In-product AI assistant
An assistant grounded in each customer’s own data inside your product, with tenant isolation enforced at retrieval rather than in the prompt.
AI Consulting for B2B SaaS Companies
Ship AI features your customers will pay for, with the unit economics, evaluation and multi-tenant isolation that turn a demo into a product line.
Frameworks I design against
- GDPR
- EU AI Act (transparency)
- SOC 2
- ISO 27001
- Customer DPAs
Short answer
Rubra helps B2B SaaS companies ship AI features that hold up commercially: multi-tenant retrieval with strict isolation, evaluation that survives customer-specific data, unit economics that work at your price point, and the security evidence enterprise buyers demand.
Last reviewed
Where I see this working
Customer-facing document intelligence
Extraction and summarisation over documents your customers upload, usually the feature with the clearest willingness to pay.
Support deflection
Answers grounded in your documentation and prior tickets, measured on genuine resolution rate rather than deflection rate.
Onboarding and data migration
Mapping messy customer data into your schema during onboarding, historically the slowest and most manual part of enterprise sales.
For SaaS companies the hard part of AI features is rarely the model. It is shipping something that works across thousands of tenants whose data you have never seen, at a cost that survives your price point, with security evidence that clears enterprise procurement.
Evaluation when you cannot see the data
Your internal evaluation set is drawn from your own documents. Your customers’ data looks nothing like it: different vocabulary, different structure, different quality. A feature that scores well internally can fail badly on a real tenant.
The workable pattern is a synthetic evaluation corpus deliberately built to span the variation you see across customers, plus opt-in evaluation with design partners on their real data, plus online metrics that tell you per-tenant quality in production. Without the last one you will hear about failures from churn rather than from monitoring.
Unit economics before launch
Instrument cost per request, per tenant and per feature before the feature is generally available, not after the first invoice surprises someone. Set per-tenant ceilings. Expect the distribution to be heavily skewed. In every deployment I have measured, a small number of accounts drove most of the spend, and knowing which ones changes both pricing and product decisions.
Isolation is architectural
Tenant filtering happens server-side in the retrieval query, derived from the authenticated session. It is tested adversarially in CI. A cross-tenant leak is the one failure in this category that ends customer relationships, and the control for it cannot live in a prompt.
Frequently asked questions
How do you guarantee tenant isolation in a shared retrieval system?
Isolation is enforced in the retrieval query itself, not by instructing the model. Tenant identity is a mandatory filter applied server-side before search executes, derived from the authenticated session rather than from anything the client sends. I test this adversarially: deliberate attempts to retrieve across tenants form part of the evaluation suite and run in CI. Prompt-level instructions to respect tenancy are not a control and should never be treated as one.
How should we price an AI feature?
Start by knowing your cost per active user, which requires per-request cost attribution before launch rather than after. Usage-based pricing aligns with cost but is unpopular with buyers who want predictable bills; a higher tier that includes AI is easier to sell and puts the variance risk on you, which is manageable once you have the data and have set sensible ceilings. The failure mode I see most is launching a flat-priced AI tier without knowing the cost distribution, then discovering that a handful of heavy accounts consume the margin from everyone else.
What do enterprise buyers ask about AI features?
Consistently: where does inference run and can it be pinned to a region, is our data used for training, how is tenant isolation enforced, what happens to prompt logs and for how long, which sub-processors are involved, and what evaluation evidence supports your accuracy claims. Having crisp answers and a published sub-processor list removes weeks from enterprise procurement. It is one of the higher-return pieces of work available, and it is mostly documentation.
Get a straight answer on your AI roadmap
A 30-minute call with the engineer who would do the work, not a salesperson. You will get an honest read on what is worth building, what is not, and roughly what it costs.
No NDA needed to talk. EU and UK hours in full, with afternoons overlapping US Eastern and Central.