Build a Little, Buy a Little: Nokia Implements AI Throughout Travel and Expense Program

Artificial intelligence may be a hot discussion point in travel management but there are as yet few documented instances of buyers walking the excitable talk. Even the 34 per cent who claimed they will apply AI in significant ways this year in a Global Business Travel Association survey published in February feels more an aspiration rather than reality.

One pioneer who truly has embraced AI, however, ranging from machine learning to generative AI, is Mário Pires, head of digital travel experience for B2B networking technology company Nokia. His team serves 78,000 employees across 130 countries. Pires is not only an early adopter of three AI tools offered by his company’s booking and expense tool provider SAP Concur, and one from CWT, one of Nokia’s two globally retained travel management companies (the other is American Express Global Business Travel). He has also repurposed an internally developed AI platform to respond automatically to the 1,000 queries a month received by the Nokia Travel and Expenses department.

Portugal-based Pires said the AI tools his team has deployed are already producing significant benefits. These include handsome time savings for travelers, travel approvers and the travel management team itself. What’s more, Pires added, the work performed by AI is often of a higher quality than the manual process it is replacing.

Those achievements align perfectly with the AI strategy of Nokia Business Services, the global shared service organisation to which the T&E department, part of the Source to Pay procurement unit, belongs. The strategy, said Pires, is “mainly to simplify the work of our employees to increase efficiency. We want to eliminate repetitive and tactical tasks and be more focused on strategic tasks. We want to do it not fast but responsibly: respecting privacy, security and ethics. We have very strict governance around AI.”

Pires is proudest of the contribution AI is making to easing the two main pain points his team and a group of Nokia travelers identified during a customer journey mapping workshop held shortly after he joined the company two years ago. The first pain point was raising a pre-trip travel request, which is a Nokia travel policy requirement. “People considered it to be duplicate work,” said Pires. “It’s true. You have to research your flight and hotel when you plan the trip and then you have to go to Concur again to search when you book.”

Even more unpopularly, “people were very unhappy about the expense submission process because it takes too long and has errors that need to be corrected and resubmitted,” said Pires. “Through AI we were able to tackle these two main points.”

Pires sees no need to build an AI solution internally if an external supplier has already developed one. But the AI tool he has created could only be developed internally. The automated response to employee travel queries uses Nokia GPT, an in-house generative AI tool that sources its information from internal documentation which cannot be fed into public domains. Nokia GPT has also been used to develop non-travel applications for in-house use. Here then are the five AI tools deployed by the Nokia T&E team:

Tool 1 – Traveler query response: Nokia GPT

Many of the 1,000 queries per month that Nokia employees raise about T&E relate to procedural matters such as obtaining a new credit card or submitting expenses. Employees raise a ticket in the workflow platform ServiceNow and the enquiry is responded to by an outsourced team.

Pires is experimenting with his outsourced team feeding the query into Nokia GPT. The tool interrogates resources such as the travel pages and policy documents stored in the company’s SharePoint platform used for internal collaboration, communication and document management. The outsourced team member copies and pastes the answer back into the ServiceNow ticket, editing it if inaccurate or incomplete.

“Seventy percent of the time it gives the correct answer and appropriate level of detail,” said Pires. “The chatbot does an amazing job because it sets out in steps and with the relevant hyperlinks what the employee needs to do. I was blown away by the answers it is giving. There is a much higher level of detail and excellence than we had before.” 

In addition to improved quality, said Pires, using Nokia GPT is reducing both average response time and outsourcing costs. Next steps will be to improve accuracy levels and automate the communication of queries and answers between ServiceNow and Nokia GPT but “in this first phase there will always be monitoring by a human until we are sure the answers will be correct.”

Tool 2 – Trip cost estimates: Concur Request Assistant

Nokia is one of the first companies to deploy Concur Request Assistant. Travelers input their destination(s) and dates of travel, and the tool automatically estimates the transport and accommodation costs based on data from previous trips by Nokia employees. Request Assistant will also advise if the trip can be made by train instead of plane, “so it’s already directing our travelers towards more sustainable choices,” says Pires. “This is one of the things we like a lot in this tool.

“At the beginning it was not accurate at all. Almost all the estimates were lower than the actual costs and that’s bad because then you don’t have the budget for the trip. We put it on hold for two or three months for Concur to improve the algorithm and now we are much happier. If our people don’t trust the output, they can stop using the assistant, but the feedback we are getting from employees is good. I would say I am 70 percent satisfied. We understand that it’s a new tool.”

Tool 3 – Expense form filling: Concur Auto-Itemization

Research by Pires found the most time-consuming aspect for Nokia employees of filling expense reports is itemization, for example entering each of the constituent costs on a hotel invoice such as room, breakfast, laundry and wi-fi. This cumbersome manual process is necessary because different elements of the bill may be subject to different tax treatments.

Nokia has started using Concur’s Auto-Itemization tool which performs the task automatically when employees upload a PDF of the hotel invoice into their expense report. Alternatively, travelers can forward the PDF to a Concur e-mail address, which will recognize the sender and enter the itemized details into the traveler’s expense report and attach the invoice to the report.

Although the auto-itemization often is inaccurate, Pires is delighted with progress.  

“Inaccuracies are things like the date or the wrong entry such as wi-fi instead of breakfast, but it only takes 1 or 2 minutes to correct these small errors, when before the whole thing could take 15 to 20 minutes,” he said. “I am sure also that the machine is learning from these errors and will improve with experience.”

Tool 4 – Expense auditing: Concur Intelligent Audit

Nokia also uses its outsourced T&E team to audit 40 percent of expense reports for compliance with 30 checkpoints such as spend limits, use of preferred suppliers, attachment of accurate receipts and confirmation of pre-trip approval.

Pires has introduced Concur Intelligent Audit, which is able to make most of these checks robotically (checks which still need to be made manually include flight class, whether a client was in attendance for a business meal and whether expenses have been split across different expense types to avoid exceeding spend thresholds).

Using the new tool, Nokia can now audit 100 percent of invoices for the same cost. The work is also completed much faster: before the traveler’s line manager approves the expense report. “Previously auditing was after approval, so if the audit discovered any problems, the claim would be sent back to the traveler to do again and re-sent to the line manager for approval again. It was a very cumbersome process,” said Pires.

Tool 5 – Business intelligence: CWT Analytics

Nokia consolidates all its travel reporting through CWT. When running a report on the CWT Analytics reporting tool, such as Nokia’s top 20 airlines, previously “I would have to bring the fields together and decide how to present it. It would take at least 15 minutes,” said Pires. 

Now CWT has enhanced Analytics with generative AI capabilities provided by a company called ThoughtSpot. “I type in what I want using natural language,” said Pires, who cautions that he has only tried the tool in English.

“I am truly happy with the results. Eighty percent of the time it returns the report you want within 20 seconds. Even if the report is not quite how you want it, all the data elements are there and it takes one minute to fine-tune what you want. It’s an amazing solution and an amazing way to speed up processes and a great example of how companies can leverage AI without having to develop it themselves.”

Pires is not stopping there with his AI strategy. “We are very excited about what we can do in the future,” he says. “We are hoping to see transformative changes, things like providing travel plans even before you think you need to do a search. We are just scratching the surface.”