AI assistants offer one of the clearest routes to business value

AI assistants offer one of the clearest routes to business value

21 October 2026 Consultancy.eu
AI assistants offer one of the clearest routes to business value

With AI use cases emerging rapidly while budgets and resources remain limited, organisations need to focus on applications that deliver tangible value. Santhul Joseph, Senior AI Engineer at Crux Digits, explains why AI assistants should be a top priority for organisations investing in AI, offering some of the clearest returns on investment available today.

Surveys across European markets show that AI adoption is accelerating. Organisations have most commonly adopted AI for text- and language-based tasks, such as summarising documents, drafting content and searching reports.

However, many organisations have yet to take the next, relatively small step in this domain: putting these capabilities exactly where questions arise. That means on websites and in WhatsApp for customers, in Teams and on the intranet for employees, and in supplier portals for partners.

Rather than deploying a general-purpose chatbot that knows a little about everything, the idea is to create an assistant that knows an organisation’s own information inside out and answers questions based on that information.

Intelligent AI assistants have the capability to answer the questions that employees, customers and suppliers ask over and over, in the language they ask them in, at any hour of the day. It is one of the clearest returns AI offers today.

Internal knowledge, answered in seconds

An example from our own work at Crux Digits illustrates the case. An automotive sales company deals every day with questions about parts: which part fits which model, what replaces a part that has been discontinued, where the right specification can be found. The knowledge to answer those questions existed, but it was spread across catalogues, technical documentation and the memory of a handful of experienced colleagues.

The company’s colleagues spent a significant part of their working day answering the same kinds of questions for everyone else, which is exactly the kind of hidden cost that never appears on a budget line but is felt in every quote that waits and every customer who is called back later.

An AI agent now answers those questions in plain language, from the company’s own parts documentation. A salesperson types the question the way they would ask a colleague across the desk, and receives the answer together with a reference to where it came from, so it can be checked in a moment.

The experienced colleagues have not been replaced. They are still the ones who handle the difficult cases, the unusual configurations and the judgement calls. They are simply no longer the search engine for the easy ones, and that changes both their own working day and the speed at which the rest of the organisation can respond to customers.

The same principle reaches outward as well. An educational institute we work with uses an assistant to answer the questions that prospective and current students ask, and it answers each student in the language they write in, from the institute’s own approved information. For an organisation that serves an international audience, that removes a barrier that used to require either multilingual staff on every channel or answers that simply arrived later.

Why these use cases work

The two cases look different, one internal and one customer-facing, but technically they share four features, and those features are the reason they succeed.

The first is that the questions repeat. A large share of what people ask is a variation of something that has been asked many times before, which is what makes an assistant worthwhile, and also what makes it testable: the last few hundred real questions become the test set that shows whether the assistant answers correctly, today and after every change.

The second is that the answer already exists. The assistant is not inventing knowledge. It retrieves it from documents and systems the organisation already maintains, and it shows where each answer came from, so if the source is right, the answer is right, and if the source is wrong, the problem becomes visible.

AI assistants offer one of the clearest routes to business value

Intelligent AI assistants have the capability to answer the questions of employees, customers and suppliers

The third is that a wrong answer has a safe exit. Neither assistant makes decisions on behalf of the organisation. When it is unsure, or when a question needs judgement, it hands the conversation to a person, and that handover rule is a deliberate design decision rather than an afterthought.

The fourth is that the assistant lives where people already are. Nobody had to learn a new tool, open a new portal or remember a new address. Employees ask where they already work, and customers ask where they already write.

The variety of domains where AI assistants can add value

A closer look at departments, processes and workflows shows that these patterns appear in almost every department of a larger organisation.

Customer service is the most obvious place, particularly for organisations that sell across borders or serve customers who are more comfortable in another language. Questions about order status, delivery times, returns, invoices and opening hours make up a large share of incoming contact, and they can be answered on the website and in messaging channels in the customer’s own language, with a person taking over the moment a conversation becomes a complaint, a negotiation or an exception.

For an organisation with several brands or several countries, the assistant can answer each from its own content while the service team works from one shared view of the conversations.

Human resources is a quieter but equally strong case. In an organisation of a few thousand employees, the same questions about leave, expenses, travel policy, pension arrangements and working-from-home rules arrive every week, and they are too often answered from someone’s memory of an older version of the policy. An assistant that answers from the current policy documents, and refers sensitive personal questions to an HR adviser, gives employees a reliable answer at the moment they need it and gives HR back the time to work on the questions that genuinely need a person.

In technical and industrial organisations, field and service technicians are often the people who benefit most. Error codes, maintenance procedures, safety instructions and equipment manuals can be made available on a phone at the machine, in the technician’s own language, instead of in a binder in the van or a document library that only works from the office. The difference shows up in how often a problem is fixed on the first visit and in how quickly new technicians become productive.

Sales and product teams face a similar problem to the automotive example: specifications, compatibility rules, configuration options and pricing conditions that live in catalogues, spreadsheets and the heads of a few product specialists. An assistant that answers from those sources means a quote no longer waits for the one colleague who knows, and new sales staff can answer customer questions with confidence much sooner.

New employees ask the same questions in their first months: how systems work, who to ask about what, where documents are kept, which procedures apply. Answering those immediately and consistently makes onboarding less dependent on whose desk a new colleague happens to sit next to, which matters more as organisations grow and work across locations.

Supply chain and purchasing functions can use the same approach with suppliers and internal requesters, answering questions about order status, delivery schedules and invoice status from the ERP, so that many of the phone calls and emails that interrupt buyers and planners no longer need a person at all.

How to start well

The organisations that get the most from assistants tend to start in the same way, and it is rarely by buying the broadest possible solution. They pick one question stream, with one audience, one channel and one body of knowledge, because a focused assistant that answers well earns trust far faster than a broad one that answers vaguely, and that trust is what allows the scope to grow.

They collect the real questions first: the last few hundred messages from the mailbox, the chat or the service desk show what people actually ask, which is usually different from what the project team expects, and they become the test set that decides when the assistant is ready to go live and whether it stays ready when anything changes.

They also fix the source before the assistant. An assistant is exactly as good as the content it draws on, and outdated pages, contradictory policy versions and incomplete product data surface within days of going live. Cleaning that up is often one of the most valuable outcomes of the whole project, with or without AI. They decide the handover rule in advance: which topics always go to a person, how the conversation is passed on with its context, and how quickly someone picks it up.

They are open that it is AI, which is not only good practice but a legal requirement, since under the EU AI Act people must be told when they are talking to an AI system and that obligation already applies. And they measure what matters from the first day: the share of questions answered without a person, the share handed over, the time to an answer and how satisfied people are. Those four numbers tell an organisation, without argument, whether to widen the scope.

Build, or configure?

In terms of implementation, there is no longer any need to build from scratch. Assistants that answer from an organisation’s own content, in any language, on the website and in messaging channels, with a clean handover to a person, can be configured and tested within days rather than months.

Building becomes worthwhile when the AI assistant has to read from internal systems, such as an ERP, a parts database, a product information system or a case management system, or where the answers carry real business rules that differ per customer, contract or country.

A sensible route for most mid-sized and large organisations is to start with a configured assistant on one well-defined question stream, learn from the questions people actually ask, and then build only the integrations that the evidence shows are worth building.

Closing

The most useful AI in many organisations today is not the most ambitious. It is the AI assistant that tells a salesperson within seconds which part fits, answers an employee’s policy question correctly at the moment it matters, or replies to a customer in their own language late in the evening. It saves time on both sides of the conversation, it is visible to the people it serves, and it can be measured from the first week.

For mid-sized and large organisations, the question is no longer whether this works. It is which of the questions you answer every day should be answered this way first.