Atlas Labs operates at the intersection of deployment and research, measuring how automation reshapes labour demand in small and medium-sized businesses.
"We don't advocate for automation: we document what happens when it becomes economically rational."
01 / Purpose
Automation adoption in small and medium-sized enterprises is accelerating. Businesses adopt these technologies for survival and competitiveness, not ideology. Yet most research on automation's labour impact remains theoretical or focused on large enterprises.
Small businesses adopt automation to reduce costs and compete. The impact on their workforce is immediate and measurable.
Atlas Labs provides empirical data from actual deployments, not simulations or projections based on theoretical models.
Current focus: We're actively researching automation impact in UK education, specifically voice AI systems handling school absence reporting. This sector offers rich data on workflow displacement, staff reallocation, and safeguarding considerations unique to public services.
02 / The Wider Field
The landmark studies shaping how economists think about AI and work. Worth reading in full.
Brynjolfsson, Li & Raymond · NBER, 2023
A Fortune 500 call centre gave agents a generative AI assistant. Productivity rose around 14%, with the largest gains for the newest workers: the tool effectively transferred the habits of top performers.
Noy & Zhang · MIT, Science, 2023
Professionals given ChatGPT finished writing tasks roughly 40% faster at higher measured quality, and the gap between strong and weak writers narrowed rather than widened.
Dell’Acqua et al. · Harvard × BCG, 2023
Consultants using GPT-4 outperformed on tasks inside AI’s capability and did worse when they trusted it beyond that edge. Knowing where the frontier sits is worth more than having access.
Anthropic Economic Index · 2025
Analysis of millions of real AI conversations shows the technology augmenting human work more often than automating it outright, with usage still concentrated in a narrow band of occupations.
Almost all of this evidence comes from large firms and controlled settings. The SME evidence, where most of the workforce actually sits, barely exists. That is the gap we are filling.
03 / Metrics
Our measurement framework captures the full lifecycle of automation's labour impact, from initial deployment through transition outcomes.
Time savings per automated workflow, measured against baseline operations
Admin, support, operations, and sales support functions impacted by automation
Distinguishing between prevented hires and actual role reductions
Duration from deployment to measurable labour substitution
Redeployment, reskilling, hours reduction, and natural attrition tracking
04 / Findings
Early aggregate indicators from internal pilots (illustrative). These figures represent preliminary data and should not be cited as definitive findings.
of automation impact occurred in admin & support roles
days median time-to-impact
FTE avoided per SME engagement
of saved time reinvested into higher-value work
Most displacement in our sample occurred through avoided hiring, not layoffs. SMEs typically chose not to backfill roles or expand headcount rather than terminate existing employees.
05 / Open Questions
The debates that keep this field interesting. Our deployments put us in a position to collect evidence on each of them.
Reinvested into higher-value work, absorbed into slack, or converted into shorter weeks? The answer decides whether automation compounds or just quietly disappears.
Our early data points to avoided hiring rather than layoffs. Whether that holds as agents take on larger shares of the workflow is the question that matters most for policy.
Validation layers and accuracy exams turn "trust me" into "check it". We are measuring whether that is what actually moves SME owners from pilot to production.
Tasks that resisted automation twelve months ago fall today. What predicts the order, and how should a five-person business plan around it?
If agents absorb the junior work, where does the next generation learn the trade? Nobody in the field has a convincing answer yet. We want one.
06 / Experiments
We evaluate what works in practice, not what sounds ideal. These are active experiments in managing the human side of automation adoption.
Converting displaced roles into AI oversight and quality assurance positions
Exploring productivity gains shared through shorter work weeks
Moving affected workers to growth areas within the same organisation
Timing automation with planned departures and retirements
Piloting models where efficiency gains translate to worker benefits
07 / Methodology
A practical, transparent approach to measuring automation's workforce impact.
Capturing operational baselines prior to automation deployment
All client data pooled without identifying information
Erring toward understatement in displacement calculations
Explicit uncertainty ranges on all published figures
08 / Next Steps
As automation adoption accelerates, understanding its labour impact becomes an infrastructure problem, not a philosophical one. We're expanding our research capabilities to meet this challenge.
Atlas Labs deploys automation in real businesses and measures what happens next.