Ask Spurs: What Tottenham’s AI agent means for your customer service

Imagine a fan who wants to know, shortly before a home match, whether his membership is still valid, where he can get the shirt of his choice and when the stadium gate opens. In the past he would have visited three pages or written an email. At the London Premier League club Tottenham Hotspur, an AI agent called “Ask Spurs” has been answering these questions for a short time now. Salesforce presented the case in a press release on 15 October 2026. For companies in the German-speaking Mittelstand, this seems far away at first glance: a football club with millions of fan accounts is not a machine builder with 80 employees. On second glance, the case contains a pattern that applies to every company with recurring customer questions: an agent that handles part of the requests itself and hands the rest over to a person.

This article classifies what the press release actually substantiates, what it explicitly does not say, and what you can derive from it for your own company without adopting anything that is not substantiated. We stick to what the source says. Figures that come from the vendor are marked as statements by the vendor.

What was published on 15 October 2026

The source is the press release “Tottenham Hotspur Becomes Agentic Enterprise With Agentforce” from Salesforce UK, published on 15 October 2026. It is therefore a statement by the vendor about one of its own customers. This matters for how you read it: the text aims to show what is possible with the product. It is not an independent assessment.

The core of the release can be summarised in a few points:

• What Ask Spurs is: A multilingual AI agent built on Salesforce Agentforce. It gives fans personal answers about tickets, memberships, merchandise and stadium events.
• Where it runs: On the website tottenhamhotspur.com and in the club app.
• Languages: The release mentions “eleven languages” but does not list them individually. Whether German is among them is not stated there.
• Handover to people: If Ask Spurs cannot resolve a question, it hands the conversation over to an employee on the club’s support team.
• Data basis: According to the release, Salesforce Data 360 merges more than 4.6 million fan accounts into “SpursID” profiles. Data 360 is connected to Snowflake via Zero Copy.
• System landscape: By its own account, the club replaced more than 30 individual systems for this and connected more than ten others.
• Key figures: The release cites an “86 % engagement rate” and “20 % of interactions resolved by Agentforce alone, all grounded in trusted Spurs fan data”.

The release also contains quotes. Rob Pickering, Chief Technology Officer of Tottenham Hotspur, is quoted as saying: “Over the past two years, we’ve been transforming the technology behind our fan experience.” Zahra Bahrololoumi CBE, CEO of Salesforce UK and Ireland, says: “Tottenham Hotspur shows what’s possible when trusted data, CRM, and AI come together to solve a real business challenge.” Both statements are promotional messages in the best sense: they are understandable, but they prove nothing about your company.

Availability: what is certain and what remains open

With any product announcement, the first question worth asking is whether you can use any of it at all. Here the answer is nuanced.

Ask Spurs itself is an implementation by the football club. The press release does not describe an agent that you can book as a product, but a case at a customer. So you cannot “introduce” Ask Spurs the way you install a program.

The platform behind it, Salesforce Agentforce, is a product of the vendor. The press release says nothing about prices, plans, prerequisites or the countries in which you can use it. There is also no statement on whether it offers functions in German. For your company, this means: availability in the DACH region and German as an agent language are not substantiated by this source. You clarify that directly with the vendor before you plan anything.

What we make of this: we do not describe the case as “usable for you now”, but as something that is coming toward companies. Customer service agents that resolve part of the requests themselves and pass the rest to people are no longer a vision of the future, but an offering you should expect in the coming months, whether from Salesforce or from other vendors. Those who are prepared decide calmly. Those who are not decide under pressure when a vendor raises the topic.

What the figures say and what they do not

The release’s two key figures deserve a closer look because they are easily misread.

“86 % engagement rate”: The release does not explain what “engagement” means here. Does every conversation that is started count? Every one in which the person writes more than one message? Without a definition, the figure cannot be compared with your own customer service. We therefore do not use it as a benchmark.

“20 % of interactions resolved by Agentforce alone”: This is the more interesting figure. It says that one fifth of the interactions was resolved by the agent alone, without a person stepping in. The sentence ends with the note that everything is based on “trusted fan data” from the club. What the figure does not say: what the remaining four fifths looked like. Many may have gone to employees. Many may have ended because the fan only wanted some information and was satisfied afterwards. Some may have abandoned the conversation. The release does not distinguish between these. Anyone who concludes from the figure that “80 percent go to people” is reading more into it than it says.

A resolution rate is only meaningful if you know what “resolved” means, what happens to the rest and who verifies it.

Both figures come from the sender. They are not confirmed by a third party in the source. We report them because they appear in the release, not because we have verified them.

Why the case is interesting for the Mittelstand anyway

A football club with more than 4.6 million fan accounts has different prerequisites than a company with 120 employees. Nevertheless, three patterns can be read out of the release that apply regardless of size.

First: the agent does not resolve everything. The release itself describes a handover to the support team when the agent cannot answer a question. This is the central design decision. An agent that always answers, even when it does not know, is more dangerous than one that admits it cannot go further. For your customer requests, this means: define in advance how the agent recognises that it has to hand over, and where it hands over to.

Second: the data is half the work. By its own account, the club replaced more than 30 systems and connected more than ten before Ask Spurs could answer in a meaningful way. The release describes this as part of the project. This, too, can be transferred: an agent can only answer what it finds in your systems. If customer data, contracts and price lists are scattered, the agent runs into the same gaps as your employees.

Third: the metric is part of it. The fact that Salesforce names a rate at all is not a given for a vendor. It shows that a customer service agent is measured by the number of cases it resolves itself, not by the number of conversations. For you, this means: anyone who introduces an agent should decide beforehand how to measure success, and check that figure regularly.

An everyday example

A wholesaler of building services supplies with 90 employees receives enquiries from installation companies every day: Is the item available? Where is my shipment? Which invoice is still open? Some of these questions can be answered from the inventory management system; others need a decision, for example when a customer demands a credit note or a delivery does not arrive. An agent could prepare the first group and hand the second cleanly to the internal sales team, including a summary of the conversation so far, so that the customer does not have to explain everything twice.

This is a thought experiment, not a statement about a customer and not a figure from the press release. But it shows what it is fundamentally about: not about replacing employees, but about making sure that routine questions do not eat up the time for the cases in which someone really has to think.

Mapping: which of our agents fits

On amazing-agents.com we describe eight agents. The Ask Spurs case touches two of them.

• The Research Specialist delivers, according to our description, market analyses and competitor checks. Its strength is the targeted finding and consolidation of information from sources. The pattern “compile answers from your own data” is its task, before an answer goes to people.
• The Inbox Manager sorts requests and writes draft replies. It is closer to what happens in customer service every day: prioritising incoming questions and preparing a reply that a person approves.

This is a comparison of tasks, not a product promise for Agentforce and not an indication that Ask Spurs works this way. Salesforce products are third-party products. Our agents work according to their own logic. But the question you have to answer for both is the same: where is the limit, and who gives the approval?

Agent with limits and an approval step: without and with

Without an approval step: The agent answers every request, even when it is unsure. A customer asks about a refund, and the agent names a deadline that is wrong. Nobody sees the answer before it reaches the customer. A routine question becomes a complaint. The Ask Spurs release does not describe this as a problem at the club, but the risk lies in the construction: an agent without a limit answers even when it would be better off staying silent.

With an approval step: The agent answers what it can answer from reliable data and passes everything else on to a person in a traceable way. For decisions with financial or legal consequences, it prepares the answer as a draft, and a person reviews it before it is sent. On our website we describe it like this: every agent runs within a defined framework with approval steps for critical actions; emails, for example, are created as drafts and not sent directly.

From the press release, it can be inferred that Tottenham has built in at least one handover to people. Whether and where there are approval steps beyond that, for example for certain types of answers, is not stated there. We therefore assume nothing.

How we proceed in the audit

We do not claim any results from client projects here, but describe how we proceed when a company wants to assess an agent for recurring customer questions:

1. Collect and sort requests. Together with you, we look at the actual requests of the last few weeks: which ones recur, which are simple, which need a decision?
2. Check data sources. Where are the answers? In which system, in what quality, who maintains them? An agent is only as good as the source it draws from.
3. Set limits. Together we define which questions the agent must not answer, and what it does in that case: hand over, ask a follow-up question, prepare a draft.
4. Design the handover. Who receives the request? In what form, with what summary, within what time?
5. Measure success. We determine which figure counts: not the number of conversations, but the number of cases completed without follow-up questions, and the number of errors.

At the end there is a process plan, not a promise. Whether you then assess a vendor such as Salesforce, choose another route or deliberately wait is your decision.

What you should prepare now

Because availability in the DACH region and language support are not substantiated by the source, this is initially not about use but about preparation. You can tackle these steps this week:

1. Write down the ten most frequent customer questions. Ask your customer service or internal sales team to collect the most frequent questions of the last four weeks. For each one, note whether the answer comes from a system or is decided by a person.
2. Record the answers centrally. Check where the answers are today: in the heads of individual employees, in scattered documents or in a system. Every agent, whether from Salesforce or from another vendor, needs an organised source.
3. Set a handover rule in writing. Write down in which cases an agent must never answer alone, for example complaints, contract questions or price negotiations, and who takes over in that case.
4. Ask the vendor. If you already use Salesforce, ask your contact whether Agentforce is available for your country and in German, and on what terms. Do not rely on marketing texts, but have the answer given to you in writing.
5. Involve data protection. Clarify with your data protection officer which customer data an agent may see and where it is processed. This is the same question for every vendor.

What the source does not say

The press release gives no information on costs, on the duration of the project or on fan satisfaction. It does not define “engagement” and does not say what happens to the interactions that the agent does not resolve alone. It does not list the eleven languages and does not comment on German. It says nothing about the availability of Agentforce in the DACH region. And it is a statement by the vendor, which by its nature presents the benefits positively. We have therefore limited every statement in this article to what is stated there, and named the gaps.

Conclusion

Ask Spurs shows where customer service agents are heading: they answer part of the requests themselves, connect the data from many systems to do so, and pass the rest to people. The figure Salesforce names is modest and instructive precisely for that reason: according to the release, the agent resolves one fifth of the interactions alone. That is neither a failure nor a miracle, but an indication that most of the work lies in designing the handover, not in the answer itself. Those who clarify this now for their own company can assess an offering calmly later. Those who do not will be caught off guard by the question when a vendor or a competitor raises it. The difference lies not in the tool, but in the arrangement: an agent with limits and an approval step instead of an agent that simply answers.

Sources

• Salesforce UK: “Tottenham Hotspur Becomes Agentic Enterprise With Agentforce”, press release, 15 October 2026, salesforce.com/uk/news (primary source, retrieved on 17 October 2026)
• Amazing Agents: description of the eight agents and the approval framework, amazing-agents.com