> ## Site Index
> Fetch the site index at: https://www.credibledata.com/llms.txt
> Use this file to discover all available pages before exploring further.

# How Form & Function built an AI OS for services firms

> How a Boston advisory firm turned twenty years of services-firm know-how into Lever, an AI OS built on open-source Malloy Publisher.

## Overview

[Form & Function](https://www.form-function.co/) is an operating partner to professional services firms and the investors behind them. On November 12 it [launched Lever](https://www.financialcontent.com/article/bizwire-2026-11-12-form-and-function-consulting-launches-ai-os-for-professional-services-firms), an AI OS that connects a firm's pipeline, delivery, staffing and financial data and helps its leaders forecast revenue, plan capacity, price work and test scenarios. Lever is built on a data model written in [Malloy](https://www.malloydata.dev/), served by [Malloy Publisher](https://github.com/malloydata/publisher), the open-source server Credible built and maintains.

## The challenge: the rules live in people's heads

A services firm's operational data sits in its CRM, its project and time-tracking tools, and its finance system. The rules for reading them are often maintained as tribal knowledge. This setup keeps complex scenario planning and financial forecasting out of reach for most firms that have become dependent on ad-hoc data exports and a tight circle of individuals pulling that data together into spreadsheets.

AI on its own can only solve so much of this problem. Pointed at the raw data, it sees none of the context behind the numbers. It doesn’t know that expected deal revenue must be spread out over time, that utilization for salaried staff is calculated differently from contractors, or how to rationalize “placeholder” staff allocations. This leaves it guessing while answering with confidence anyway. For a firm deciding whether to hire, a number that looks right and isn't is worse than none at all.

So Form & Function ran a small test. The team [put the same five questions](https://www.form-function.co/ideas/why-your-ai-data-analyst-wont-work-without-a-semantic-layer) about a client's real operational data to Google Gemini twice: once on the raw BigQuery tables, and once on a Malloy data model. Each question ran five times and the results below speak for themselves. Data flowing through Malloy was faster to retrieve and more accurate.

> With Malloy we get a query language easier for LLMs and humans, and a semantic layer that captures our business definitions. I can’t recommend it enough for conversational analytics.
>
> — Adam Ribaudo, Partner, Form & Function

## With and without Publisher, by the numbers

- **$4.1M vs. $2M.** Asked for expected 2026 deal revenue, the raw-table AI counted every deal in full on its close date and roughly doubled the answer. The model returned the right figure in 32 seconds and two queries.
- **24 seconds vs. 130.** For average billable utilization of salaried staff, the model needed two queries every time. The raw-table AI averaged 130 seconds and ran into its 40-tool-call limit.
- **Under a minute** for every answer on the model. On raw tables, the hardest questions never finished.
- **About $1.8M** in new annual revenue one smaller client found capacity for, without adding headcount.

## Twenty years of advising, written down as a data model

Form & Function's edge was never the software. It's two decades of advising services firms on how they sell, staff, deliver and measure work. Lever is that experience written down where an AI can use it, in three layers: a Malloy data model that holds the math, a domain-specific semantic layer that grounds the AI's understanding of services firms, and operating principles that are specific to each client.

### The math lives in the model

The model defines what utilization, backlog, pipeline and project health mean, and how the CRM, project and finance data join up. Joins are declared in the model, so a one-to-many join can't quietly double-count revenue. As Adam Ribaudo, Partner at Form & Function, put it: "Malloy provides the guardrails to avoid traps like query fan-outs, while also giving us the hooks we need to inject our domain expertise as context to agents." The AI composes its answer from that context rather than guessing.

### Every client starts from what Form & Function already knows

A new client doesn't start from a blank page. Onboarding connects the firm's systems, reads its documents, and interviews its people through Lev, Lever's assistant. The answers fit Form & Function's model of how services firms work to the way this firm works. Publisher then serves the result over MCP, so the same definitions answer in Slack, in ChatGPT or Claude, and in Lever's own web interface.

### Built in the open, with Credible

Form & Function’s introduction to Credible was made through Scott Breitenother, one of Credible's angel investors. Scott founded Brooklyn Data, a well-regarded data consultancy, which was acquired by Velir where Adam worked at the time. At the time of the introduction, Adam was looking for semantic layer vendors who felt as strongly as he did that context management would make or break agentic analytics solutions. His early experiences with both the Credible team and the wider Malloy open source community convinced him to commit to Publisher as the engine of his multi-tenant operational analytics solution. Adam’s commitment to the open source project has resulted in several improvements ranging from faster compile checks to hardening Publisher’s security posture as it relates to multi-tenant deployments. Adam has gone on to write up his lessons learned working across multiple semantic layer vendors in a [comparison of two production deployments](https://www.noisetosignal.co/ideas/semantic-layers-in-production/).

## The impact: an AI OS Form & Function owns

1. **Correct where it counts.** The model carries the business rules an AI on raw data can't see, like when revenue counts.
2. **Fast to roll out.** Each client starts from Form & Function's model of a services firm, so onboarding takes days.
3. **One model, every surface.** The same definitions answer in Slack, ChatGPT, Claude and Lever itself.
4. **Owned by Form & Function.** The model, the product and the choice of AI are theirs, on a stack nobody can take away.
5. **Better for everyone.** What Form & Function needed from Publisher shipped in the open-source project.

> We use Lever in our own work because we need a continuous, up-to-date understanding of the businesses we work with.
>
> — Dave Valliere, Founder & CEO, Form & Function

## What's next: the package that fits

Lever is live in Form & Function's advisory engagements, where its consultants use it alongside the firms they advise.

Form & Function chose open source, and Credible is glad to support it. As Credible [explained in July](/blog/posts/open-sourcing-skills), trust comes from letting people look inside, and an open foundation gets better every time someone builds on it. There are three ways in, all first class:

- **Run the open source yourself**, the way Form & Function does. No account, no sales call, no meters.
- **Plug it into the agent you already use.** Point Claude, ChatGPT or Gemini at any Publisher server, yours or Credible's.
- **Use Credible**, if you'd rather it run for you: connect your data, tell Credible what it means, and build apps and agents on it, with access control built in.

It's the same data model whichever you pick, and you can move between them any time. The goal is that everyone can use this technology, in the package that makes the most sense for them.

### About Form & Function

[Form & Function Consulting](https://www.form-function.co/) is an operating partner to services firms and investors, based in Boston. It combines twenty years of hands-on advisory experience with Lever, its AI OS for services firms. Form & Function is a Kantata Consulting Partner.

**Build on the open-source core, or let Credible run it for you.**

[Malloy Publisher](https://github.com/malloydata/publisher) is free and open
  source: clone it and start building today. Or start on Credible and get the
  same data model, run for you, with apps, agents and access control built in.
