The spreadsheet, the duplicate book and the WhatsApp group are not the problem. They are the accurate record of how your business actually works. The problem is that none of them carries a clock, an owner or an identifier — so nothing downstream can be measured, joined or automated.
What you are actually replacing
"Digital transformation" is a term that has cost SMEs a great deal of money by being vague. Concretely, there are only three things you are moving:
- Transaction records — what happened. An order, a delivery, an issue from stores, a payment. Today these live in duplicate books, Excel and the accounting package, and they disagree.
- Status — where something has reached. Today this lives in a WhatsApp thread and in someone's head. It is the single most expensive gap, because every "where are we with X?" question is an interruption.
- Documents — quotes, LPOs, delivery notes, invoices, certificates. Today these are files named final_v3(2) across several desktops.
Everything in this roadmap is about moving those three things — and nothing else — into a form that has a timestamp, an owner and a shared identifier.
Stage 0 — Measure before you spend (one week)
Do not skip this. It is free, it takes a week, and it is the only thing standing between you and a system nobody uses.
- Pick one flow. Order-to-cash or purchase-to-pay. Not the whole business.
- Trace ten real transactions backwards from the bank statement to the original enquiry. List every artefact and the system it lives in.
- Count the handoffs — every re-keying, printing, copying or verbal relay. Healthy flows have 2–4. Most SMEs find 9–14.
- Time the gaps, not the tasks. The work takes minutes; the waiting takes days. Record elapsed hours between artefacts.
- Write the number on one page and show it to the leadership team. The number is the mandate.
Stage 1 — One identifier, everywhere (weeks 1–2)
Before any software: agree a single order number, generated once, written on every artefact including the printed job card and the WhatsApp message. No new system required.
This is the highest-return change available to an SME and it costs nothing. It converts a pile of unrelated documents into a chain that can be followed — by a person today, by software later. Firms that do only this typically recover one to two hours of reconciliation per day.
Stage 2 — Capture at source (weeks 3–6)
The event is recorded once, digitally, by the person who caused it — not by an office clerk two days later from a pile of paper.
Start with your highest-volume paper form. In most SMEs that is the delivery note, the goods-received note or the job card. Replace it with a phone form that writes to a shared table. Requirements, in order of importance:
- It must work on the phone the person already owns, in poor network, with the data captured offline and synced later.
- It must be faster than the paper it replaces. If it is slower, it will be filled in at the end of the week from memory, and you have bought a worse record.
- It must print or send the same document the customer already expects. Do not ask a customer to change their process to suit your transformation.
Keep the paper running in parallel for two weeks. Retire it only when the digital record is demonstrably complete.
Stage 3 — Move status out of WhatsApp (weeks 5–8)
WhatsApp is an excellent interface and a terrible database. The mistake is trying to remove it; staff and customers will not comply. The fix is to change what it is for.
- Keep WhatsApp as the notification channel. It is where people actually look.
- Stop treating the thread as the record. Every status change gets written to the shared table — by a form, a bot, or a person clicking a button in the message.
- Make the system push, so nobody has to pull. When a job moves stage, the customer and the internal group get a message automatically. This alone removes most "any update?" traffic.
Stage 4 — Excel becomes a report, not a database (weeks 8–12)
Do not attempt to remove the spreadsheet. Change its role.
Today the workbook is where data is authored, stored and analysed at once, which is why there are nine versions. The target is: data is authored in a form, stored in one table, and the workbook reads from that table. Analysts keep every skill they have. They simply stop being the storage layer.
Migrate authorship one field at a time, starting with the fields that cause the most conflict — status, owner, next action, price.
Stage 5 — Automate the deterministic steps, then add AI (from week 12)
By now you have timestamped events joined by an identifier. Automate the parts that need no judgement at all: status notifications, document generation, reminder ladders, reconciliation matching, threshold alerts, scheduled reports. These are rules, not models. They cannot hallucinate, they are auditable, and they deliver most of the early gain.
Only what is left after that — reading unstructured documents, drafting, classifying, summarising — is the legitimate AI surface. And by now you have the data to evaluate whether it worked.
What to buy, and what not to
- Do not buy an ERP yet. An ERP encodes a process. If your process is not written down, you will pay a consultant to invent one, and your staff will work around it. ERPs fail in SMEs for this reason far more often than for technical ones.
- Do buy, or build, a shared operational table with forms in front of it. Budget band: KES 250,000–900,000 for a first flow, depending on integration.
- Do pay for data migration and training — typically 30–40% of the project, and the first line cut when budgets tighten. Cutting it is what turns a working system into shelfware.
- Do not pay per-seat for staff who only need to receive information. Send it to them; do not license them.
The 90-day plan
- Days 1–7: Trace ten transactions. Publish the handoff count and elapsed times.
- Days 8–14: Agree and mandate a single order identifier on every artefact.
- Days 15–45: Digitise the highest-volume capture point. Run parallel with paper.
- Days 46–60: Push status notifications automatically; stop using the thread as the record.
- Days 61–80: Point one management report at the new table. Retire its manual predecessor.
- Days 81–90: Re-measure the same ten steps. Report the change. That number funds the next stage.
The failure modes, honestly
- Digitising the whole business at once. The organisation has a fixed appetite for change. Spend it on one flow that visibly works.
- Building for the exception. Design for the 85% of transactions that are ordinary. Handle the rest by hand and route them to a queue.
- No named owner. A system without someone accountable for its data quality degrades within a quarter.
- Skipping to AI. An AI layer over a fragmented process automates the guessing, then presents it confidently. That is worse than no output, because it is now trusted.
The point
Transformation in an SME is not a platform purchase. It is the disciplined act of giving your existing process a clock, an identifier and a single place to live. Do that and the automation — and later the AI — becomes almost mechanical. Skip it and no amount of software will help, because there is nothing for the software to read.
