The best, worst and strangest ways AI is really being used at work

Turning AI’s vast potential to transform professional life into a day-to-day reality is a challenge occupying workers and their bosses everywhere.

Click to read about an industry

| | | | |

Much attention is focused on the extent to which AI will result in job cuts. Amid a sea of predictions, from mass lay-offs to supercharged human productivity, understanding exactly where new tech tools can assist rather than replace workers in sectors from consultancy to law is not straightforward.

Here, FT writers speak to industry specialists to uncover the concrete benefits and drawbacks of AI in their working lives today — and some of the weirdest uses.

Consulting

Of course the PowerPoints are quicker to make. All that text box shuffling and graphic tweaking — the junior consultant’s lot in life — has indeed been made easier by generative AI. But the biggest efficiencies for a consultant might not be on the output side so much as on the inputs.

All the large consulting firms have been focused on how to speed up the process for employees who need to quickly get educated on a new market, new client or new problem when they are assigned a project. That has involved making vast troves of institutional knowledge — historical slide decks, lessons from previous projects, proprietary market data — searchable through in-house tools such as Lilli at McKinsey, Sage at Bain & Company and PwC’s ChatPwC.

Executives say this can shave days off basic research, leaving consultants more time to brainstorm with each other or clients and to explore a wider range of solutions to a problem. The latest frontier: making all this available in formats workers can digest on the go. McKinsey consultants can now get the information they need in a bespoke podcast, which they can listen to, interrupt and even ask questions.

Drawbacks

Consulting firms initially exhorted staff to learn and use new AI tools and factored this into annual performance assessments.

Getting a grip on what bosses want has been complicated in more recent months by a newfound concern about the cost of the “tokens” by which some AI models charge. This has led some smaller firms to impose limits or remove AI access from people who seem to use it less.

The pressure to demonstrate AI use has caused some unusual behaviours. “I know my employer monitors my usage of AI and . . . wants me to use it more frequently,” says one consultant on an industry forum. “If I haven’t had a reason to ask it a question all day, I will ask Claude something like, ‘Why are we here?’ That way, I get credit . . . for wasting tokens.”

Weird use

Clients might be surprised to learn that AI does not just dictate what consultants do but which consultants they might get. McKinsey seeks AI help to make sure a new project team has the right set of technical skills, expertise and personal “chemistry”, according to one person in the industry. If an “80 per cent perfect” team is available now, but a 93 per cent perfect team can be assembled next week, it might even suggest telling the client to wait.

Oil and gas

The world’s largest oil companies are also some of the largest bureaucracies: Shell and BP both have nearly 100,000 employees, many of whom work in the back office. Much of their drudgery can now be outsourced to AI.

“We do close to 5,000 non-disclosure agreements every year,” says Wael Sawan, Shell’s chief executive, outlining one of many examples where workers are turning to computers. “Every one of them would have required a lawyer in the past. Now you can do the mundane ones, 90 per cent of them, through AI and that allows our lawyers to focus on the more material ones that require a bit more thoughtfulness.”

Dashboards are proliferating throughout the industry, whether for bosses such as Sawan, who now has all the data he needs to make investment decisions on his screen, or for engineers manning pipelines or reservoirs, who can see flow rates and pressure readings and get predictive suggestions on where maintenance is needed.

Exploration experts are using AI to decode seismic readings of oil and gas fields, and to calculate the best way of drilling an asset. BP says it used to take weeks or months to draw the optimal trajectory through the rock to the reservoir, but engineers now cycle through hundreds of options in a day.

Drawbacks

The ability of engineers to do more, and faster, may not help oil and gas companies hit their net zero targets. While many oil majors have set a target to make their own operations carbon neutral by 2050, one recent peer-reviewed study found their productivity gain was likely to significantly boost emissions.

Weird use

Can AI predict when drivers are going to drink more coffee? BP is using it to manage stock levels in its convenience stores depending on weather conditions, sales history and regional tastes.

Banking

When preparing for client meetings, many investment bankers recognise that collaboration, while encouraged, can be treacherous. How well each performs affects who gets the credit when allocating the millions of dollars in fees from a big deal.

Rather than dividing up research between them, bankers are turning to AI tools — often internal models such as JPMorgan’s LLM Suite or a third-party service like Rogo — to quickly swot up on everything they have been talking about with clients and what they might want to hear about now, including updates on potential takeover targets.

“You walk into a meeting with 30 minutes of preparation. We use AI with internal information to quickly summarise all the internal written communication about a client,” says one banker at a large Wall Street firm.

AI is also helping ease the burden on junior bankers of time-consuming tasks such as the first drafts of financial models and client presentations. That’s no small thing when 80-hour workweeks are not uncommon.

Drawbacks

There are worries among more seasoned bankers that an over-reliance on AI can creep into the work, especially in an industry where getting an intuitive feel for the numbers and client pitches can become second nature through the laborious work of pulling these together.

Bankers say it is becoming increasingly clear when something has been written by AI and not updated. Sometimes it is a style point, but it can also seem so crisp as to be inauthentic.

“You’re having to put typos in, dumbing it down to look like it’s not an AI email even though it’s an AI email,” says one banker. “Because people don’t want to get an AI email.”

Weird use

AI chatbots have given entry-level analysts a rung beneath them in Wall Street’s hierarchy and some are getting a kick out of it. One banker talks about analysts scolding their AI bot if they make a mistake or do not deliver something properly.

Law

Trawling through reams of dull documentation has long been part of working life for many lawyers, but AI is changing that.

“If you’re a junior lawyer, you’re no longer sitting up at night reading 40 clauses from a contract to work out what to do,” says Sam Newhouse, global vice-chair of Latham & Watkins’ M&A and private equity practice. “You can get the answer to that very quickly and [spend time on] proper critical thinking.”

“If you need to analyse 400 agreements and find out the way something’s treated, you can now do that, though not perfectly.”

So, are lawyers getting more sleep? “I think that is probably a debatable topic,” he says. “You’re able to go further and deeper with analysis than before so . . . there’s additional work that now you can do. I think our teams are still incredibly busy. If [your client] is making a significant investment, why wouldn’t you turn over that further stone, why wouldn’t you diligence further?”

Drawbacks

Several law firms have been embarrassed when courts have censured them over hallucinations — where an LLM generates misleading information — in legal documents.

This year, elite firm Sullivan & Cromwell apologised to a New York federal judge for misquoting the US bankruptcy code and citing cases incorrectly, errors caused by its use of the technology.

Boies Schiller Flexner, the firm on the other side of the case, spotted the mistakes.

BSF itself had made similar errors the previous year, when in a case against Amazon, a document prepared using AI tools contained what a partner at the firm admitted were “material citation errors”.

Another problematic consequence for lawyers is that some clients have become so cautious about the risks of their sensitive data being fed into large language models that they have used tech tools to prevent their lawyers from downloading and even printing the information. In some cases that has forced law firms to assign more associates to a case and work in a less efficient way than in the pre-AI era.

Weird use

It is now easier than ever to write a plausible-sounding legal document and file it in a court case. In the US, a publicity-seeking company submitted a filing known as an “amicus brief”, drafted using AI, in a closely watched Trump lawsuit. This may have got it noticed but the trend risks clogging up court dockets and slowing down processes.

Pharmaceuticals

The biggest promise of AI for the pharmaceutical industry is shortening the drug discovery process. On average it takes about four to five years for molecules to reach pre-clinical testing — AI could cut that to between 12-18 months, according to Thorsten Rall, global industry lead for life sciences at Capgemini.

It is helping researchers to process the copious amounts of data generated in drug development at a scale not possible before. In practice, this means using AI to assess how a treatment might work in different patients.

Danielle Belgrave, vice-president of AI and machine learning at British drugmaker GSK, says identifying biomarkers “is the huge thing that AI is unlocking at the moment”. “Can I identify that feature or . . . trait within human biology that predicts whether someone will respond or not?”

Drawbacks

Pharma is one of the most regulated industries, and dreams of much quicker trials and drug development are some way in the future as regulators work out the rules.

The World Health Organization has warned that AI in drug discovery risks replicating biases that already exist in the human-led process. Patient safety could also be at risk if algorithms turn out false-positive or false-negative recommendations that are not properly checked and tested.

Scientists in the US recently used AI to create previously unknown viruses that raised alarms about biosecurity.

Weird use

Researchers in Germany found that hallucinations could actually improve the process of drug discovery. The researchers demonstrated in a study that incorporating LLM-generated hallucinations enhanced the performance of the models.

Advertising

Advertising creatives are automating many of the more mundane tasks in production, media buying and planning, giving them more chance to work up ideas.

Smaller companies can now use AI-powered systems that bring together all the tools they need to make and distribute ads without the expense of hiring a large agency. This enables start-ups and independents to better compete with larger agencies using AI to scale content and automate tasks that once required big central functions.

AI tools also allow advertising to be adapted countless times to different audiences and countries at scale.

Drawbacks

Some advertising executives say the benefits are not fully being recognised, with teams only travelling at the speed of their slowest members, who are often not yet fully embracing the new technology.

AI is also upending the traditional ways to earn money. Most agencies still operate on payment for the hours worked, leading to a mismatch in client demands for AI-based cost reductions despite rising token inflation. Executives worry that, as well as the real threat to more junior jobs, a reliance on AI for pitches will reduce the level of genuine human creativity and push advertising into more boring, cookie-cutter moulds of what has worked in the past.

Weird use

Market researchers are using AI to create “synthetic” focus groups and audiences to test ideas and tweak them to different demographics. Thousands of AI agents — or synthetic respondents — are given “personalities” and motives to replicate human behaviour (potentially even mapped on social media data) and simulate populations. They are asked questions from their favourite chocolate bar to who they would vote for in the next election. Their synthetic answers lead to decisions that affect millions in real life.