What Starbucks’ AI retreat tells HR about the frontline workforce

Starbucks has retired an AI inventory tool across North America, nine months after deploying it across its stores according to Reuters. The tool was designed to automate inventory counts and improve visibility of product shortages but had reportedly miscounted and mislabelled items, including similar milk types. 

The Starbucks example raises a wider question – as AI moves into stores, warehouses, hospitals, restaurants, factories and field operations do organisations understand frontline work well enough to make the technology useful?

One reason this matters is because the frontline is often treated as a single workforce category. In practice it contains very different jobs, skills, risks, working patterns and employee expectations. A tool that helps one group may slow another down while a process that works in a corporate pilot may behave differently in a busy café, ward, warehouse aisle or production line. 

So are organisations designing AI around frontline reality? BCG research, published today, suggests this is already a live issue for employers. Some 75% of UK frontline employees now use AI at least several times a week. Yet 49% fear they will lose their job to AI, well above the 36% global average. At the same time 70% of UK frontline AI users say AI has increased their day-to-day joy and satisfaction at work and 67% say AI has changed the skills expected of them.

Steve Rockey, head of people and culture Europe at hotel company The Zetter, says the figures “don’t feel wrong” from his experience in hospitality. He is already seeing AI appear in written communication between employees, managers and guests. But he says employers need to be clear about who benefits. “From a business perspective you end up looking at this thing as a way to control things better,” he says. “The classic thing missing from most of that is, does it make the employee’s life easier?”

New research from The Josh Bersin Company challenges the habit of talking about frontline workers as though they are one group. Its report, Five Types of Frontline Workers: A One-Size Talent Approach Doesn’t Fit, notes that the frontline workforce comprises 80% of the global workforce or around 2.7 billion workers worldwide. In the United States frontline roles account for nearly 72% of total employment and 60% of all wages, according to the report’s analysis of US labour market and Lightcast employment data. Frontline work is a major business issue – and also a major HR issue.

Why Starbucks is a useful warning

The Starbucks case is useful because inventory counting sounds like a contained, rules-based process. In practice it happens amid product movement, storage variation, time pressure, customer demand, staffing constraints and daily execution challenges.

According to Reuters the Starbucks app used LiDAR (light distance and ranging) sensors and camera data on tablets to scan shelves for milks, syrups and other beverage products. It was expected to be faster and more accurate than manual counts.

The reversal suggests accuracy, workflow fit, employee trust and operational governance matter as much as the model or device. This is where HR needs a stronger voice. AI implementation may be led by technology, operations or transformation teams but the consequences are carried by workers, managers, trainers and teams who absorb the extra work when systems perform poorly.

For Tom Kegode, founder of SparkShift, the lesson is that frontline employees need to be part of the redesign rather than brought in after the technology decision has been made. “The people on the front lines are closest to how work actually gets done – they're best placed to redesign it. Go tech first and you'll struggle for adoption. Go culture first, bring people into co-creating the future, and not only will you get adoption – you'll get advocacy.”

Rockey says the challenge is that frontline AI must still work around real human lives. “A rota can be written incredibly correctly by AI and relate to forecasting and other good stuff”, he says, but “it’s then the practical application of that with whoever it is that needs to leave work early that day because they need to pick their son up from school, for example.”

He notes that systems that may once have got a rota “80% correct” are now “well into the 90s”. But the final 5-10% is where the human reality sits: anxiety, childcare, buses, late finishes, missed shifts and all the things that cannot be neatly forecast. “It can’t be seen as the panacea. It still can’t be the predictor for the human behavioural element,” he says.

Why frontline worker is too broad a category

The Bersin report identifies five distinct types of frontline workers: customer-facing associates, back-office associates, high-skilled specialists, licensed specialists and credentialed professionals. Each group needs a different talent strategy, operating model and technology approach. 

This distinction is important for AI because frontline technology is increasingly being used across hiring, scheduling, learning, safety, performance support, workforce planning, task allocation, inventory, compliance and service delivery and these use cases do not land in the same way for every group.

Customer-facing associates may need faster onboarding, simple scheduling and real-time performance support. Back-office associates may need multilingual communication, safety alerts and more predictable access to pay. High-skilled specialists need visible career paths, scenario-based learning and tools that reduce daily friction. Licensed specialists and credentialed professionals need reliable credential tracking, workforce planning, safety governance and technology that protects scarce expertise. A single frontline AI strategy will miss these differences.

The report also finds that more than 60% of frontline occupations are high-skilled jobs that are difficult to backfill, replace or automate. Yet organisations invest less than one-third as much in skilling frontline workers as they do for white-collar employees: around $400 per frontline worker annually, compared with more than $1,500 for office-based roles.  Organisations are asking frontline workers to absorb more technology, more data and more change while often giving them less development, less voice and less design attention than corporate employees.

Rockey says this rings true in hospitality. Rolling out a new engagement or communication tool to reception is very different from engaging chefs working in basement kitchens or engineers who may see immediate practical value in being able to show a problem in real time through an app.

The frontline manager is the missing link

Frontline AI will often succeed or fail through the manager. Line managers translate new tools into daily practice, answer questions, correct misunderstandings, manage resistance and keep service or production moving when the technology does not behave as expected.

The Bersin report repeatedly points to manager capacity as a critical issue. It argues that organisations should reserve manager time for coaching and critical decisions, while also investing in structured coaching, succession models and practical support. A manager who is already stretched cannot easily become the adoption lead, troubleshooter, coach, data interpreter and trust-builder for every new system.

For Rockey the starting point for frontline retention and performance remains the manager. “It always comes down to the manager,” he says. “When you invest in managers and engage them, they then invest in and engage their teams.”

This is why pre-rollout questions should be used as a way of testing whether the organisation understands the work, the worker group and the manager capacity needed to make AI useful.

The underinvestment problem

The frontline workforce is large, diverse and commercially important but it has often received less strategic attention than the office workforce. Bersin describes this as an underinvestment paradox: organisations spend less on frontline skills and support, then absorb the costs of turnover, vacancies, lower service quality, safety incidents and lost productivity. 

This is where the AI conversation needs to become more grounded. There is a risk that organisations see AI as a way to compensate for underinvestment in frontline work. Used well AI can indeed improve scheduling, reduce repetitive administration, provide real-time guidance, support safer work and help managers make better decisions. But used poorly it can increase monitoring, add friction, undermine trust and create more workarounds.

The difference lies in how the work is designed. A frontline AI strategy should sit alongside a broader frontline workforce strategy. This means understanding which roles carry the highest operational risk, which skills are hardest to replace, where turnover is most damaging and where technology can genuinely improve the day-to-day experience of work.

It also means resisting the temptation to treat adoption as the main measure of success. 

What HR leaders should do now

HR leaders do not need to own every frontline technology decision. They do need to be close enough to influence the questions being asked before those decisions are made.

Map the frontline workforce properly. Use the Bersin five-part framework to identify different risks, skills and support needs across customer-facing associates, back-office associates, high-skilled specialists, licensed specialists and credentialed professionals.

Bring managers in earlier. They know where work gets stuck, where systems create friction and where teams rely on informal workarounds.

Involve frontline employees before rollout. Understand how the tool changes pace, judgement, autonomy, safety, customer interaction and workload.

Measure the human consequences. Alongside productivity, speed and cost track trust, workload, error recovery, manager time, attrition, learning needs and employee confidence.

Build feedback into the operating model. Frontline work is dynamic. Products move, customers behave unpredictably, staffing changes and physical environments vary.

The frontline AI opportunity is still real

None of this argues against AI in frontline work. Many frontline workers deal every day with poor systems, fragmented information, unpredictable scheduling, inadequate training and constant pressure from customers, patients, colleagues or operations. Well-designed AI could help.

It could make learning more immediate, help managers anticipate staffing gaps, support safer work, reduce low-value administration and make information easier to access during the working day.

The most valuable frontline AI will be shaped around the people doing the work. A barista, warehouse operative, field technician, nurse and pilot do not need the same tools, training or operating model. The frontline is where many organisations experience their biggest workforce risks and their clearest opportunities for business value. AI makes understanding that workforce more urgent.