How teams use GRAM
From one-person data teams to enterprise SaaS platforms. Compile the graph from your schema, hand it to any AI, ship answers.
The one-person data team
A 150-person distribution company. The “data person” is the ops manager who built the original Access database and migrated it to SQL Server. She gets 15 ad-hoc data requests a week from sales, warehouse, and finance. She connects the database, imports the schema, builds the graph in one sitting. Now the VP of Sales checks pipeline on his phone. The warehouse supervisor checks stock levels between shifts. She went from bottleneck to enabler in an afternoon.
Schema import · Analyst/Strategist workflow · mobile dashboards · SMB sweet spotThe consultant’s first-week win
A BI consultant walks into a new healthcare client engagement. Day one, she connects to their PostgreSQL claims database, imports 40 tables, and has a working graph by lunch. By Friday, the client’s operations director is looking at a live dashboard showing denial rates by payer, average time-to-payment, and top billing codes. The graph itself becomes a deliverable: a structured map of what the client’s data means, versioned in Git.
Speed-to-value · graph as artifact · consulting use case · unfamiliar databaseThe multi-location executive cockpit
A restaurant group with 12 locations across three states. One analyst builds a graph against their POS database. Each regional manager gets a portal with their locations only. Daily covers, food cost percentage, labor ratio, comp sales. The GM checks from the floor on a tablet. The owner checks all 12 from the airport. No Power BI, no Tableau, no training.
Portals · mobile dashboards · non-technical end users · bite-sized analyticsThe legacy database translator
A manufacturing company running a 20-year-old ERP with tables named OM1R06 and columns like CSTMR_NBR. No foreign keys. No documentation. The engineer who built it retired. The graph becomes the Rosetta Stone: each cryptic table and column gets a business alias, each unreadable foreign key gets a role alias, and missing relationships are declared explicitly. The AI reasons over clean business language while the SQL hits the original table names untouched.
The Snowflake accuracy rescue
A mid-market e-commerce company on Snowflake. They tried Cortex Analyst, spent two weeks writing YAML semantic models, and accuracy was still disappointing on anything beyond simple lookups. They point GramSpec at the same Snowflake warehouse, compile a graph from the schema, and the GRAM layer handles what YAML couldn’t: role-playing date dimensions, multi-path joins, derived metrics. Same data, same warehouse, different architecture, dramatically better results.
Complementary positioning · enterprise stack compatibility · separable accuracyThe product-embedded semantic layer
A SaaS company building a project management tool wants to add “ask your data” to their product. Their dev team builds the graph in Grammar, exports the JSON via the Builder tier, and wires it into their React app with their own LLM. Their customers get natural-language queries against their project data, powered by a graph the SaaS company maintains and versions alongside their codebase.
Graph export · Builder tier · open standard · any-LLM portability · ISV use caseThe question no metrics library anticipated
A quarterly review. The CFO asks which salespeople give the most aggressive discounts, weighted by pre-discount line value. No metrics library on earth carries that entry, because nobody predicted the question. The model derives the measure live from the graph’s constraints and the conversation, shows the SQL and the interpretation it chose, and the controller pins the answer to the board. The SQL freezes in the widget. The measure is now ratified, shared, and reproducible. Consistency by ratification, not by committee.
On-demand measures · disclosed derivation · pin to ratify · DashboardThe premise that was wrong
A VP asks the analytics chat why Q3 revenue is down. An answer machine would pick a comparison and produce a chart. The Analyst scopes the question first: down versus which quarter? Then it runs the queries and reports the actual truth: against the comparison the VP had in mind, Q3 is up 22 percent. It says so plainly, cites every query it ran, and lists what the VP might have actually meant. Wrong premises get corrected, not decorated.
Analyst · scope before solve · cited queries · premise checksLists ready to query
An operations team runs its world on SharePoint lists: orders, suppliers, regions, approvals. The lists sync to SQL, Grammar compiles the graph over the synced schema, and the same hard questions a warehouse graph answers run against the lists: Pareto breakdowns, cohort revenue, month-over-month change, weighted discount rates. And when a question needs a measure this particular graph does not carry, the system says so and says what is missing, instead of guessing. The lists stay in Microsoft 365. The analysis grows up around them.
SharePoint sync · Microsoft 365 · graph-scoped answers · honest refusalSee yourself here?
Connect your database, build the GRAM layer, start asking questions. It takes minutes, not months.