Platform & Model
Every recommendation, every prediction and every value atwork produces comes out of the atwork AI Middle Layer — our own set of models, built over years and trained on cross-organisational feedback and outcome data that took just as long to gather. The models can be described. The data behind them cannot be copied.
The architecture
Feedback and business data go in. Recommendations and quantified predictions come out. In between sits the part nobody else has.
Measure + Performance
Feedback management
API landscape
Act + Impact
Real-time dashboards
Analysis of the key drivers
Forecasting outcomes and their effects
Action management
The components
No single model does all of this. The layer is a set of specialised models that hand work to each other — and every one of them is ours.
01
The Impact Framework
Employee feedback, business data and years of survey history brought into one structure. Established scientific models describe how influencing factors, HR outcomes and business outcomes hang together — which is what connects every answer to a business outcome.
02
Machine Learning and a Multivariate Gaussian Graphical Model reveal how each survey variable influences your HR and business KPIs, directly and indirectly. The output is a quantified relationship, not a correlation left to interpretation.
03
Retrieval over three bodies of knowledge: best practice, thousands of real action plans including the ones that failed, and peer-reviewed science. It makes a recommendation specific to your industry and size instead of generic advice.
Powers Act
04
Meta-studies, the expertise of top universities and practitioners, and the accumulated evidence of what has worked across industries and company sizes. A new customer gets a credible answer long before they have their own history.
Powers Act
05
An artificial neural network with Natural Language Processing converts qualitative feedback into measurable data. What people write in free text becomes a variable in the model instead of a quote in an appendix.
06
Every output is narrowed to who it is actually for: organisational unit, employee group, industry and company size. The right action for a 40-person ward is not the right action for a plant with 2,000 people.
Powers Act
Privacy by architecture
Not a policy, a sequence. Identities are removed and free text is redacted before anything is processed — nothing identifiable ever reaches a model, and no general-purpose language model sees your raw data.
Answers are separated from identities, minimum group sizes are enforced, and Azure automatically redacts personal information from free-text comments before processing.
Filtering, retrieval, prediction and monetisation run inside atwork, on infrastructure hosted in Switzerland.
Only the finished, anonymised output is passed on for wording — with AI Act documentation and human oversight throughout.
