Services

AI & Intelligent Automation

Most of what is sold as enterprise AI is a demonstration. The useful question is narrower: which specific piece of work in this business is repetitive, high in volume, rule-shaped, and currently done by someone who could be doing something better.

The situation

What we usually find

For organisations under pressure to have an AI position, who would rather have one working process than five pilots.

Kenya published a national AI strategy in March 2025, and the pressure on boards to show an AI position has risen steadily since. That pressure produces poor decisions: pilots with no owner, tools bought before the problem was named, and proofs of concept that impress in a meeting and never reach anybody’s daily work.

The work that actually pays is unglamorous — invoice and delivery-note capture, document classification, reconciliation matching, demand forecasting, exception detection. It is measurable, it has an owner, and it either removes a cost or removes a delay. None of it makes a good slide.

Readiness decides the outcome. A model built on figures that disagree between systems produces confident wrong answers faster than a person produced uncertain ones, and anything touching customer or employee records brings the Data Protection Act with it — registration with the Data Protection Commissioner applies from five million shillings of turnover or ten employees. That is a lawful-basis question before it is a technical one.

What changes

From where you are, to where this gets you

Most organisations arrive at this practice from a recognisable place. This is the distance it covers.

What this practice changes. Documents keyed in by hand, one at a time becomes: Extracted, validated and posted for approval. Forecasts built on last year plus a feeling becomes: Forecasts built on what the data actually shows. Problems noticed when somebody complains becomes: Anomalies flagged as they appear. Skilled people spending their week on repetition becomes: Skilled people on the work that needs judgement.

  1. Documents keyed in by hand, one at a time

    Extracted, validated and posted for approval

  2. Forecasts built on last year plus a feeling

    Forecasts built on what the data actually shows

  3. Problems noticed when somebody complains

    Anomalies flagged as they appear

  4. Skilled people spending their week on repetition

    Skilled people on the work that needs judgement

What this practice covers

The work itself

Engagements draw on whichever of these the situation needs. Very few use all of them, and we will say which we think apply before you commit to anything.

Working out what is worth doing

An honest assessment of where automation would pay and where it would not, before any commitment. Most candidate ideas should die at this stage, and most do.

  • AI strategy
  • AI readiness assessment

Taking out the routine

The high-volume, rule-shaped work that consumes skilled people’s weeks — captured, classified and routed without anyone keying it in.

  • Intelligent automation
  • Intelligent document processing
  • Process mining

Seeing what is coming

Forecasting and detection built on your own operating history, so planning starts from what the data shows rather than from last year plus a feeling.

  • Predictive analytics
  • Demand forecasting
  • Anomaly detection

Putting it in people’s hands

Interfaces that let people ask a question in their own words and get an answer from the systems they already have.

  • Enterprise AI assistants
  • Natural language interfaces
  • AI-enabled reporting

Built for your case

Where nothing off the shelf fits the problem, built to your process and run as a service rather than handed over as a project.

  • Custom AI solutions
  • AI as a Service

How it runs

What happens if you call us

Every engagement has a decision point at the end of each stage. You can stop at any one of them, and what you have paid for up to that point is yours to take elsewhere.

  1. 01

    Assess readiness honestly

    What data exists, whether it agrees with itself, what volumes actually run, and where skilled people’s time is going. This stage frequently concludes that the data has to be fixed first, and we will say so.

  2. 02

    Score the candidates

    A shortlist of processes scored on volume, clarity of rules, data quality and value at stake, with the scoring shown. Most ideas do not survive contact with this list, which is the point of having it.

  3. 03

    Build one, properly

    The strongest candidate only, built into production and measured against how the work is done today. One process working end to end is worth more than four pilots, and it is the thing that earns the second one.

  4. 04

    Hand it over with its rules

    Monitoring, a named owner, and a written statement of what the system decides on its own and what it escalates to a person — because the day it is wrong is the day that matters.

What you get

The things you keep

Advisory work is easy to buy and hard to hold on to. These are the artefacts that remain with you afterwards, and they belong to you whether or not the engagement continues.

  • A readiness assessment covering data quality, systems, volumes and lawful basis
  • A scored shortlist of candidate processes, with the reasoning behind each score
  • A working automation in production — not a demonstration, not a sandbox
  • A measured before-and-after against how the process ran previously
  • Written human-in-the-loop rules: what the system decides, and what it must escalate

Where this is not the answer

We do not sell AI as a category. If the answer is a better-designed form, a report the business already owns, or a rule configured in the ERP, that is what we will recommend — it will cost less, it will work sooner, and it will still be working in two years.

Explore AI Services

Tell us what you are trying to improve and we will tell you, plainly, whether we are the right people for it.

Book an Assessment

Also in services

Other practices

Back to top