Automation
Where automation actually pays first
The automations that return their cost quickly are almost always document-shaped and high-volume: invoice and receipt capture, order intake from messaging channels, payment reconciliation, recurring reporting. What rarely pays first is anything requiring judgement on thin data — pricing, credit, dispute resolution. Our test is simple: count the touches, multiply by the time each takes, then check the process runs the same way twice. If it does not, fix the process first, or you will automate the inconsistency and make it permanent.
Buying Technology
Build, buy or integrate
Buy where the process is genuinely standard — accounting, payroll, email. Integrate where two systems that already work refuse to talk to each other; this is the most common right answer and the most under-used. Build only where the process is your actual commercial advantage, or where no product fits without forcing you to work around it. A large share of “we need custom software” briefs turn out to be undecided processes, and software makes indecision expensive and permanent.
Transformation
Why transformation programmes stall
Five failure modes account for most of them. No single accountable owner with authority to decide. A scope that starts too wide to produce a visible result. A success measure nobody baselined before the build. Rollout treated as a launch announcement rather than a change programme. And a system that never integrated with the numbers finance already trusts, so the organisation quietly keeps two versions of the truth. Each has a governance answer, and every one is cheaper to prevent than to recover from.
Applied AI
AI without the hype: what it is actually good at
Applied AI is reliable today at bounded work with a checkable output — extracting fields from documents, classifying and routing, drafting from your own material, summarising a record. It is unreliable as an unsupervised decision-maker on thin or contested data. So we deploy it with confidence thresholds, an exception queue with a named owner, and a full log of what ran and why. If the business cannot supervise it, we do not ship it. That constraint is a feature, not caution.
Operations
The hidden cost of running a business on WhatsApp
Message-based operations feel free and are not. The cost appears as rework, as disputes you cannot settle because the record lives in somebody’s phone, as institutional memory that leaves when staff do, and as a month-end nobody fully trusts. The migration path is not “stop using WhatsApp” — people will not, and the channel is genuinely good. It is to keep the channel your customers and staff already use and put one operational record behind it.
Web
Your website is a sales asset. Measure it like one.
Four numbers tell you whether a website earns its place: qualified enquiries per month, cost per qualified enquiry, real load time on a mid-range phone over mobile data, and the share of visitors arriving directly on pages with commercial intent. Traffic, bounce rate and time on page tell you very little. Instrument the four before redesigning anything — otherwise the redesign has no baseline to argue with, and you are left debating taste.
Mobile
Designing for the mid-range Android
Applications tested on a flagship phone in an office with good wifi fail in the market. The decisions that determine whether an app survives real conditions are taken at the very start and cost a rebuild to retrofit: offline capture with a genuine sync and conflict strategy, small payloads, lean build size, and flows that assume a session can end mid-task. Mobile money and cash on delivery are payment flows to design around, not integrations to bolt on at the end.
Data
One version of the truth your finance team accepts
Reporting fails on definitions, not technology. If finance and operations do not already agree what counts as an order, a completed job, or revenue recognised, no dashboard will reconcile them — it will just give both sides a more confident way to disagree. The unglamorous work decides the outcome: write the definitions down, name an owner for each measure, agree the quality rules, then build. That sequence is why some dashboards get used and most get ignored.
Market Entry
Systems for companies entering West Africa
The assumptions that do not transfer are specific and predictable: saved-card payments, reliable street addressing, email-first identity, always-on connectivity, and a stable back-office workforce. Systems designed elsewhere usually break on the first two, then get blamed for everything downstream. Plan instead for mobile money, landmark-based addressing and delivery instructions, phone-number identity, offline tolerance, and processes documented well enough to survive turnover.
