OpenAI's Decisions API went generally available October 22, 2026. Here's what service businesses need to know about using it to route leads and triage inquiries.
Ido Cohen · Published 2026-10-23 · AI for Service Business
OpenAI quietly shipped one of the most practical AI tools of the year this week: the Decisions API went generally available on October 22, 2026, and it could fundamentally change how service businesses handle the first 150 milliseconds of every new lead. If you run a plumbing company, dental practice, law firm, or any other service business that lives and dies by inbound inquiries, this one is worth your full attention.
Think of it as a very fast, very cheap "traffic cop" for AI. Instead of asking a full language model to read a lead, think about it, write a summary, and figure out what to do next — which is slow and expensive — you ask the Decisions API a specific question with a fixed list of possible answers and get one back in roughly 150 milliseconds.
According to TechCrunch's coverage, the launch marks OpenAI entering "the market for fast, cheap intelligence" — a segment that already pushed rival startup TypeSafe to a $7.6 billion valuation on the back of its competing Jev model. The Decisions API is built on GPT-6 Luna and, as OpenAI describes it, returns "typed answers" — a probability, a choice from your list, or a score on a scale — rather than free-form text. It can process text or images, accept up to 400 questions per request, and charges only for input tokens at $0.10 per million (with cached inputs dropping to $0.01 per million). Output is free.
That pricing structure means a plumber triaging 1,000 inbound web form submissions a month — a serious volume for most service businesses — would spend roughly $0.10 or less on the decision layer alone.
One to three seconds doesn't sound like a long time. But in a service-business lead funnel, it's an eternity.
According to reporting by Unite.AI, the API reduces task completion times from the standard 1–3 seconds for a full model call down to 100–180 milliseconds. That's a roughly 10x speed improvement, confirmed across multiple early customers who reported the gains in OpenAI's launch materials. Speak, an AI tutor platform, reportedly saw 10x faster response times in their integration.
For a service business, here's why that gap matters in practice:
The speed isn't about vanity metrics. It's about keeping humans from noticing the seam where AI takes over from human judgment.
The Decisions API is purpose-built for what DataCamp's tutorial calls "repeated, narrow decisions where the next action already exists in the surrounding system." Service businesses run on exactly those decisions, all day, every day.
Here are the four that matter most:
1. Lead routing and qualification
A contact form comes in. Within 150ms, the Decisions API can answer: Is this lead high-fit, medium-fit, or low-fit? Is this an emergency or a scheduled inquiry? Route to the emergency line, the sales team, or the nurture sequence? Tools like storypros.io have already documented how to wire this into HubSpot or Salesforce via webhook — you define the possible answers (e.g., book_now, nurture, spam, human_review), send the form data as input, and take the label back.
2. Inbound inquiry triage
For businesses handling dozens or hundreds of service requests per day — HVAC, legal, dental, financial advisory — a triage layer that classifies severity before a human even opens the ticket saves real time. A question like "How urgent is this service request?" with options ranging from emergency_dispatch to schedule_within_week to general_inquiry gives your dispatcher or front-desk staff a sorted queue instead of a raw inbox.
3. Review and reputation signals
The API accepts images as well as text. For a med spa or dental practice managing Google reviews and social comments, a Decisions API call can classify incoming reviews as positive_feature_mention, complaint_requiring_response, spam, or neutral before any staff member sees them — triggering different response workflows automatically.
4. Chatbot "next action" decisions
If you're running any kind of website chat or AI receptionist, the Decisions API is exactly what OpenAI designed it for: "choosing an agent's next action." Rather than asking a full model to reason about what comes next in a conversation, you ask the Decisions API: does this message mean the customer is ready to book, still researching, or escalating a complaint? The agent then acts on a label rather than reasoning from scratch each time.
You might be thinking: "I already run leads through ChatGPT or a Zapier AI step — is this actually different?" The honest answer is yes, meaningfully so — but with limits.
The Decisions API sits between dumb keyword rules (which break on anything unexpected) and full AI generation (which is overkill and slow for yes/no decisions). According to FourWeekMBA's pricing breakdown, OpenAI's $0.10 per million input tokens is about 2.4x the price of TypeSafe's competing Jev model, so it's not the cheapest option in its category — but it runs on GPT-6 Luna, which many developers already use, eliminating a vendor to manage.
The practical tradeoff: use the Decisions API for classification steps where you just need a label, and a full chat model for anything where a human will actually read the AI's output.
General availability on October 22 isn't just a status change. According to Cryptobriefing's launch coverage, OpenAI added several meaningful capabilities between the October 6 public beta and the GA release:
On compliance: zero data retention is supported and the API is HIPAA eligible. If you operate in healthcare-adjacent services, that's a non-trivial signal that OpenAI built this with regulated industries in mind.
The Decisions API is genuinely useful, but it's not magic and it's not for everyone on day one.
It requires a developer. This is an API — a raw programming interface. You won't log into a dashboard and flip a switch. You or a developer need to write code that sends your lead data to the endpoint and acts on the response. Tools like Make.com, Zapier, and n8n are already being used to wrap API calls without full coding, but you still need someone who understands the plumbing.
It only returns labels you define. If your intake process has genuinely complex, nuanced routing logic that changes frequently, the closed-answer format is a constraint. The Decisions API cannot surprise you with an answer you didn't anticipate — which is mostly a feature, but occasionally a limitation when edge cases fall outside your predefined buckets.
It's priced higher than direct competitors. At $0.10 per million input tokens versus TypeSafe Jev at $0.042, you're paying a premium for the OpenAI ecosystem integration. At low volumes, the dollar difference is negligible. At enterprise scale, it adds up.
Confidence scores need thresholds. The API returns not just a choice but a confidence score. That's powerful — you can auto-route high-confidence decisions and flag low-confidence ones for human review. But you need to set those thresholds intentionally, and you need to monitor them over time as your customers' language and behavior shift.
Start with one narrow decision and a human fallback. Don't try to automate your entire intake process out of the gate.
You don't need to rebuild your entire intake funnel this Friday. Here's a staged, low-risk path to get value from this before your competitors do:
1. Identify your highest-volume repetitive triage decision. What's the one question your front desk or dispatch team answers a hundred times a week? "Is this a new or existing customer?" "Is this an emergency or can it wait?" "Is this request in our service area?" Pick one.
2. Map the possible answers. The Decisions API requires a closed list. Write out every answer your team would give to that question. If it's more than six, simplify — either the decision is too complex for this tool, or you're conflating two decisions that should be separate.
3. Talk to a developer (or your Zapier/Make.com builder) about a webhook from your intake form to the API. The endpoint is documented at OpenAI's developer site. The plumbing is a few hours of work for someone who knows their way around APIs. Ask them to build a version that logs every decision with its confidence score — you'll want that data for the next step.
4. Set a confidence threshold for human review. Before going live, decide: at what confidence level do you trust the API to act automatically, and below what threshold does a human review the call? A reasonable starting point is auto-route above 0.85 confidence, flag for human review below that.
5. Run it in shadow mode for two weeks. Have the API make its call on every inbound inquiry, but have your team still make the real routing decision. Compare the two. If the API is right 90%+ of the time, you have a strong signal to start acting on its output.
---
What exactly is OpenAI's Decisions API?
It's a developer endpoint — generally available as of October 22, 2026 — that accepts text or images plus a question with a fixed list of possible answers, and returns one answer within roughly 150 milliseconds. It's built on GPT-6 Luna and is designed for classification, routing, and scoring tasks, not for writing text or generating explanations.
Does my service business actually need this, or is it only for big tech companies?
If you handle inbound leads, service requests, or support inquiries at any meaningful volume — even 100 per month — there's a real use case here. Lead routing, urgency triage, and review classification are all small-business problems. The barrier is that you need a developer or a no-code automation builder to wire it up, which isn't trivial.
How much will this cost a typical service business?
At $0.10 per million input tokens with free output, a small service business routing 5,000 leads a month through a short form would spend well under $1 on the decision layer. Cost is not the obstacle — implementation is.
Is the Decisions API HIPAA compliant? Can dental offices or med spas use it?
OpenAI has flagged the Decisions API as HIPAA eligible and supports zero data retention. That makes it a viable tool for healthcare-adjacent service businesses, but you still need a Business Associate Agreement (BAA) with OpenAI and should consult your compliance team before routing patient-identifiable data through any API.
How is this different from what I'm already doing with ChatGPT in Zapier?
A standard ChatGPT call in Zapier takes 1–3 seconds, generates free-form text, and requires you to parse that text into a decision downstream. The Decisions API returns a typed label (e.g., emergency, standard, spam) in ~150ms, with no parsing required — your system can act on the label directly. It's faster, cheaper per call, and far easier to build reliable automations around.
---
Sources: