STEP-BY-STEP

SF AI Playbook

This Playbook, developed by the City and County of San Francisco’s Emerging Technologies Team, offers guidance to help City departments use artificial intelligence (AI) responsibly in their services. It outlines key steps to consider when exploring AI adoption or planning pilot projects.

Department of Technology (DT)

The SF AI Playbook helps City staff use AI in a careful and fair way to make services better for everyone in San Francisco.

It explains how to choose the right problems to solve, decide if AI can help, use good data, focus on equity, measure results, and stay open with the public. Staff can follow the scan–pilot–evaluate–scale steps: start small, test ideas, learn what works, and scale successful projects.

Print version

SF AI Playbook + Implementation Worksheet

1

Define the Problem You Aim to Solve

Before choosing any AI approach, start by clearly defining the problem and the outcome you want to achieve. Describe who will be affected, what will improve, and why it matters. Keep the problem statement neutral, and connect it to City goals to ensure the work delivers clear public value.

2

Assess Whether and How to Use AI

AI isn’t always the right tool. First, confirm whether the problem needs AI at all. If it does, choose the simplest approach that works.

3

Put People and Their Rights First

AI tools can sometimes make mistakes, be unfair, or invade people’s privacy. They can also change the kinds of skills workers need and how their jobs are done. Identify these risks early, weigh them against public benefits, and address them with enforceable safeguards and mitigations.

4

Make Sure the Data Fits the Task

AI systems are only as good as the data they use. If the data is incomplete or biased, the results will be too. Make sure the data used is accurate, complete, and fair, and follow City data policies for how they are stored and managed.

5

Set Up an Evaluation Plan Before Starting Your Pilot

Evaluate technical performance, public impact (with specific attention to equity, civil rights, and community trust), and overall outcomes. This means assessing whether the AI tool works as intended, and whether it advances equity, protects rights, and builds public trust.

6

Assess Whether to Build or Buy

Decide whether to build in-house or buy from a vendor based on staff skills, capacity, and existing infrastructure. First, see if the City already has tools that can be adapted for your needs. Only add new technology when it clearly advances public value.

7

Disclose AI Use Through the AI Inventory

Be transparent about AI use. City law (Chapter 22J) requires departments to disclose AI systems they use, along with their purpose. These disclosures are published in the public AI Use Inventory.