Your KPI definitions decide the number
A plain-English question resolves to the metric in your semantic model — so a KPI means the same thing in a chat as it does on your dashboard.
Get the analysis a senior analyst would produce — in plain English, in minutes, without waiting on the data team.
Every answer resolves to your KPI definitions and stays inside your access controls — so the number in a chat matches the number on your dashboard.
Illustrative — the shape of a governed answer, drawn with synthetic data.
High promise for AI — it should invite skepticism
Already built —a decision-intelligence system for the asset-management team at one of the largest global REITs, answering performance-review questions on their own enterprise data.
Delivered work02 The difference
A decision you’re accountable for needs a different system.
A general chat assistant is built for open-ended work. A governed decision system is built for the number you will act on.
Microsoft Copilot and Claude are excellent at open-ended work. The difference here is what sits under the answer: your KPI definitions, your row-level security, and the fields, filters, and logic behind the number.
Two jobs, side by side
General chat assistantA general chat assistant
A general chat assistant
(Copilot, Claude)
Best atOpen-ended drafting, summarizing, reasoning over text and code.
Answers fromWhatever data or text you hand it, in the moment.
Right whenThe work is exploratory and a person checks the output.
This pageGoverned decision intelligence
Governed decision intelligence
(this)
Best atQuestions you will act on, answered against your KPI definitions.
Answers fromYour governed semantic model, with row-level security applied.
Right whenSomeone is accountable for the decision and the number has to hold up.
- The hard part: any tool can return a number. A decision-grade answer needs a system that knows your metrics, your rules, and your access model — and reasons like a senior analyst.
- Who it’s for: teams accountable for decisions on a governed model.
- Not for: a raw chatbot over ungoverned tables, or a staffing bench sold as a capability.
- SquareShift builds the governed layer underneath, so the business gets speed and you keep control of what a metric means.
See it in your industry
Try it on a question you would actually ask.
Pick your industry, then open the question a business user in that seat brings to your team.
Store performance and sales trends
“Which stores beat their target last quarter, and what drove it?”
Try this Retail & E-commerceInventory and stock-out risk
“Which SKUs are about to stock out before the next replenishment cycle?”
Try this Retail & E-commercePromotion and discount effectiveness
“Which promotions actually lifted margin instead of just moving volume?”
Try this Retail & E-commerceCustomer basket and repeat purchase
“What do our repeat customers buy together that new customers don’t?”
Try this Retail & E-commerceCategory margin and price mix
“Which categories are dragging down margin even though revenue looks fine?”
Try this Retail & E-commerceFulfilment and delivery performance
“Which regions are missing delivery SLAs, and why?”
Try this Retail & E-commercePortfolio performance and exposure
“Where is our exposure concentrated, and how has it shifted this quarter?”
Try this Financial ServicesFraud and risk signals
“Which accounts show the fraud pattern we flagged last month?”
Try this Financial ServicesBranch and channel productivity
“Which branches are underperforming their peer group, and on what metric?”
Try this Financial ServicesRegulatory and compliance reporting
“Can you pull the numbers behind this quarter’s compliance filing?”
Try this Financial ServicesCustomer attrition and retention
“Which customers show the early warning signs of churn this month?”
Try this Financial ServicesLiquidity and cash position
“What does our cash position look like over the next 13 weeks?”
Try this Financial ServicesPortfolio occupancy and yield
“Which properties are dragging down portfolio yield right now?”
Try this Real Estate & REITsLease expiry and renewal risk
“Which leases expire in the next two quarters with no renewal signed?”
Try this Real Estate & REITsAsset-level operating cost
“Which assets have the highest operating cost per square foot this year?”
Try this Real Estate & REITsCapital project tracking
“Which capital projects are over budget, and by how much?”
Try this Real Estate & REITsTenant mix and concentration risk
“How concentrated is our rent roll in any single tenant or sector?”
Try this Real Estate & REITsEnergy use and sustainability targets
“Which buildings are furthest off track on our energy targets this year?”
Try this Real Estate & REITsPatient throughput and capacity
“Which units are running over capacity during peak hours?”
Try this Healthcare & Life SciencesClaims and reimbursement trends
“Which claim types have the longest reimbursement delays this quarter?”
Try this Healthcare & Life SciencesClinical trial enrolment
“Which trial sites are behind on enrolment against plan?”
Try this Healthcare & Life SciencesSupply and formulary usage
“Which formulary items are trending toward a shortage?”
Try this Healthcare & Life SciencesStaffing and roster coverage
“Which shifts are understaffed against forecasted patient volume?”
Try this Healthcare & Life SciencesReadmission and outcome risk
“Which discharged patients are at highest risk of a 30-day readmission?”
Try this Healthcare & Life SciencesProduct usage and feature adoption
“Which features are new customers not adopting in their first 30 days?”
Try this Hi-Tech & SaaSChurn and retention signals
“Which accounts show the usage drop-off pattern that precedes churn?”
Try this Hi-Tech & SaaSPipeline and conversion analysis
“Where are deals stalling most in the pipeline this quarter?”
Try this Hi-Tech & SaaSSupport load and resolution time
“Which issue categories are driving the longest resolution times?”
Try this Hi-Tech & SaaSCloud and infrastructure cost
“Which services are driving the increase in our cloud spend this month?”
Try this Hi-Tech & SaaSRelease quality and incident trend
“Which releases correlate with a spike in production incidents?”
Try this Hi-Tech & SaaS03 How a trusted answer works
Every answer resolves to your KPI definitions, inside your access controls, with the calculation shown.
Four behaviors are what stop one KPI arriving at two different numbers in two decks. SquareShift engineers each one. No platform ships them by default.
01
02
Your access controls still apply
Your existing row-level security applies to every answer, and to every chart or report it produces.
03
The fields, filters, and logic are shown
Anyone can check the number before they act on it, and your team can check it after someone quotes it in a meeting.
04
Every number is read from your governed data
The number comes back from a query against your governed data. The model writes the explanation around it.
The answer, and the calculation behind itIllustrative · synthetic example
Governed answer
“Q4 revenue grew 6% — down from 11% in Q3 — with Asia-Pacific the main drag.”
MetricRevenue growth % = (this period − last period) ÷ last period
FiltersRegion: all · Period: Q4 vs Q3
LogicRevenue summed by region, then period-over-period change
PermissionScoped to your region and product access
Illustrative — a synthetic example, built for this page.
The architecture behind it
A governed semantic model on Looker, answered with Google’s Vertex AI and Gemini.
This is the path a business question travels before it becomes an answer you can defend.
Business question, asked in natural language.
Resolves to the metric definition in your semantic layer.
Runs inside that user’s row-level security.
Returns a governed answer with its calculation attached.
A follow-up question stays in the same governed context.
Looker — governed semantic modelAccess controls & row-level securityVertex AI & Gemini — Google model platformBigQuery — the data layerExplainability — shows the calculation
What SquareShift builds
SquareShift designs the semantic-model coverage, the permission model, the conversation-context handling, and the explainability. That governed layer is the engineering work; connecting a model to it is the easy part.
How we deliver
Your current reporting keeps running. We start on your own data, prove it on the decisions that matter, then expand.
01
Start with the first decisions
We set up the governed data behind your first questions — nothing you don’t need yet.
02
Test on real questions
We check the answers against the questions your team actually asks, before anyone relies on them.
03
Pilot with a feedback loop
A small group uses it for real, tells us what’s off, and we tune it.
04
Expand once it’s trusted
Add more decisions after the first set is working and trusted.
Start a working session on your own data.
A scoped session that ends with a working prototype on your own data.
04 Flagship engagement
Inside the decision-intelligence system we built for one of the largest global REITs.
One experience for the asset-management team — with the semantic layer, access controls, and audit trail underneath that keep every number consistent.
The architecture — one experience, governed underneath
To the user it is one experience. Underneath, responsibilities are separated so each answer stays governed and traceable.
01ExperienceHosts the user experience. The business user asks: “Which properties need attention, and what evidence explains the risk?”
02RuntimeA Cloud Run application hosts the experience, its services, and agent execution.
03OrchestrateAn agent orchestrator understands intent, checks scope, routes the work, and synthesizes the result.
04AgentsSpecialized agents — a data agent and an evidence agent — reason with Gemini.
05OutputA grounded response: metrics, explanation, citations, and an artifact you can act on.
The stack underneath
BigQuery + Vertex AI SearchGoverned data and document search.
Semantic layerBusiness terms resolve to one consistent meaning.
Business knowledge layerThe domain context the agents reason over.
Planning & orchestration layerIntent, scope, routing, and synthesis.
Answers, charting & reporting layerGrounded responses, charts, and artifacts.
The controls that keep every number consistent
01RBAC
02Grounding
03Feedback
04Audit
05Evaluation
06Observability
A performance-review question runs as a process: metrics → variance → reports → commentary → a point of view. SquareShift’s system does the retrieval, reconciliation, and first-pass synthesis, so the asset manager spends the time interpreting and deciding.
01
Business workflow
What a useful answer actually looks like.
02
The underlying data
So answers stay consistent.
03
Business meaning — the semantic layer
So business terms resolve correctly.
04
Orchestration and routing
So the right source answers the right question.
05
Controls
So every result is traceable and reviewable.
The governed data-and-BI foundation this stands on
Google Cloud Premier PartnerLooker Consulting PartnerGlobal EdTech learning platformDelivered near-real-time governed Looker reporting, replacing lagging nightly-batch BI.
American semiconductor manufacturerDelivered centralized sales, partner, and training dashboards in Looker Studio, cutting analyst dependency.
Digital skilling platform, 1.5M+ learnersDelivered an end-to-end Domo-to-Looker migration with business continuity.
26,000+ BI and Looker assets migrated across the practice.
Delivered work
A conversation layer is only as good as the reporting underneath it.
Reporting a business can defend — a BI migration, real-time analytics, and compliance reporting SquareShift delivered.
Digital learning platform
Daily reporting for 1.5M+ learners moved from Domo to Looker without going dark.
Read the case study Global EdTech companyNightly batch replaced with real-time Looker numbers that match the web experience.
Read the case study International aviation group40,000 device records validated on Elastic; compliance reports audit-ready on demand.
Read the case study05 Book a session
Book a Decision Intelligence working session.
Bring one decision that currently waits in the analyst queue. You leave the session with a scoped view of what a governed answer to it takes, on your own data.