You add AI services to a GSA Schedule through existing MAS SINs — primarily 54151S for AI professional services, 511210 for AI software, and 518210C when the offering is cloud-delivered. The work is in the labor categories and the pricing: AI titles are new enough that a sloppy LCAT description or an unsupported rate draws a clarification letter almost every time.
I spent eighteen years in federal acquisition as a Contracting Specialist and Contracting Officer at GSA, IRS, DoD, and DOI, and I now build these offers for a living. Agencies are buying AI services through the Schedule today, under SINs that have existed for years. The contractors losing that business are not the ones without an AI SIN — there is no waiting for one — they are the ones whose Schedule still describes the company they were in 2019.
Which GSA Schedule SINs cover AI services?
There is no single "AI SIN." AI offerings map to the existing MAS Information Technology SINs based on what you actually deliver: labor hours map to 54151S, software licenses to 511210, and cloud-hosted platforms to 518210C. Most AI services firms need 54151S first and add the others only if they sell product.
| SIN | What it covers | AI offerings that fit |
|---|---|---|
| 54151S — IT Professional Services | IT labor by the hour: development, integration, data services, systems analysis | ML model development, LLM integration, RAG pipeline builds, data engineering for AI, AI strategy and governance consulting |
| 511210 — Software Licenses | Commercial term and perpetual software licenses | Licensed AI products, ML platforms, AI-enabled COTS applications |
| 518210C — Cloud Computing and Cloud-Related IT Services | Cloud services and the professional services to implement them | AI delivered as SaaS, hosted model APIs, cloud AI platform implementation |
| 54151HACS — Highly Adaptive Cybersecurity Services | Specialized cybersecurity services | AI-enabled security operations, adversarial testing of AI systems — only if cyber is genuinely your offering |
Two cautions from the review side. First, do not chase every SIN — each one you add carries its own experience and documentation requirements, and a thin justification on a second SIN can slow the whole offer. Second, watch the NAICS mapping: custom AI development sits in NAICS 541511/541512 territory, which is exactly where 54151S points, so a services-first firm rarely needs anything more exotic. GSA has been steering emerging-technology purchases through these existing SINs and curated buying channels rather than standing up new special item numbers, so build for the structure that exists.
How do you write AI labor categories a Contracting Officer will accept?
Write each AI LCAT with a functional description of duties, minimum education, and minimum years of experience — broad enough to survive five years of task orders, specific enough that a CO can price it. The failure mode is copying a job posting: postings describe a person; LCATs describe a category.
As a Contracting Specialist at GSA, I reviewed hundreds of applications and saw the same mistakes repeatedly on new-technology labor categories: titles with no functional description behind them, "or equivalent experience" language with no substitution table, and single-tier categories that forced the contractor back into eMod within a year to add senior levels. Structure AI LCATs in tiers from day one.
| LCAT | Functional responsibility (summary) | Typical minimums |
|---|---|---|
| Machine Learning Engineer (I/II/III) | Designs, trains, evaluates, and deploys machine learning models; builds data pipelines and MLOps infrastructure; integrates models into production systems | Bachelor's in a technical field; 2/5/8 years by tier; allow degree-for-experience substitution |
| AI/Prompt Engineer | Designs and optimizes prompts, retrieval strategies, and evaluation harnesses for large language model applications; tunes model behavior against accuracy and safety criteria | Bachelor's; 2–4 years software or data experience; write duties broadly — the title will evolve, your LCAT should not have to |
| AI Governance Specialist | Develops AI risk-management documentation, impact assessments, and compliance artifacts; maps systems to the NIST AI Risk Management Framework and agency AI policy requirements | Bachelor's; 4+ years in IT governance, risk, or compliance |
| Data Scientist (I/II/III) | Applies statistical and machine learning methods to derive insight from structured and unstructured data; designs experiments and communicates findings | Bachelor's or Master's in quantitative field; 2/5/8 years by tier |
One structural decision matters more than any wording: keep the category names close to what the market already uses. A CO pricing "Machine Learning Engineer II" can find comparables in minutes. A CO pricing your invented "Cognitive Solutions Architect — Applied GenAI" cannot, and what a CO cannot benchmark, a CO questions.
How do you price AI labor categories defensibly against CALC data?
Benchmark every proposed rate against awarded Schedule rates in GSA's CALC / buy.gsa.gov pricing tools, then document why your rate sits where it sits. AI titles command premiums, but a premium without a written justification reads as an unsupported rate — and unsupported rates are the single most common reason IT offers bounce back.
- Pull comparables. Search buy.gsa.gov pricing tools for the closest awarded titles — machine learning engineer, data scientist, senior software engineer — filtered to comparable experience and education minimums.
- Triangulate where CALC is thin. Newer titles like prompt engineer have fewer awarded comparables. Bridge with adjacent categories plus published compensation surveys, and say explicitly in your narrative which proxy you used and why.
- Anchor to your commercial reality. Under Transactional Data Reporting (TDR) you will not submit Commercial Sales Practices data, but your proposed rates still need to reconcile with what you actually bill commercially and on other government work. COs check.
- Write the price narrative rate by rate. For each LCAT: the comparables, where your rate falls in that range, and the justification for any premium — scarcity of cleared AI talent, niche domain expertise, demonstrated outcomes.
- Leave escalation room. AI compensation is moving fast. Propose a defensible annual escalation method rather than starting high and hoping.
From the CO seat, the applications that got approved fastest were the ones where the pricing narrative anticipated my questions — the offeror had already told me why their ML Engineer III sat above the CALC median before I had to ask. Across our 70+ proven GSA contract awards, that one habit shortens negotiation more than any other.
What does a CO actually review on an AI-heavy Schedule offer?
The same four things as any services offer — relevant experience, LCAT quality, price reasonableness, and compliance documents — plus two AI-specific pressure points: whether your project experience proves you have delivered AI (not just talked about it), and whether your rates survive benchmarking.
- Corporate and project experience. Two years in business and relevant project documentation per SIN. For AI services, the projects must show AI delivery — a web-development project relabeled "AI-adjacent" is the kind of stretch reviewers flag immediately.
- LCAT descriptions. Functional duties, education, experience, substitution methodology, and tier logic that holds together.
- Price reasonableness. Comparison against CALC data, other awarded rates, and your own commercial or government billing history, negotiated under the MAS pricing framework in GSAR 538.270.
- Compliance and clause posture. Standard MAS documentation, plus awareness that GSA now applies AI-specific contract language to Schedule holders — your offer should not contradict obligations you will sign onto at award.
- Consistency across the offer. The technical narrative, the LCATs, and the pricing must describe the same company. When the narrative promises ML engineering but the LCATs are generic programmers, the whole offer loses credibility.
What if you already hold a Schedule — how do you add AI to it?
Existing contract holders add AI through modifications in eMod: an Add SIN modification if you lack the right SIN, an Add Labor Category modification for new AI LCATs under a SIN you hold, and pricing support for both. This is weeks-to-a-few-months of work, not a new offer.
The sequencing I recommend to clients: add the AI labor categories under your existing 54151S first — that is the fastest path to quotable AI rates — then pursue additional SINs only when a real opportunity demands them. Every LCAT mod needs the same defensible pricing package described above, so build the CALC benchmarking once and reuse it. If you want the SIN strategy, labor categories, and rate justification built by people who have sat in the reviewing chair, start with our GSA Schedule consulting page.
What Is the Bottom Line?
- Do not wait for an AI SIN. AI services sell today through 54151S, AI software through 511210, and cloud-delivered AI through 518210C.
- Write tiered, market-standard LCAT names. "Machine Learning Engineer II" prices in minutes; invented titles trigger questions.
- Benchmark every rate against CALC before submission. Justify premiums in writing, rate by rate, and bridge thin comparables with adjacent categories.
- Prove AI delivery in your project experience. Relabeled legacy projects are the most-flagged weakness on AI-heavy offers.
- Existing holders: use eMod. Add AI LCATs under 54151S first, then add SINs only when opportunity justifies the documentation burden.
Frequently Asked Questions
Is there a dedicated GSA SIN for artificial intelligence?
No. GSA routes AI purchases through existing MAS SINs — 54151S for professional services, 511210 for software, 518210C for cloud — supplemented by curated buying guidance for agencies. Structure your offer around the SINs that exist rather than waiting for a special AI item number.
Can I add a prompt engineer labor category to my GSA Schedule?
Yes. Write it with a functional description covering prompt design, retrieval optimization, and LLM evaluation, plus education and experience minimums. Because awarded comparables are still thin, support the rate with adjacent LCATs like software or data engineers and a written proxy justification.
What is CALC and how do COs use it?
CALC — the Contract-Awarded Labor Category tool, now part of GSA's buy.gsa.gov pricing tools — is a database of hourly rates already awarded on GSA contracts. COs use it to benchmark your proposed rates against the distribution for similar titles, so you should run the same search before you submit.
How long does it take to add AI labor categories to an existing Schedule?
An Add Labor Category modification through eMod is typically a weeks-to-a-few-months process depending on your assigned center's queue and the quality of your pricing support. A complete package with CALC benchmarking and a clear rate narrative moves fastest.
Do AI services require special compliance documentation in a MAS offer?
The offer itself follows standard MAS requirements. But GSA applies AI-specific contract language to Schedule holders, and ordering agencies increasingly request AI risk-management evidence — such as NIST AI RMF alignment — at the task-order level, so have that documentation ready even though the Schedule offer does not demand it.
Can a small business win AI task orders against large integrators on Schedule?
Yes — the Schedule is where small AI firms are most competitive, because set-aside orders under FAR 8.405 and small-business participation goals work in your favor. The prerequisite is having the SIN, the LCATs, and the rates already on contract when the order drops; agencies rarely wait for you to mod.