Business Workflow Automation Kenya: Automate the Right Things and Keep a Person Where It Matters
Business workflow automation Kenya is sold in two ways and bought for a third. It is sold as efficiency — fewer hours on repetitive work — and increasingly as intelligence, with AI agents that will handle whatever the business throws at them.
It is bought, in practice, because someone is tired: tired of typing the same M-Pesa confirmation into a spreadsheet, of calling the director for every approval, of sending the same reminder to the same customers on the same day every month, of generating invoices by hand from a template that is always one version out of date.
That tiredness is the right starting point, because the processes that exhaust people are almost always the ones worth automating — repetitive, rule-based, high-volume and error-prone — and the processes that need judgement, relationship or accountability are almost always the ones that should stay with a person.
The failures come from confusing the two: automating a decision that needed a human, or leaving a human on a task a rule could do. They also come from automating a process that was broken, so it now fails faster; from integrations that snap the first time a system changes; and from AI placed where its errors have consequences nobody will own.
This guide covers doing it properly: what automation is and is not, which processes qualify and which do not, mapping before building, triggers including M-Pesa payments and forms, approvals and routing, notifications, document generation, data entry elimination, connecting systems, scheduled tasks, exceptions, human-in-the-loop, where AI agents genuinely fit and where they do not, guardrails, escalation, audit, roles, testing, change management, measurement, common failures, maintenance, data protection, security, vendor questions and cost.
The value of business workflow automation Kenya is a business where people do the work that needs people, and business workflow automation Kenya that mapped the process, kept judgement human and treated AI as a component rather than a promise is what delivers it — which is why business workflow automation Kenya should begin with a process map rather than a tool.
Table of Contents
- What Automation Is and Is Not
- The Kenyan Context
- Which Processes Are Worth Automating
- Which Are Not
- Map the Process Before Automating It
- Triggers: What Starts the Workflow
- M-Pesa Payment as a Trigger
- Forms and Intake
- Approvals and Routing
- Notifications: SMS, WhatsApp and Email
- Document Generation
- Eliminating Data Entry
- Connecting Systems
- Scheduled Tasks and Reminders
- Exception Handling
- Human in the Loop
- Where AI Agents Genuinely Fit
- Where AI Does Not Belong
- Guardrails for AI in Workflows
- Escalation
- The Audit Trail
- Roles and Permissions
- Testing Before Go-Live
- Change Management and Staff
- Measuring Whether It Worked
- Common Failures
- Over-Automation
- Brittle Integrations
- Maintenance
- Data Protection and Automated Decisions
- Security
- Questions to Ask a Vendor
- Build or Configure
- Costs and Phasing
- Frequently Asked Questions
What Automation Is and Is Not {#what-is}
Precision about the term prevents buying the wrong thing.
Automation is a rule that runs without a person: when this happens, do that.
A trigger, conditions, actions and exceptions.
It is not custom software, though it may run inside it.
It is not AI, though AI may be one of its components.
It is not replacing a person; it is removing a task from a person.
It is only as good as the rule, since a rule that is wrong runs wrong every time.
The distinction matters, since business workflow automation Kenya that connects an M-Pesa confirmation to a receipt and a ledger entry is a rule, and business workflow automation Kenya that promises to handle customer queries intelligently is something else and should be assessed as such.
The Kenyan Context {#kenyan-context}
Local conditions shape what automation is worth doing.
M-Pesa confirmations arrive as messages and are re-entered by hand into spreadsheets and systems everywhere.
WhatsApp is how customers and staff communicate, which makes it the notification channel that is actually read.
SMS remains reliable for customers without smartphones and carries a per-message cost.
Approvals happen by phone call to a director, who is in traffic.
Invoices, receipts and quotations are generated by hand from templates.
Small teams do everything, so a task removed from a person is a real gain.
Staff turnover means process knowledge leaves with people.
Connectivity fails, so automations that depend on it must handle failure.
Informal processes vary by who is doing them that day.
Regulatory requirements including electronic invoicing develop and automation must keep pace.
Each of these is what business workflow automation Kenya should address, and business workflow automation Kenya designed for a market with card payments and email as the primary channel would automate the wrong things.
Which Processes Are Worth Automating {#worth-automating}
The candidates share characteristics.
Repetitive: done many times in the same way.
Rule-based: the decision can be written down.
High-volume: enough occurrences that the saving is real.
Error-prone when done by hand.
Time-sensitive: a delay costs something.
Cross-system: data moves from one place to another by re-typing.
Recurring on a schedule.
Examples: M-Pesa confirmation to receipt and ledger, invoice generation from an order, payment reminders, approval routing, onboarding steps, report distribution, stock reorder alerts, appointment confirmations.
The test is whether the rule can be written, since business workflow automation Kenya that starts from the tasks staff describe as mindless finds the candidates, and business workflow automation Kenya that starts from what a tool can do finds features looking for a problem.
Which Are Not {#not-worth}
Some processes should stay with people and the reasons should be stated.
Decisions requiring judgement about a person: hiring, discipline, credit to a customer in difficulty.
Relationship conversations: complaints, negotiations, bad news.
Anything where the rule cannot be written because the cases are all different.
Low-volume tasks where the automation costs more than the task.
Processes that are still changing, since automating a moving target means rebuilding monthly.
Processes nobody understands, since automating what is not understood encodes the confusion.
Anything where an error would be serious and nobody would notice.
Leave these alone, since business workflow automation Kenya that tried to automate a credit decision for a customer in arrears has removed the judgement the decision needed, and business workflow automation Kenya that automated the reminder and left the conversation to a person has drawn the line correctly.
Map the Process Before Automating It {#map-first}
Mapping is the step most projects skip and the reason most fail.
Who does what, in what order, with what information, using which systems.
Where the process actually starts and ends, not where people think it does.
Every decision point and its rule.
Every exception and what happens to it.
Where it fails today and why.
Documented, since a map in someone’s head cannot be built from.
Agreed by the people who do the work, since the owner’s version and the clerk’s version differ.
Automating a broken process produces a broken process that runs faster, and business workflow automation Kenya that mapped and fixed the process first has something worth automating, while business workflow automation Kenya that built the automation from the owner’s description will discover the exceptions the clerk knew about in production.
Fix the process, then automate it.
Triggers: What Starts the Workflow {#triggers}
Every automation begins with an event and the event must be reliable.
A payment received.
A form submitted.
A record created or changed.
A date or time.
A message received.
A threshold crossed: stock below reorder, invoice overdue.
A status change: order approved, delivery completed.
The trigger must be detectable by the system, since business workflow automation Kenya triggered by a status change that staff forget to set never fires, and business workflow automation Kenya triggered by an event the system itself records — a payment callback, a form submission — fires every time.
Prefer system events to human-set flags.
M-Pesa Payment as a Trigger {#mpesa-trigger}
The payment confirmation is the most valuable trigger in this market.
A Paybill or Till callback arrives with sender, amount, reference and transaction code.
Matched to a customer, an invoice, an order or an account.
On match: receipt issued, ledger updated, service activated, order released, customer notified.
On no match: held in suspense and a person alerted.
Verification against the provider before anything irreversible happens.
Idempotency so a repeated callback does not run the actions twice.
The M-Pesa integration itself is its own subject, and the reconciliation article on the estate side shows what happens without it, but business workflow automation Kenya built on the payment callback removes the largest single re-entry most businesses have, since business workflow automation Kenya that turns a confirmed payment into a receipt, a ledger line and an activated service in seconds has replaced a clerk, a spreadsheet and a delay.
Never release goods or services on an unverified callback.
Forms and Intake {#forms}
Forms are the trigger for anything that starts with a request.
Customer orders, applications, bookings, service requests, staff requests.
Structured fields so the automation has data rather than free text.
Validation at entry so bad data does not enter the workflow.
Attachments where needed.
Confirmation to the submitter.
Routing based on the answers.
The form replaces the phone call and the WhatsApp message that someone had to interpret, and business workflow automation Kenya that begins with a structured form has clean input for everything downstream, while business workflow automation Kenya that tries to automate from free-text messages is parsing rather than processing.
The church public pages article shows forms landing in the record without re-entry.
Approvals and Routing {#approvals}
Approvals are where directors spend their phone time and where automation earns trust.
Rules for what needs approval and by whom.
Thresholds: below this amount, auto-approve; above, route to a manager; above that, to the director.
Routing to the right person by category, amount or site.
Notification to the approver with the information needed to decide.
Approval or rejection from the phone.
Escalation if no response within a period.
Record of who approved what, when.
The director stops being the bottleneck for small things and stays the decision-maker for large ones, since business workflow automation Kenya with thresholds means a KES 3,000 expense is approved by rule and a KES 300,000 one reaches the director with the quote attached, and business workflow automation Kenya that routes every approval to the same person has automated the queue rather than the decision.
The project management article’s approval thresholds apply.
Notifications: SMS, WhatsApp and Email {#notifications}
Notifications are the most visible output and the easiest to overdo.
Which channel the recipient actually reads: WhatsApp for most, SMS for those without smartphones, email for business and records.
Templates with the specific information: amount, reference, date, next step.
Triggered by events: payment received, order shipped, appointment tomorrow, invoice overdue.
Frequency limits so a customer is not messaged five times for one order.
Opt-out for anything beyond transactional.
Delivery status tracked.
Cost per message managed, since SMS at volume is a line item.
A notification that says what happened and what to do next is service; one that says “update” is noise, and business workflow automation Kenya that sends “Payment of KES 2,500 received for invoice 1042, thank you” has informed the customer, while business workflow automation Kenya that sends a message at every internal status change has taught them to mute it.
Transactional messages are expected; marketing requires consent, which the data protection section addresses.
Document Generation {#documents}
Documents produced by hand from templates are produced wrong and late.
Invoices, receipts, quotations, contracts, delivery notes, statements, certificates.
Generated from the record with the current data.
Numbered sequentially.
Branded and formatted consistently.
Sent to the right person by the right channel.
Stored against the record.
Compliant with whatever the document must contain, including any electronic invoicing requirements, which should be confirmed with the revenue authority.
The template is always current because there is one, and business workflow automation Kenya that generates the invoice from the order the moment it is confirmed has removed the delay and the version problem, while business workflow automation Kenya that leaves invoicing to a person with a template file has the invoice going out on Friday with last quarter’s terms.
Eliminating Data Entry {#data-entry}
Re-entry is where time goes and errors enter.
The same information typed into a spreadsheet, a system and a message.
M-Pesa confirmations copied by hand.
Orders from WhatsApp typed into the system.
Reports assembled by copying figures.
Each re-entry is a delay and an opportunity for a wrong digit.
Automation captures once and moves the data, since business workflow automation Kenya that takes the form submission into the system, the payment callback into the ledger and the order into the invoice has removed three re-entries, and business workflow automation Kenya is measured partly by how many times a piece of information is typed after it first exists.
Count the re-entries in the process map; each is a candidate.
Connecting Systems {#integrations}
Most automation crosses system boundaries and the connections are the fragile part.
The POS, the accounting system, the CRM, the payment provider, the messaging channel, the spreadsheet.
Integration through APIs where systems offer them.
Import and export where they do not.
Mapping fields between systems.
Handling differences in identifiers, formats and timing.
Error handling when one system is unavailable.
The API integration article covers the engineering, and business workflow automation Kenya is largely integration with rules on top, since business workflow automation Kenya that moves an order from the website to the POS to the delivery system to the customer’s phone touches four systems and must survive any of them being down.
Fewer systems is easier, and a business that consolidates onto a platform with the functions built in has fewer joins to maintain.
Scheduled Tasks and Reminders {#scheduled}
Some automation runs on the calendar rather than on events.
Payment reminders before and after due dates.
Daily sales reports to the owner.
Weekly arrears summaries.
Monthly statements.
Renewal reminders for contracts, licences and subscriptions.
Stock reorder checks.
Scheduled rather than remembered, since business workflow automation Kenya that sends the rent reminder three days before the due date to every tenant every month does it without anyone remembering, and business workflow automation Kenya that produces the owner’s daily summary at close means the owner reads it rather than asks for it.
The PPPoE and collection articles show scheduled reminders changing collection rates.
Exception Handling {#exceptions}
Automations fail on the cases the rule did not anticipate, and those cases need a path.
A payment that matches nothing.
A form with impossible values.
An approver who is unavailable.
A system that is down.
A customer who replies to an automated message.
A record missing required data.
Each exception routed to a person with the context.
Not silently dropped, since an exception nobody sees is a customer nobody served.
Not silently guessed, since an automation that guesses the customer for an unmatched payment has credited the wrong person.
Logged and counted, since a rising exception rate means the rule needs work.
Exceptions are where automation meets people, and business workflow automation Kenya that routes every exception to a named person with the information needed has kept the process whole, while business workflow automation Kenya that handles the ninety-five percent and loses the five has automated the easy part and abandoned the hard part.
Human in the Loop {#human-loop}
For some steps the automation should prepare and a person should decide.
Draft prepared, person approves and sends.
Match suggested, person confirms.
Classification proposed, person accepts or changes.
Action queued, person releases.
The person does less work and keeps the decision.
Which steps stay human is a design decision, since business workflow automation Kenya that prepares a customer’s refund with the evidence and lets a person approve it has kept accountability where it belongs, and business workflow automation Kenya that refunds automatically on a rule has made a decision nobody will own when it is wrong.
Keep a person on anything irreversible, anything with money leaving, and anything about a person.
Where AI Agents Genuinely Fit {#ai-fits}
AI adds value where the input is unstructured and the output is a draft or a suggestion.
Reading a free-text message and classifying it: order, complaint, query, spam.
Extracting fields from a document: an invoice, an ID, a delivery note.
Drafting a reply for a person to review.
Summarising a long thread or a report.
Suggesting a match where rules are ambiguous.
Answering questions from a defined knowledge base, with a route to a person.
Flagging anomalies for review.
In each, the AI proposes and a rule or a person disposes, since business workflow automation Kenya that uses AI to read incoming WhatsApp orders into structured fields for a person to confirm has removed typing without removing judgement, and business workflow automation Kenya that lets the AI classify and a rule route has used it where it is strong.
The applied AI page is where these are built, and the honest framing is that AI is a component in a workflow rather than a replacement for one.
Where AI Does Not Belong {#ai-not}
AI should not make decisions with consequences that nobody will own.
Approving credit or refunds.
Rejecting applications.
Disciplining staff.
Sending anything to a customer without review where an error would matter.
Interpreting legal or regulatory requirements.
Handling a complaint end to end.
Making financial entries without verification.
Anything where being confidently wrong is worse than being slow.
AI errors are plausible-looking, which is what makes them dangerous in these places, and business workflow automation Kenya that keeps AI to drafting, classifying and extracting has used it safely, while business workflow automation Kenya that let an AI agent decide a customer’s dispute has placed an error where it will cost a relationship and possibly more.
The buyer should be suspicious of any proposal that puts AI at a decision point with money or people on the other side.
Guardrails for AI in Workflows {#guardrails}
Where AI is used, it should be constrained.
Defined scope: what it may do and what it must hand off.
Confidence thresholds: below a level, route to a person.
Human review on anything outbound where an error would matter.
Logging of what the AI proposed and what was done with it.
Testing against real cases before go-live.
Monitoring of error rates.
A way for staff to correct it and for the correction to stick.
Clear disclosure to customers where they are interacting with an automated system.
Guardrails are what make AI usable in a business, since business workflow automation Kenya with an AI that drafts replies for review and escalates anything it is unsure of is a productivity tool, and business workflow automation Kenya with an AI that answers customers directly and unmonitored is a liability waiting for its first confident mistake.
Escalation {#escalation}
Every automated path needs a way to reach a person.
Timeouts: approval not received, reply not sent.
Failures: system down, integration error.
Exceptions: unmatched, unclassifiable, out of range.
Customer request: “I want to speak to someone”.
Escalation to a named role with context.
Escalation itself monitored, since a queue nobody clears is a process that has stopped.
The person is the safety net, and business workflow automation Kenya that escalates cleanly has kept the customer served when the rule ran out, while business workflow automation Kenya that loops or drops has produced the automated failure everyone has experienced.
Test escalation as carefully as the happy path.
The Audit Trail {#audit}
Automation acts on the business’s behalf and every action should be traceable.
What triggered the workflow.
What conditions were evaluated.
What actions were taken, when, on what record.
What the AI proposed, where used.
What a person approved or changed.
What failed and what was escalated.
Searchable by record, date and workflow.
The trail is how a business answers “why did that happen”, and business workflow automation Kenya with a complete log can explain why a customer received a message or a payment was applied, while business workflow automation Kenya that acts without record leaves the business unable to explain its own behaviour.
The database and cybersecurity articles both depend on this.
Roles and Permissions {#roles}
Automation should act within the permissions of the process it replaced.
An automation that issues refunds has the refund permission and its use is logged.
An automation that sends customer messages has the messaging permission and its templates are approved.
Who can create and change workflows.
Who can approve a workflow going live.
Who can override.
Automations are users too, since business workflow automation Kenya where a workflow runs with administrator rights it does not need has an exposure, and business workflow automation Kenya where each workflow has exactly the permissions its actions require is contained.
Changing a workflow is changing how the business operates and should be controlled accordingly.
Testing Before Go-Live {#testing}
An automation that has not been tested on real cases will fail on the first one.
Test with real historical data.
Test every branch, including exceptions and failures.
Test the integrations with the actual systems.
Test notifications to real numbers.
Test escalation.
Run in parallel with the manual process for a period.
Compare outputs.
Fix, then switch.
Parallel running is the safeguard, since business workflow automation Kenya that ran beside the clerk for a month and matched her output is ready, and business workflow automation Kenya switched on directly will find its bugs in front of customers.
The project management article’s acceptance discipline applies.
Change Management and Staff {#change}
Automation changes jobs and how it is introduced determines whether staff support it.
Explain what is being automated and why.
Explain what staff will do instead.
Involve the people who do the work in mapping and testing.
Address the fear that automation means redundancy, honestly.
Train on the new process including exceptions.
Keep the person who knew the old process, since they know the exceptions.
A clerk who understands that the automation removes the typing and leaves her the customers is an ally, and business workflow automation Kenya introduced with staff involvement works because staff make it work, while business workflow automation Kenya imposed without explanation is worked around.
Measuring Whether It Worked {#measuring}
Automation should be measured against what it replaced.
Time per process before and after.
Errors before and after.
Delay from trigger to completion.
Exception rate.
Escalation rate.
Staff time redeployed.
Customer response where measurable.
Cost of running versus cost of the manual process.
Exception rate is the health measure, since business workflow automation Kenya with a falling exception rate is learning, and one with a rising rate has a rule that no longer fits, which business workflow automation Kenya should surface weekly.
If it cannot be measured, it should not have been automated.
Common Failures {#failures}
The failures recur and each is avoidable.
Automating a broken process.
Automating from the owner’s description rather than the clerk’s reality.
No exception handling.
No escalation.
AI at a decision point.
Notifications that overwhelm.
Integrations that break on the first upstream change.
No testing on real data.
No parallel run.
No measurement.
No owner after go-live.
Tool chosen before process mapped.
Each appears in the sections above, and business workflow automation Kenya that avoided them is unremarkable in operation, which is the goal, while business workflow automation Kenya that made several will be remembered as the project that made things worse.
Over-Automation {#over-automation}
More automation is not better automation.
Every workflow is something to maintain.
Every integration is something that can break.
Every notification is something a customer can mute.
Every rule is something that can be wrong.
Automating a rare task costs more than doing it.
Automating a changing process means rebuilding.
Removing every human touchpoint removes the relationship.
Restraint is a design principle, since business workflow automation Kenya that automated the ten processes that mattered and left the rest is manageable, and business workflow automation Kenya that automated everything has a business nobody fully understands.
Start with the process that hurts most and stop when the returns fall.
Brittle Integrations {#brittle}
Integrations fail when the systems they connect change, and they will change.
An API updated by the provider.
A field renamed.
A spreadsheet column moved.
A payment provider’s callback format revised.
A system replaced.
Monitoring so failures are noticed.
Error handling so a failure pauses rather than corrupts.
Documentation so the next person can fix it.
Version awareness so upstream changes are anticipated.
The API integration article’s points apply, and business workflow automation Kenya built with monitoring and graceful failure survives the upstream change, while business workflow automation Kenya that assumed the connected systems would never change will stop silently one Tuesday.
Maintenance {#maintenance}
Automation is not finished at go-live.
Rules change as the business changes.
Integrations need updating as connected systems change.
Exception patterns reveal rules to refine.
Templates need updating.
Regulatory changes need reflecting.
An owner for each workflow.
Periodic review of what is running and whether it is still needed.
The maintenance article covers software generally, and business workflow automation Kenya with a named owner reviewing exceptions monthly stays fit, while business workflow automation Kenya with no owner accumulates workflows nobody understands running on rules nobody remembers.
Data Protection and Automated Decisions {#data-protection}
Automation processes personal data and may make decisions about people, and the Data Protection Act applies to both.
Data flowing between systems is processing, and each system and each flow should have a basis.
Automated decisions affecting individuals — credit, eligibility, pricing — carry particular obligations, and the position on solely automated decisions with significant effects warrants qualified advice.
Notifications to customers require a basis: transactional is expected, marketing requires consent.
AI processing personal data is processing, with the same obligations and additional ones where it profiles.
Retention across every system the automation touches.
Access to the audit trail, which contains personal data.
Third-party AI services receiving personal data are processors, with contractual and location considerations, and the cloud article’s residency points apply.
Confirm obligations with qualified advice before automating anything about people, since business workflow automation Kenya that sends customer data to an external AI service without a basis or a contract has created an exposure, and business workflow automation Kenya that established the position first has automated within it.
Security {#security}
Automation holds credentials and acts at scale, which makes it a target.
API keys and credentials stored securely, never in scripts or messages.
Least privilege per workflow.
Webhook and callback verification so forged events do not trigger actions.
Rate limits so a runaway workflow does not send ten thousand messages.
Logging and alerting.
Review of what workflows can do.
A compromised automation acts as the business, since business workflow automation Kenya with a forged M-Pesa callback accepted without verification would release goods for payments that never happened, and business workflow automation Kenya that verifies every callback with the provider before acting has closed that.
The cybersecurity article’s points on credentials and verification apply.
Questions to Ask a Vendor {#vendor-questions}
The right questions separate a process partner from a tool seller.
Will you map the process before proposing anything?
What happens to exceptions?
How does a customer reach a person?
Where do you propose AI, and where do you refuse to?
What guardrails and review does the AI have?
How are integrations monitored and what happens when an upstream system changes?
What is the audit trail?
Who owns the workflow after go-live and what does maintenance cost?
Can we run in parallel before switching?
What data goes to which external services and on what basis?
A vendor who answers the AI question with “we keep it to drafting and classification and route decisions to people” understands the risk, and business workflow automation Kenya from a vendor who proposes AI at every step has been sold the promise, while business workflow automation Kenya from one who mapped first and proposed rules where rules suffice has been designed.
The consulting and project management articles cover vendor selection generally.
Build or Configure {#build-configure}
Automation can be configured on a platform or built as software, and the choice depends on the process.
Configure where a platform the business already uses offers workflows, approvals and notifications.
Build where the process is specific, the integrations are custom or the platform’s workflows do not fit.
Hybrid where a platform handles the common paths and custom code handles the specific ones.
Configuration is faster and constrained; building is flexible and costs more.
The custom software and SaaS articles cover the build side, and business workflow automation Kenya configured inside the POS, the property system or the billing platform the business already runs is often the fastest win, while business workflow automation Kenya built as custom integration is warranted where the process crosses systems the platforms do not connect.
Start with what the existing platforms can do.
Costs and Phasing {#costs}
Cost depends on scope and the right approach is phased.
Process mapping and design, commonly from around KES 100,000 to KES 400,000 depending on complexity.
Configuration on an existing platform, frequently modest.
Custom integration and workflow build, commonly from around KES 300,000 upward per significant process, with M-Pesa and multi-system flows at the upper end.
AI components as an additional layer with their own running cost.
Messaging costs per notification.
Maintenance as a recurring cost.
Phase it, since business workflow automation Kenya that started with the one process that hurt most — usually payment-to-receipt-to-ledger — proved the approach and paid for the next, while business workflow automation Kenya that tried to automate the whole business at once had nothing working for months.
Weigh it against the hours, since a clerk’s afternoon every day on M-Pesa re-entry is a cost the business already pays, and the automation that removes it is measured against that rather than against zero.
Frequently Asked Questions {#faqs}
What is worth automating?
Processes that are repetitive, rule-based, high-volume and error-prone — the ones staff describe as mindless. M-Pesa confirmation to receipt and ledger, invoices from orders, payment reminders, approval routing, report distribution. The test is whether the rule can be written. If the cases are all different, or the decision is about a person, keep it with a person.
Why map the process first?
Because automating a broken process produces a broken process that runs faster, and because the owner’s description of a process and the clerk’s reality differ — the clerk knows the exceptions. Map it with the people who do it, fix it, then automate it. This is the step most projects skip and the reason most fail.
What is the most valuable trigger?
The M-Pesa payment callback. It arrives with sender, amount, reference and transaction code; matched to a customer or invoice it can issue the receipt, update the ledger, activate the service and notify the customer in seconds, removing the largest single re-entry most businesses have. Verify it with the provider before anything irreversible, and hold unmatched payments for a person.
Where does AI actually help?
Where the input is unstructured and the output is a draft or a suggestion — reading a WhatsApp order into structured fields, extracting data from a document, classifying a message, drafting a reply for review, summarising, flagging anomalies. The AI proposes; a rule or a person disposes. Treat it as a component in a workflow, not a replacement for one.
Where should AI never be?
At a decision point with money or a person on the other side — approving credit or refunds, rejecting applications, disciplining staff, handling a complaint end to end, interpreting regulation, or sending anything to a customer unreviewed where an error would matter. AI errors look plausible, which is exactly what makes them dangerous there. Be suspicious of any proposal that puts AI at such a point.
What happens when the automation hits a case it wasn’t built for?
It routes the exception to a named person with the context, logged and counted. Never silently dropped — that is a customer nobody served — and never silently guessed — that is the wrong customer credited. A rising exception rate means the rule needs work, which is the health measure to watch weekly.
Will staff resist it?
Not if the truth is told: the automation removes the typing and leaves them the customers. Involve the people who do the work in mapping and testing, address the redundancy fear honestly, keep the person who knew the old process because they know the exceptions, and run in parallel before switching. Imposed without explanation, it is worked around.
How do we know it worked?
Measure against what it replaced: time per process, errors, delay from trigger to completion, exception and escalation rates, staff time redeployed, running cost against manual cost. If it cannot be measured, it should not have been automated. Business workflow automation Kenya that shows a falling exception rate and a clerk’s afternoon returned every day has done what it was bought for.