Skip to content

Compounding Intelligence

An AI project can leave the company something that makes the next one faster and less costly. We work to make that happen.

Models will be available to everyone; what a company knows remains its own

Before long any company will be able to use the same models as its competitors, and what will set it apart is its data, the rules by which it decides and the experience of its people. It is like a balsamic vinegar cellar: anyone can buy the grape must, while the value lies in the set of barrels, that is, in what each vintage leaves to the next.

The principle of compound interest, applied to knowledge

A rule written for one agent also holds for those that follow, and a shared definition of “customer” serves every process that uses it. When this asset is not preserved, the next initiative starts from scratch and the company pays twice for the same work.

  1. 01Complaints handling
  2. 02Credit assessments
  3. 03Branch assistant
  • What gets built
  • What is already in place

What accumulates

  1. Process knowledge

    In practice
    Tasks, volumes, systems, constraints
    Where it lives
    Company Brain
  2. Data

    In practice
    Connected sources, shared definitions, an owner for every data item
    Where it lives
    Data Foundation
  3. Analysis and models

    In practice
    Indicators, forecasts and scores already proven
    Where it lives
    Data Intelligence
  4. Rules

    In practice
    Decision criteria, exceptions, each agent's limits
    Where it lives
    Company Brain and Governance
  5. Controls and security

    In practice
    Agent permissions, run log, evidence for checks
    Where it lives
    Governance and Security
  6. Integrations and components

    In practice
    Connectors to systems, access already authorised, reusable components
    Where it lives
    The client's platform
  7. Skills

    In practice
    Using, assessing, correcting, deciding
    Where it lives
    The people

From the first project to the third, in a manufacturing company

  1. 01

    Demand forecasting

    What gets built

    Sales and warehouse data connected, a single definition of product and of customer, a team prepared to work with forecasts

    What is already in place

    Nothing: it is the first

  2. 02

    Production planning

    What gets built

    The lines' capacity constraints and the priority rules

    What is already in place

    Connected sales and warehouse data, the product definition, demand forecasts

  3. 03

    Customer service assistant

    What gets built

    The answers on orders and deliveries

    What is already in place

    Product and customer master data, order status, priority rules, people who know how to work with an agent

We measure it on the process and on the project that follows

We compare the result with an agreed baseline, counting the time people spend on checks as well. At close we write down what the project leaves that can be reused, and at the start of the next one what was actually reused.

  1. An agreed baseline
  2. The time people spend on checks
  3. At closeWhat the project leaves that can be reused
  4. At the start of the nextWhat was actually reused

Why the next project often starts from scratch

The asset is usually lost because the project stays closed in on itself, with the rules written into one agent's instructions and the data prepared in a supplier's environment, or because rules and data end up in formats the company cannot export. There is also a cost to budget for from the start: writing a rule so that it serves other agents too, or agreeing a definition with every function that uses it, takes longer than solving the problem of the current project alone. The first project pays for that time, while the return shows from the second onwards, as we explain in the article The asset one AI project leaves to the next.

The company's asset in the three ways of working

The comparison describes common situations, not every contract, because much depends on how it is written.

What we look atProject by project, with an integratorA licensed platformWith JoinX
Where rules and data end upProject by project, with an integratorOften in individual agents' instructions and in the environments chosen for each projectA licensed platformIn the supplier's platform, in the formats the platform providesWith JoinXIn the company's systems, in open, exportable formats
What the second project finds readyProject by project, with an integratorDepends on how much the first documented and handed overA licensed platformThe platform's features, within what the licence coversWith JoinXThe definitions, rules, connectors and access left by the first project
Who can run the solution afterwardsProject by project, with an integratorUsually whoever built it, or whoever receives adequate documentationA licensed platformThe supplier and its partners, within the limits of the licenceWith JoinXThe company's people, another supplier, or us with Presidio
How you exitProject by project, with an integratorWith a handover, to be negotiated if the contract does not provide for oneA licensed platformBy exporting data and configurations in the ways the supplier allowsWith JoinXWith the knowledge transfer, written into the contract from the first project

Starting with the first process, with the second already in mind