A company can introduce new software, train employees and report high adoption without making its operations more productive. Staff may complete individual tasks faster while customers still wait for the same approvals. A department may save time by passing additional work to another team. These outcomes can look encouraging in a project report while leaving the commercial result uncertain.
Digital investment needs a test that follows work through to completion. The question is whether the organisation produces a useful result with fewer resources, better quality or a service improvement that customers value. Logins and features can help explain adoption, but they do not establish that result.
For businesses assessing automation and AI, this distinction has financial consequences. A technology budget creates continuing commitments, including subscriptions, support and changes to working practices. Management needs evidence that those commitments improve the operation. This article examines how firms can develop that evidence without assuming every useful tool must immediately reduce employment or transform the whole company.
Start with the meaning of productivity
Productivity concerns the relationship between output and inputs. The US Bureau of Labor Statistics distinguishes labour productivity, which compares output growth with growth in hours worked, from total factor productivity, which considers a broader combination of inputs. These are statistical measures, but the distinction is useful when evaluating business projects.
A company that saves employee time by purchasing an expensive external service may improve output per employee hour without reducing its overall cost. That can still be worthwhile if the service raises quality or enables more sales. The business case needs to explain which benefit it expects and which additional resources it consumes.
Revenue per employee also needs careful interpretation. It can rise because prices increase, low-value activities are outsourced or the sales mix changes. Managers should avoid treating a favourable ratio as proof that a particular software investment caused the improvement. Operational measures need to sit alongside financial results rather than substitute for them.
Adoption is the beginning of the analysis
Digital tools can support market access and more efficient operations, but the result depends on how firms use them. The OECD's 2025 D4SME Survey examines digital maturity and practices across ten OECD countries, in collaboration with digital platform partners. Its scope should be kept in mind when applying its findings elsewhere.
The report considers the barriers involved in adapting business processes as well as adopting technology. A firm may have access to a capable platform while lacking time, skills or reliable information to use it well. Installing software does not settle how employees should handle difficult cases.
Usage data remain useful. Low adoption may explain why an investment produces little benefit. High adoption may show that staff find a tool convenient. Neither result answers whether the completed service is faster or less costly. A credible review connects use of the tool with a defined change in work.
Follow one customer outcome across departments
Consider a hypothetical company processing customer orders. Sales enters the request, finance checks payment terms and operations confirms availability. Each team may finish its own activity efficiently, yet the order can spend days waiting between teams because responsibility for missing information is unclear.
A new application that speeds up data entry addresses only part of that process. If the main delay occurs during an approval, the customer may notice little improvement. Mapping the sequence makes it possible to identify where a technology investment could change the result and where a policy or staffing decision is required instead.
The most useful starting point is often a specific completed outcome, such as an accepted order or a resolved service request. That gives departments a shared result to assess. It also exposes situations where one team meets its target by creating more corrections, enquiries or waiting time elsewhere.
Better information can matter more than additional features
Integration is difficult when departments use different definitions. One team may treat an account as active after a contract is signed, while another waits until the first payment. Software can move those records quickly while preserving the disagreement. The apparent technical problem may therefore originate in business rules.
The OECD's analysis of SME digital transformation describes wider adoption gaps for more sophisticated applications and process integration. It also identifies skills and organisational constraints. Those findings do not imply that every smaller firm needs a comprehensive enterprise platform. They support a closer assessment of the prerequisites for effective use.
Cleaning a small set of important records can be more practical than attempting to standardise every database simultaneously. Customer identity, product definitions and payment terms may be sufficient for an initial project. A controlled scope helps the business learn whether clearer information improves performance before committing to a wider programme.
Evidence from AI needs context
Academic research provides examples of productivity gains, but businesses should resist applying a headline finding to every occupation. The study Generative AI at Work, published in The Quarterly Journal of Economics in 2025, examines a conversational assistant used by customer support workers. The researchers found improvements in productivity, with effects varying across workers.
An author manuscript is also available on arXiv. This is evidence about a particular setting and task. It does not establish that any organisation purchasing an AI subscription will obtain the same outcome. The technology was integrated into an existing workflow, and the study examined operational performance rather than simply asking whether workers liked the tool.
For management, the important lesson is methodological. Measure the task in which assistance is used, consider who benefits and examine quality alongside speed. A tool that helps less experienced staff may have a different value proposition from one intended for specialists. Training, task allocation and supervision can influence whether a benefit becomes visible in company results.
Baselines prevent attractive but weak comparisons
Before changing a process, a business needs a reasonable record of its existing performance. Average handling time alone may be insufficient if work varies substantially. Straightforward requests and complicated cases should be distinguishable so that a change in the workload is not mistaken for a technology gain.
Seasonality matters too. A pilot during a quiet period may encounter fewer interruptions than a rollout during peak demand. Comparisons should account for relevant differences in volume, staff experience and customer mix. Perfect experiments are often impractical in operating businesses, but transparent limitations improve the usefulness of the result.
A baseline should include rework and waiting time where those affect the outcome. Employees may spend little time actively processing a request while the customer waits several days for completion. Recording both helps identify whether the project changes effort, service speed or both. Those benefits can justify different commercial decisions.
Quality must remain in the calculation
Faster completion is valuable only when the result meets the required standard. An automated response that generates a second customer enquiry may move work rather than remove it. A quickly prepared document that requires extensive checking may not save much time overall.
ISO's discussion of quality management emphasises consistent performance and continual improvement. Its principles offer a useful reference for viewing connected processes rather than isolated activities. This does not mean that a firm must seek certification to evaluate a software project; the management logic can inform ordinary operational reviews.
Quality measures should fit the task. Relevant indicators could include corrections, unresolved cases or complaints attributable to the process. Some errors are much more consequential than others, so a single average may conceal an important risk. Where the service involves sensitive decisions, speed should be assessed within the applicable standards for accuracy and oversight.
Time saved is not automatically money saved
An employee who spends less time preparing a weekly report does not necessarily cost the business less. The time becomes available for other work. It produces a financial benefit when the organisation uses that capacity effectively, avoids additional hiring or reduces a genuine recurring expense.
The distinction helps prevent inflated business cases. A project may release a few minutes across many employees without creating a practical reduction in staffing needs. Multiplying those minutes by salary costs can suggest a saving that never appears in cash expenditure. The released capacity may still improve responsiveness, but it should be described honestly.
Conversely, a project can create value without a direct cost reduction. Better scheduling may allow more customer appointments. More reliable information may help staff answer enquiries without repeated escalation. Management should explain the route from operational improvement to the commercial benefit rather than rely on an abstract total of hours saved.
Include costs that continue after launch
The initial purchase price is only one part of digital investment. Integration, migration and staff training can affect the first year. Monitoring, licence renewals and support affect later years. More complex workflows may also need specialist employees or external help that a small business cannot absorb easily.
Businesses should examine the cost of changing direction. Moving data out of a platform, rebuilding integrations or replacing a custom workflow may involve expense. A low entry price can be less attractive if the company becomes dependent on a configuration that is difficult to maintain or transfer.
An evaluation need not predict every future cost precisely. It should identify the assumptions that matter most and test whether the investment remains useful if those assumptions weaken. A project that requires exceptional adoption, uninterrupted service and permanent low pricing deserves closer scrutiny than one with a reasonable margin for operational variation.
Employees can reveal where the process fails
Frontline staff often know why records need correction or why customers contact the business repeatedly. Involving them in evaluation can uncover problems that an executive dashboard misses. Their contribution is particularly useful when formal procedures differ from the steps needed to complete a real task.
Participation should not become an unlimited request for unpaid adaptation. Employees need time to learn the system and report difficulties. If the organisation adds technology while maintaining every old procedure, the workload may initially increase. That transition cost belongs in the implementation plan.
Training also needs to support judgment. Staff should understand when a tool is appropriate, how to recognise a questionable result and how to continue when it fails. Familiarity with menus is useful, but it does not establish that employees can handle exceptions. A productive workflow makes those responsibilities clear.
Expansion should follow a demonstrated benefit
Once a project improves a defined outcome, management can assess whether the result will transfer to other teams. Different customer populations, data quality or approval requirements can change performance. Repeating the deployment without examining those conditions risks treating a local success as a universal template.
Expansion can also increase dependence on the platform. A service disruption that affects one team during a pilot may affect the whole business after rollout. Continuity arrangements and support capacity should therefore develop alongside usage. The benefit and the exposure can both grow with scale.
Some projects will justify only limited adoption. Others may show that a simpler process change captures most of the value. Those are legitimate findings. An evaluation is useful when it improves the allocation of resources, including by stopping an investment whose ongoing demands exceed the evidence of benefit.
Questions businesses are asking
What should a productivity pilot measure
It should measure a defined completed outcome, the resources needed to achieve it and the quality of the result. Adoption data help explain performance. They should accompany measures of correction, waiting and ongoing costs so that management can assess the full process.
Can a digital project succeed without reducing headcount
Yes. It may improve service, support additional volume or help employees perform more complex work. The business case should identify the intended benefit and demonstrate how released capacity is used. A staffing saving should be claimed only when it is supported by an actual change.
Why do useful tools sometimes disappoint financially
A local task may improve while the wider process remains constrained by approvals, information or customer behaviour. Recurring costs may also absorb the gain. Following work from request to completion helps distinguish a convenient tool from an investment that improves company performance.
Businesses do not need a perfect productivity measure to make better digital decisions. They need a clear definition of the result, a credible comparison and an honest account of the resources involved. That discipline gives management a stronger basis for funding technologies that improve daily work and reconsidering those that add activity without delivering enough value.
References
1. US Bureau of Labor Statistics. Overview of BLS Productivity Statistics. Source.
2. Bianchini, M. and Lasheras Sancho, M. SME digitalisation for competitiveness The 2025 OECD D4SME Survey. OECD, 2025. Source.
3. OECD. The Digital Transformation of SMEs. 2021. Source.
4. Brynjolfsson, E., Li, D. and Raymond, L. Generative AI at Work. The Quarterly Journal of Economics, Volume 140, Issue 2, 2025, pages 889 to 942. Source.
5. International Organization for Standardization. Quality management The path to continual improvement. Source.
6. Brynjolfsson, E., Li, D. and Raymond, L. Generative AI at Work. Author manuscript, arXiv 2304.11771, revised November 2024. Source.