in five languages: English, German, French, Dutch, and Italian, and it has put Spanish and Portuguese support on its future roadmap. It successfully expanded its service into several new European markets without hiring native-speaking agents for every country it enters. Foreign-language inquiries that reach human agents were probably also handled smoothly within the existing team using translation tools.

The working culture inside the team changed too. Adam Cox, a senior product manager, explained, “It’s taken a lot of pressure off the team. We can now be more proactive, and provide the kind of service our customers expect.” Agents who had been buried in simple repetitive inquiries could focus on more personalized, in-depth consultations, and opportunities opened up for them to take on new roles in the organization.

The challenges ahead and what companies can learn

MPB is now preparing two next steps. The first is combining AI Insights, its AI-based analysis, with the CX Score. The idea is not to stop at checking a fragmentary satisfaction score at one moment, but to use AI and large language models to understand the causes and context of customers’ difficulties fully and in real time.

Chris Beattie said, “We feel that AI and LLMs are now providing us with the insights we need to fully understand the voice of the customer and be able to turn all their valuable insights into tangible actions to keep pace with customers’ increasing expectations.” Gideon Knight also expected that this effort will fundamentally change how the company looks at customer satisfaction.

The second step is to connect Fin directly to internal systems so that it can handle even complex transactional inquiries. Today’s Fin does not reach real-time transaction data directly. Once the system integration is done, MPB plans to have Fin look up real-time order data itself to check detailed order status, and even carry out order cancellations and refunds directly.

The MPB case offers a clear lesson to companies that want to bring in AI customer support. The success of an AI support agent depends on the structure of the base knowledge a company holds and on its control rules, more than on the model’s performance. Before switching on the model, MPB rewrote its help center documents so that the AI could understand them easily, and it first put dedicated staff in place to manage the knowledge.

It also strictly separated the areas the AI should answer from the areas that must be handed to specialist tools. A typical example is the decision to stop the AI from making up used-equipment prices, which swing widely, and to connect customers to the quote tool instead. A practitioner thinking about adopting AI should look not only at a tool’s automation rate but first at whether their own guidance knowledge is organized so the AI can read it well, and whether rules are ready to control surprises on the front line.

---

Sources

1. Intercom Fin — How MPB is scaling the recommerce support experience with Fin
   https://fin.ai/customers/mpb (previously https://www.intercom.com/customers/mpb)
   Publication date: unknown (not displayed on page; 2025-01-01 in structured data is a generic placeholder for fin.ai customer stories). Identical text and figures present in 2025-08-30 web archive; accessed 2026-09-24