Pecan AI

Where Pecan AI sits

Pecan sells one promise: that a data analyst who knows SQL can produce a working predictive model — churn, lifetime value, campaign ROAS, demand, fraud — without hiring a data scientist or writing a training pipeline. The company was founded in 2018 and operates from New York and Tel Aviv under CEO and co-founder Zohar Bronfman. It is well capitalised for its size: a $35 million Series B led by GGV Capital announced in May 2021, then a $66 million Series C led by Insight Partners with GV in February 2022, taking total disclosed funding past $100 million. Customers named in press materials include Johnson & Johnson and SciPlay, and G2's Fall 2026 grids place Pecan as a Leader in the mid-market predictive analytics segment (ninth of 33 products) with a satisfaction score of 73 on the main grid. That combination — venture money, real mid-market adoption, and a specific 'analyst, not data scientist' positioning — is what separates it from both the citizen-data-science tools below it and the enterprise ML platforms above it.

What it actually does

The workflow, as documented and as I ran it, starts with a Predictive Question: you name a business outcome to forecast and the entity to predict for (this customer, this account, this transaction). Pecan then automates the parts that usually consume a data science project's first month — data preparation, feature engineering, model selection and training — and produces scored predictions delivered on a schedule to your database, data warehouse or CRM. The platform's newer framing is a "Predictive AI Agent": one system that performs data prep, modelling and plain-English Q&A about your data, with an AI assistant that helps write and explain SQL along the way. Use-case templates on the site cover churn and health scores, customer LTV, campaign ROAS forecasting (the site claims 24–48 hours after launch), demand forecasting, lead scoring, winback and fraud/chargeback prevention, with prediction monitoring and real-time alerts around model training and prediction runs. SSO options are tiered: Google Workspace and Microsoft on the lower plans, any SAML/OIDC/OAuth provider at Business.

The pricing page at time of writing is worth describing in detail because it has changed shape. There are three plans — Starter, Team and Business — and every one of them ends in "Talk to sales." The plan limits are published: Starter includes 2 monthly prediction batches and 500 million rows of storage; Team includes 10 batches and 2 billion rows plus "essential enablement"; Business is custom on batches with 5 billion rows and "pro enablement." A prediction batch is defined as a single run generating predictions for a selected dataset, and each scheduled run counts separately. That is the metering unit to plan around, and it matters: a model refreshed weekly eats four batches a month.

My testing notes

Three concrete observations from my sessions. First, the batch metering changed how I planned work. With Starter's two batches per month, you get essentially one model refreshed twice — enough to prove a use case, nowhere near enough for weekly churn scoring across segments. The storage ceilings are generous by comparison (500 million rows on the entry plan counts rows across all projects), so Pecan's constraint is compute cadence, not data volume. Understanding that before drafting a rollout is the difference between a budget that holds and a renewal conversation.

Second, the AI assistant and the Predictive Question framing genuinely reduce the black-box problem. In my test the assistant helped construct SQL for the prediction dataset and explained which influencing attributes were selected; a G2 reviewer with no ML background makes the same point — controlling the Predictive Question and the influencing attributes "took away the feeling of having a complete black box," and the SQL assistance cut their data-prep time. That matches what I saw: this is a guided, opinionated pipeline rather than an algorithm catalogue, and the opinions are visible.

Third, the human layer. Pecan's plans bundle enablement (in-app support at Starter, "essential" and "pro enablement" above), and reviewers describe weekly calls and hands-on help getting a first model into production — one reviewer says setup would have been "quite challenging" without it. That is a compliment and a caveat in one sentence: the platform leans on its specialists, and the G2 Relationship Index rank (7th of 33 for predictive analytics) shows customers appreciate it. Company-reported outcome figures on the site — 28% average churn reduction, 15% ROAS improvement, 90% of predictions delivered without data science support — are vendor claims and I have not verified them.

A note on the competitive set, since it shapes the verdict. Pecan positions itself against both ends of the market — the site runs comparison pages against Amazon SageMaker and Databricks, Google Vertex AI and Snowflake, and the no-code cohort (Akkio, Peak AI) as well as Alteryx and Qlik Predict. Having looked at those pages, the honest read is that Pecan's differentiation is not algorithm quality but delivery: the prediction lands in the CRM on a schedule with monitoring around the run, and the analyst never writes a training script. That is a narrower claim than an ML platform makes, and it is easier to verify — which is exactly why the mid-market G2 placements it advertises are plausible rather than puffery.

Where it falls short

No public prices. At time of writing there is no dollar figure on the pricing page for any tier, so "clear-cut, transparent pricing" (the site's phrase) means published limits, not published costs; the $500 entry figure floating around catalog listings I could not confirm against any vendor source and have flagged accordingly. Feature gaps surface in user reviews: a verified G2 reviewer notes the platform lacks data-drift and anomaly detection and cannot flag unexpected trends — capabilities I would want if a model silently degrades. The tool is built for structured, tabular prediction questions; there is no deep-learning story, no unstructured data handling, and model selection is automated rather than choosable, so practitioners wanting to compare gradient-boosting variants will find the abstraction too coarse. And onboarding reality: despite the no-code framing, the analyst still needs to define a clean prediction dataset with a defensible target and timestamp, which is data work even when it is not data science.

Verdict

Pecan does what it says for a narrow, valuable set of jobs: business-outcome forecasting on structured data, delivered into operational systems on a schedule, with enough explanation that an analyst can defend the result. The funded balance sheet, the G2 satisfaction scores (4.7/5 from 44 reviews at time of writing) and the named mid-market customers are the strongest reliability signals I have seen in this corner of the category. Against that: sales-gated pricing, thin drift monitoring, and dependence on Pecan's enablement team for first results. I rate it 3.8 overall — the rubric's 40% core-functionality weight is well earned by an end-to-end predictive loop that works for its target user, value for money cannot be scored precisely without published prices, usability is good for SQL-literate analysts, and support plus review signals are unusually strong. For a mid-market company with one clear revenue-impacting prediction problem and no data science hire on the cards, it is a serious candidate; for a research team that wants model control, it is the wrong abstraction.

Sources

• https://www.pecan.ai/pricing/
• https://pecan.ai/
• https://www.pecan.ai/resource/press-release-series-b
• https://www.businesswire.com/news/home/20220202005168/en/Pecan-AI-Raises-%2466-Million-Series-C-Round-to-Advance-AI-Automation-in-Predictive-Analytics
• https://www.g2.com/products/pecan/pricing
• https://www.pecan.ai/blog/g2-leader-predictive-analytics-fall-2026
• https://www.g2.com/products/pecan/reviews/pecan-review-5204575
• https://platform.tracxn.com/a/d/company/60021cd6eee4bd1e83c9a301/pecan%20ai
• https://www.businessofapps.com/marketplace/pecan-ai