AtScale launches AI suite to boost enterprise accuracy

AtScale has launched AtScale for AI, a new suite aimed at improving accuracy in enterprise AI.

The product brings together existing and new components under what AtScale calls its AI Computation Engine, including tools for semantic modelling, question clarification, governed calculation and answer testing.

Central to the launch is the argument that large language models cannot reliably answer business questions unless companies first define terms such as revenue, reporting periods and calculation methods. AtScale says its software links natural-language queries to shared business definitions, allowing AI agents to use the same metrics as analysts and business intelligence systems.

Chris Lynch, Chief Executive Officer of AtScale, said the problem for companies is not only model quality but also the lack of agreed business logic. "The AI industry keeps telling enterprises to buy a smarter model, but a model can't reason its way to your definition of revenue. Enterprises need to define what an answer means, compute it correctly, and prove it holds up. AtScale for AI brings those capabilities together so accuracy becomes something businesses can measure, improve, and demand."

How it works

The suite includes AtScale CoModel, which uses AI to help teams build and maintain semantic models covering metrics, dimensions and relationships. Any proposed changes remain subject to human review.

Another part of the package, AtScale Ava, is designed to interpret user questions before a query runs. If a user asks for revenue, the system can ask whether that means bookings, recognised revenue or contract value, then pass the request on for calculation.

Those calculations are handled by ACE, which applies metric definitions and calculation logic to enterprise data. The aim is to produce consistent results across AI agents, applications and analytics tools.

Another element, AtScale Canary, is intended to test AI answers against validated results. AtScale says the framework draws on the BIRD-Interact benchmark as well as a customer's own business question-and-answer pairs to identify mistakes and monitor regressions over time.

Together, these tools form a feedback loop in which user questions and detected errors can be used to refine semantic models, while people retain control over changes to definitions and logic.

Customer evidence

HSBC is among the customers AtScale cited in support of the launch. Dr Kevin McLafferty, Head of Data & Analytics, UK Commercial Banking, HSBC, described the role of governed business definitions in the bank's use of AI-driven analytics.

"In banking, an AI answer that's almost right is still wrong. The challenge was never model intelligence; it was making sure every agent calculated metrics with precision and accuracy - no exceptions. By building on the governed definitions we already trusted, AtScale gave our agents that context and precision without reinventing the logic behind it: on a 10 TB table in Google Cloud, with AtScale as the semantic layer, accuracy on AI-driven deterministic analytics took a quantum leap in performance from 65% to better than 99%."

Carrefour also backed the approach. "AI shouldn't create a second version of the truth. At Carrefour, we already have a semantic layer governing thousands of KPIs across the business. Extending that foundation to AI means our agents can use the same definitions as our analysts and BI tools, rather than producing different answers depending on how someone asks the question," said Nicolas Treanton, Data & Analytics Director at Carrefour.

Broader push

The launch reflects a broader effort by AtScale to position the semantic layer as infrastructure for AI as well as traditional analytics. It continues to support Apache Ossie, formerly Open Semantic Interchange, as part of work on shared semantic standards intended to preserve business logic across models, agents and data platforms.

AtScale also pointed to separate research in which it evaluated seven large language models using BIRD-Interact. According to the company, adding its semantic layer improved accuracy across all the models tested, and those findings informed the design of Canary.

AtScale Ava and ACE are available now, while CoModel and Canary are due to follow.