Get early access
Menu

Off-the-shelf decision models don’t know your business. Yours will.

Train an open decision model on how your deals, tickets, and meetings actually turned out. It’s live in 30 minutes, answers fast, and costs little to run.

Get early accessSee how it works
SOC audited · GDPR compliantVisit the Trust Center

Data to a decision model in 30 minutes.

Your data

  • Salesforce
  • HubSpot
  • Slack
  • Attio
  • Notion
  • Gmail
  • Zendesk
  • Pylon
  • Intercom
  • Linear
  • GitHub
  • Fireflies
  • Stripe Connect
  • Airtable
  • Gong
  • 99+

Training today from Attio, Pylon, and Fireflies

Decisions LabExample run
  1. Ask a question

    Will this deal close in 7 days? Yes / no

  2. Review the data plan

    We draft a plan from your records. You check it before anything trains.

    The outcomeWhich dealsDetails to useDeals kept for testing

  3. Train Kev 4B

    Pick a starting model. It learns from your past deals and how they turned out.

  4. Check the results

    Tested on past deals it hasn't seen, next to a simple guess.

Your model

Deal close forecast Will this deal close in 7 days? PreparingStartingLive · Version 1Version 2
Who can use it
Your teamPublic
Endpoint
POST …/predict
API key
dl_live_••••7f3aCopyCopied
Keeps learningRetrain every week
  1. New outcomes arrive

    Your connected data records how each deal ended.

    Renewal · ClosedExpansion · Didn't close
  2. Version 2 trains

    On schedule, using your saved setup.

  3. You choose

    Check it first. Version 1 stays live until you switch.

One model per question.

Example models for Attio, Pylon, and Fireflies. Each answers new records with how likely each answer is.

See what your model learned.

Test on examples kept out of training, compare with a baseline, and inspect the mistakes before you use it.

Deal close forecastVersion 1 · Kev 4B · tested on 6 deals it never trained on
Example run Attio
Your model
5 of 6 right
Baseline
3 of 6 rightAlways answers No, the most common outcome.
To inspect
1 mistakeExpansion, APAC: answered No at 57%. It was won.
Held-out deals: what happened, the baseline’s answer, and your model’s answer.
Held-out dealWhat happenedBaselineYour model
Renewal, 120 seatsLegal reviewing the order form.WonNoYes 84%
New logo, fintechWent quiet after pricing.Not wonNoNo 79%
Upsell, support teamSigned off by the VP.WonNoYes 88%
Pilot, retail chainChampion left the company.Not wonNoNo 72%
Expansion, APACSecurity review still open.WonNoNo 57%
Inbound, 10 seatsAsked for a quote. No call booked.Not wonNoNo 66%

Start from a recipe.

First models worth trying with the tools you already use. Pick the base model each one trains.

A growing world of decision models.

We’re starting with Clef-Flash and Kev, fine-tuned to your data with LoRA. Get to know the wider ecosystem, too.

Ready to train today: Clef-Flash, Kev 0.8B and Kev 4B. More open models are coming to Decisions Lab; closed models link to their research.

A few useful answers.

About your data, training, and what a model can and can’t tell you.

Is this only for businesses?

No. Decisions Lab is for individuals, founders, developers, and teams with a specific question and examples to learn from. Use a model for your own tasks, or build it into a product.

Do I need to be an ML engineer?

No. You start by describing your goal in plain words. You still review the prepared data and choose how to use the results. For now, the first training run and deployment need setup support from us.

What data can I start with?

Training currently starts from Attio, Pylon, or Fireflies. Use examples with known outcomes or labels, and only information available at prediction time. For other sources or your own datasets, contact us about the setup.

Start with a question you can evaluate
Do I see what goes into training?

Yes. Before anything trains, we draft a data plan from your records: the outcome to predict, which records to use, the details the model can see, and the examples held back for testing. You review it first.

How do I know whether training helped?

Run an evaluation on held-out examples. Inspect accuracy, mistakes, and available baseline comparisons. Training doesn’t guarantee an improvement; the results help you decide whether a version is useful.

What to check before you deploy a model
Does a prediction guarantee an outcome?

No. Predictions are estimates based on the data and model used. They can be wrong. Use them alongside your judgment and other information; you remain responsible for the actions you take.

How do I use a model once it’s live?

Ask it in the app, call it from your product with an API key, or run it across records in a workflow you can save and run again. Each prediction records the version that made it.

What happens when new outcomes come in?

On the schedule you set, a new version trains on the fresh outcomes. It’s a candidate: you check its results first, and your live version keeps answering until you switch.

Retraining is a new candidate, not an automatic upgrade

Still have a question?

Tell us what you want to predict. We’ll help you set up your first model.

Talk to usEmail us

A closer look.

Practical guides to the data, models, and choices behind the result.

View all articles
Evaluation · 2 min read

What to check before you deploy a model.

Use held-out examples, inspect the mistakes that matter, and compare available baselines before deciding whether a trained prediction model is ready to use.

Read article Data preparation · 2 min read

Start with a question you can evaluate.

Turn a broad goal into a prediction task with a clear outcome, usable examples, and inputs that reflect what you will actually know at prediction time.

Read article Model lifecycle · 2 min read

Retraining is a new candidate, not an automatic upgrade.

Understand why fresh outcomes can justify another training run, what to compare between model versions, and why deployment remains a separate review step.

Read article

Start with one question.
Make the model yours.

Get early accessTalk to us