From Gut Feel to Numbers: A 90-Day Plan for SMEs

Somewhere in the past year you have probably said a version of "we need proper numbers." Then the quarter got busy, a vendor quoted something enormous, and the idea got put off again. That is normal, and waiting has cost you less than a bad platform purchase would have. Gut judgement carried the business this far, and the plan below is built to back it with numbers: ninety days of work in three phases, ending with one number you trust.

In short: An SME can get from gut feel to reliable numbers in about 90 days, using the systems it already owns. Days 1 to 30: map where your numbers live and pick the one decision you keep making on gut (a jinq audit, from SGD 4,000 over two weeks, covers most of this month). Days 31 to 60: connect those systems and put that first number on a live view. Days 61 to 90: automate the manual jobs that cost the most hours. If you want it run for you, a Fractional AI Officer does this embedded in your team, from SGD 7,500 a month.

Why 90 days?

Ninety days is long enough to produce a number you actually trust, and short enough that the work stays a project with an end date. A common way data projects in small companies stall is by growing: what started as "see our sales properly" turns into a platform evaluation that never ships. A fixed 90 days forces the useful discipline of picking one decision that matters and building only what answers it.

We covered what this should cost, and why the big quotes are scoped for someone else, in the previous post. This one is the plan itself.

Days 1 to 30: where do your numbers live today?

No building yet. The first month is a clear map of what you have:

• List the systems. Accounting, POS or invoicing, CRM or customer list, stock records, and whatever lives in files someone keeps by hand. For each: what it holds, and who touches it.
• Pick the one decision. The question you keep answering on gut, for example which products make money, which customers are slipping, or whether you can afford the next hire. One is enough.
• Time the current path. How long it takes today to answer that question, and how old the number is when it arrives.

This is exactly what an audit engagement does: two weeks, remote, from SGD 4,000, and you get back a ranked map of what to connect and what each step would cost. You can also do this month yourself with the lists above; the point is that month one produces a scope, and the scope keeps month two small.

Days 31 to 60: how do you get one number you can trust?

The second month joins the systems that hold the pieces of your chosen answer. This is an integration layer between the tools you already pay for, and for a plan this size it usually touches two or three systems, no more. We walked through a concrete version of this for retailers in One Source of Truth for Retail Data; the shape is the same in any industry.

The deliverable is deliberately narrow: one live view that answers the month-one question, updating on its own, for example margin by product for the current week with nothing exported by hand. When the number on that screen matches reality for two straight weeks, you have a figure everyone trusts.

Keep it at one view for now. Every tile you add is scope; bring the extra questions in after day 90.

Days 61 to 90: which manual jobs get automated first?

With the data connected, the third month automates the manual jobs that cost your team the most hours each week. The usual candidates:

• The report someone assembles by hand.
• The reconciliation between two systems.
• The copy-paste that moves orders from one tool to another.

These run on top of the connections you built in month two, which is why they come third. The payback usually shows up here first: one Singapore tuition chain we worked with ran payroll close by hand across spreadsheets; after rostering and payroll were connected and automated, the monthly close went from 3 days to 4 hours. A clinic group that stopped re-keying records across three systems got back around 1.5 staff-days every week.

What does the plan look like on one page?

Phase	The work	What you have at the end
Days 1 to 30	Map your systems, pick the one decision, time today's path (audit from SGD 4,000 covers this)	A scope: what to connect, in what order, at what cost
Days 31 to 60	Connect the two or three systems that hold the answer	One live, trusted number, updating itself
Days 61 to 90	Automate the top two or three manual jobs	Hours back every week, and a base to build on

Where do 90-day plans go wrong?

Three places, all avoidable:

• The scope grows into a platform project. Someone proposes doing it "properly" and the two-system integration becomes a warehouse project. Keep the scope to the one decision you picked in month one.
• Nobody owns it after week two. Everyone in an SME already has a full-time job, so the data work stalls the first busy week. This is the strongest argument for having someone embedded a day or two a week whose job it is; that is the Fractional AI Officer role.
• Charts nobody acts on. Ending the 90 days with charts instead of a decision is common. The month-one discipline of picking one decision exists to prevent it.

What does the 90 days cost?

The audit that scopes it is from SGD 4,000 over two weeks. The build in months two and three is priced from that scope, so you see the number before committing to it. If you want the whole plan run for you, a Fractional AI Officer is from SGD 7,500 a month, one to two days a week, and stays past day 90 to keep it used. The previous post sets these figures against what an enterprise build costs, if you want the full comparison.

What to do next

You can start days 1 to 5 this week, without spending anything:

• Write the list of systems your numbers live in.
• Name the one decision you keep making on gut.
• Time how long that answer takes to produce today.

If the list is longer than you expected, that is normal, and it is what month one of the plan is for.