Production AI for your ecommerce stack —
scoped, built, and shipped in 4–12 weeks.

Most ecommerce AI implementations run 4–12 weeks end-to-end, depending on data complexity and integration count. A single-model Shopify inventory forecasting build typically takes 6–8 weeks. A multi-system recommendation engine across Shopify, a data warehouse, and an email platform takes 10–12 weeks. The first working result is always in week 4 or earlier — you're not waiting until week 12 to see anything. Scope is fixed before we start, so the timeline is firm, not an estimate.

A milestone is one specific, verifiable deliverable with a date and a fixed cost. A project is a series of milestones. We always start with a single milestone — you decide after seeing the first working result whether to continue. This structure means your risk is capped at one milestone at a time, not an open-ended engagement.

Engagements typically run between $12,000 and $60,000 per milestone, depending on data complexity, the number of integrations, and the type of model. A single Shopify inventory forecasting implementation is toward the lower end. A multi-system recommendation engine across Shopify, a data warehouse, and an email platform is toward the higher end. We price by milestone, not by hour — you know the exact cost before work starts. Book a call for a specific number based on your use case.

We assess data readiness before committing to a milestone, so you don't need to have everything perfect upfront. For inventory forecasting, you typically need 12–24 months of order history and current stock levels. For recommendations, transaction history and product catalogue data. For reconciliation, transaction exports from your payment processors. If there are gaps, we either fix them as part of scope or recommend a separate data infrastructure engagement first. We'll tell you exactly what you need on the first call.

No. The AI Fit Sprint is for businesses that aren't sure what to build or which opportunity has the best ROI. If you already know the problem — stockouts, reconciliation, recommendations — we scope directly into an implementation. The sprint is a 30-day assessment; if you're past that question, skip it.

It means the AI runs in your real environment, on your real data, connected to your real systems. It's not a notebook, a demo, or a prototype you have to rebuild before shipping. It's monitored (you'll see when it drifts), documented (you'll know what every parameter does), and comes with a runbook so your team can operate it without us after handover. You also own the code and model weights — no licensing back to you.

We use a mix of open-source ML — scikit-learn, XGBoost, PyTorch, and Hugging Face where relevant — and LLM APIs (OpenAI, Anthropic) for use cases that benefit from language models. We don't require proprietary tooling. Everything runs in your cloud environment (AWS, GCP, or Azure) on infrastructure you already control. We won't introduce a dependency that locks you into a third-party platform.

You do. Full IP transfer is standard on every milestone. We don't license models back to you, we don't require ongoing access to keep them running, and we don't retain any claim to the code. The handover includes the full codebase, model weights, data pipeline configs, and a runbook.

We assess data readiness before committing to a milestone. If there are meaningful gaps — missing history, inconsistent IDs, broken integrations — we'll tell you on the first call. We either include a data-cleaning step in the milestone scope, or we recommend addressing infrastructure first in a separate engagement before the AI build begins.

Ecommerce: Shopify, WooCommerce, Amazon Seller Central, Magento, and custom stacks. Data warehouses: Snowflake, BigQuery, Redshift, and direct database access. Accounting: QuickBooks, Xero, and NetSuite. Shopify and WooCommerce are the most common — we have pre-built connectors that cut integration time significantly. If you're on a platform not listed, tell us on the call — we've integrated with most stacks in the $1M–$20M ecommerce range.

AI consulting typically ends with a report, a roadmap, or a recommendation deck — you pay for advice, then still need someone to build. AI implementation ends with working software in your production environment. At TwoDots, we do not do consulting-only engagements. Every engagement ends with a deployed, documented system that your team can operate. If you need help deciding what to build first, the AI Fit Sprint is a 30-day assessment that produces a ranked opportunity list — but it ends with a roadmap, not code. Implementation is the next step after that.

A SaaS AI tool is a generic model trained on data from many businesses, accessed through a subscription dashboard. It's fast to start and low-cost upfront, but it can't learn your specific data patterns, you don't own the model, and you pay monthly indefinitely. Custom AI implementation means the model is trained on your data, runs in your environment, and is handed over to you permanently — no ongoing licensing fee. For ecommerce businesses doing $1M–$20M, the SaaS route often hits a ceiling: the model improves up to a point, then plateaus because it doesn't have enough signal from your specific catalogue, customers, and operations. Custom implementation is the right choice when you've tried a SaaS tool and it's not moving the metric, or when the use case is specific enough that a generic model won't cut it.