Cosensible

Our method

AI that speaks in your name is built differently.

When software talks to your customers, every sentence is yours. So we set constraints that do not depend on a model's goodwill: they live in the code, they are checked on every answer, and they are measured every day.

Two developers at their screens
  • 8rules
  • 0invented figures
  1. 1

    The AI only says what it knows

    Our products answer from the company's own data: catalogue, product pages, instructions, history. The language model does the phrasing; it does not decide the content. When the data is missing, the product says so and steps aside, rather than filling in the gap.

  2. 2

    Figures are checked, not hoped for

    Prices, quantities, totals, delivery times: every number in an answer is compared with what was given to the model. A number from anywhere else gets the answer rejected; the model is told about its mistake and tries again. Amounts on screen are read back from the database and rendered from a template, never copied out of a sentence.

  3. 3

    Customers' language, the way they write it

    Darija in Latin letters, Arabic, French, English, typos, several languages in one sentence: our products understand what people write on a phone, not what they would write on a form. A lexicon maintained with our customers, three searches combined, and an answer in the language of the person asking.

  4. 4

    One customer, one isolated database

    Every company is partitioned inside the database, on every table. Automated tests verify that a user can never read a row belonging to another company, even by tampering. One merchant, one space; nothing shared.

  5. 5

    Data is handled by your rules

    Our customers' data, and their own customers' data, is encrypted, backed up every day, exportable at any time, and hosted in data centres that comply with the laws of your country, the location being set out in your contract. Before any call to a model, identities are pseudonymised, and processing is declared to the competent authority.

  6. 6

    Cost is measured from day one

    The risk with an AI product is that inference costs more than the subscription. We log the cost of every answer, we cap the context and the history, we cache identical questions, and we set a quota per customer. That is what makes a 249-dirham subscription possible.

  7. 7

    A human stays in charge

    Our products advise, sort, prepare and hand over. They take no payment, turn nobody away, and send nothing binding without approval. Every conversation, every suggestion, every unanswered question is visible to the customer, who corrects it in one gesture.

  8. 8

    Installed without a developer

    A product that demands an integration project is no use to a small business. Ours install in a few minutes, from the customer's dashboard, on the platforms they already use.

These eight rules are verified by automated tests on every release, and we answer any question about our methods. Together with our AI charter, they form the framework we hold ourselves to; they line up with the responsible-AI frameworks of the continent's authorities, with the NIST AI Risk Management Framework and with the ISO/IEC 42001 standard, which we take as our reference.

See the method at work.

Install Conseilor on your shop tonight: fourteen days' trial, no credit card, and a first qualified order before the day after tomorrow.