What Drives the Cost of an AI Project in East Africa

This is the single largest swing factor. Clean records in one system with reliable identifiers is a different engagement from the same requirement spread across three systems, part of it on paper, with names spelled inconsistently and no shared key. The modelling work barely changes; the engineering ahead of it changes enormously.

It is also the variable you can improve before procurement. Knowing which systems hold the records, in what format, and how they reconcile turns a wide estimate into a narrow one.

Each integration carries its own cost: credentials, a counterpart who owns that system, a test environment, rate limits, and a failure mode to handle. Integrations against a vendor system with no documented interface and no responsive owner are the ones that quietly consume schedules.

Counting integrations explicitly in the brief, and naming an owner for each, is one of the cheapest things a buyer can do to reduce their own price.

Calendar time is cost. A project where security review, data access and sign-off take a fortnight each behaves differently from one where those decisions happen in days, and public sector procurement and audit requirements legitimately add steps.

This is why assessment phases exist. A two to four week assessment prices the delivery phase against observed reality, including how quickly your own organisation can move.

Data residency requirements, sector regulation and the choice between cloud and on-premise hosting all change the architecture and the recurring bill. Regional data protection law is real: Uganda's Data Protection and Privacy Act, 2019, Kenya's Data Protection Act, 2019, and comparable regimes elsewhere in the region impose obligations that belong in the design rather than in a later remediation project.

Recurring cost deserves its own line in the evaluation. A cheaper build on an expensive architecture is not cheaper.

Handover scope moves price legitimately. Repositories, infrastructure access, documentation, and training for the team who will operate the system all take real work, and omitting them produces a lower number and a long-term dependency.

Support scope does the same. A defined post-handover window, ours is typically around ninety days, is a different commitment from an open-ended retainer, and the two should never be compared on headline price alone.

Describe the process, not the technology. List the systems involved and name an owner for each. State where the data may live. Say explicitly that you expect code and infrastructure ownership at handover, and ask for assessment and delivery to be priced separately. Quotes built that way can be compared; quotes built from a one-line brief cannot.