Rubansi Vincent
Analytics has a strange exemption in many budgets. A company that would never buy a delivery van without a utilization plan will commission a data platform on the strength of a demo and a feeling. Then, eighteen months later, the same company concludes that "data did not work here," which is roughly like concluding that vans do not work because nobody planned the routes.
I come to analytics from accountancy, and I appraise analytics spend the way an FP&A team appraises any capital allocation: identify the cash flows, test the assumptions, time the payback, and define in advance what success will look like on the financial statements. This article is the due-diligence checklist I run before recommending that a client build anything.
Where analytics actually pays back
Analytics creates value through exactly four mechanisms. Every credible business case names which one it is pulling and traces the arithmetic to the P&L or the balance sheet.
- Better decisions: pricing, assortment, credit, campaign allocation. Value = the margin difference between the decision made with evidence and the incumbent decision, times frequency.
- Cheaper operations: automation of reporting, reconciliation, and manual data preparation. Value = hours reclaimed times loaded cost, which is usually the easiest number in the case to defend.
- Risk reduced: fewer stockouts, earlier fraud detection, covenant breaches foreseen. Value = expected loss avoided, stated with honest probabilities.
- Working capital released: sharper forecasts shrinking inventory and receivables. Value = cash freed times its carrying cost. This one belongs on the balance sheet, and it is chronically underclaimed.
If a proposed dashboard, model, or pipeline cannot be attached to one of these four, it is not an investment; it is decoration with a refresh schedule.
"An analytics proposal that cannot name its payback mechanism is not asking for investment. It is asking for patronage."
Rubansi Vincent
The due-diligence checklist
1. Start from the decision inventory
Before any tooling conversation, list the recurring decisions the business makes: their frequency, their financial magnitude, who makes them, and what information they currently run on. The product of frequency and magnitude, discounted by how bad the current information is, ranks your opportunities. In nearly every engagement, the top of that ranking is not the glamorous machine-learning idea; it is a weekly decision worth real money that currently runs on a stale export.
2. Price the full cost, not the license
The subscription fee is the visible fraction. A truthful total cost of ownership includes implementation and data engineering, the ongoing cost of data quality (validation, reconciliation, stewardship), training and adoption effort, and the analyst time to maintain the semantic model as the business changes. As a working heuristic, first-year TCO runs two to four times the license line. Writing that down early is cheaper than discovering it in month nine.
3. Demand a payback schedule, in periods, not adjectives
"Transformational" is not a payback period. A fundable analytics case looks like a loan amortization: value begins at zero, ramps as adoption ramps, and crosses cumulative cost at a stated month. My portfolio rule of thumb for small and mid-sized businesses: automation and reporting work should pay back inside six months; decision-support work inside twelve; anything longer needs either a contractual anchor or a smaller first phase.
4. Sequence build-versus-buy by differentiation
Buy what does not differentiate you (ingestion, storage, visualization; the market has solved these). Build what encodes your business logic: the metric definitions, the KPI tree, the semantic model, the reconciliation to your ledger. That layer is cheap to build with SQL and DAX, expensive to outsource badly, and it is where institutional knowledge compounds. Teams that invert this (custom infrastructure, off-the-shelf understanding) buy themselves the maintenance burden without the insight.
5. Pilot at the smallest decisive scale
The unit of purchase should be one decision instrumented end to end: one recurring question, one reconciled model, one owner, one quarter. A pilot small enough to fail cheaply and decisive enough to prove the mechanism beats a platform rollout on every dimension that matters, including politics. Scale is a reward the first decision earns.
6. Define the success metrics before the kickoff
Adoption is necessary but not sufficient; the case closes on financial evidence. Fit the metric to the mechanism you claimed:
| Payback mechanism | Proof metric |
|---|---|
| Better decisions | Decision log: variance between forecast and outcome, before vs after |
| Cheaper operations | Hours per reporting cycle, headcount-equivalent reclaimed |
| Risk reduced | Incident frequency and severity against the prior baseline |
| Working capital | Inventory days, DSO, cash conversion cycle trendline |
Agree the baseline measurement before work begins. A benefit without a baseline is a story, and auditors do not sign stories.
The costs of not investing
Due diligence cuts both ways, and the status quo should be appraised as ruthlessly as the proposal. A business running on gut feel and month-old spreadsheets is already paying: in the spread between good and mediocre pricing decisions, in analyst hours consumed by manual reconciliation, in inventory bought against anecdote, in the discount every lender and acquirer applies to numbers that cannot be traced. These costs are real, recurring, and compounding; they simply lack an invoice. Part of the analyst's job is to give them one.
"The most expensive analytics program in most companies is the one they think they are not running: decisions made daily, at full financial magnitude, on unexamined data."
Rubansi Vincent
What this looks like in practice
The engagements that clear these hurdles share a shape. They start with a decision inventory workshop, not a tool selection. They spend the first weeks on unglamorous foundations: a reconciled data model, a metric dictionary, boundary validation. They ship one decision instrument to one forum, log its use for a quarter, and present the payback evidence in the language of the P&L. Then, and only then, they scale to the next branch of the KPI tree.
None of this requires enterprise budgets. It requires the same financial discipline you already apply to vans, stock, and hires, pointed at information for once.
Make the shift to data-centric decision-making
If you are weighing analytics spend this year, do not start with a platform shortlist. Start with the four payback mechanisms, your decision inventory, and a pilot sized to prove the mechanism in one quarter. Fund analytics like an investor, and it will repay you like an asset.
This appraisal is precisely what I offer founders and small teams: a CPA-trained data analyst who writes the business case, builds the pilot with SQL, Python, and Power BI, and reports the payback in numbers your accountant will accept. If you want an honest answer to "what would analytics actually be worth to us," send me your hardest recurring decision and I will come back with an approach, a timeline, and a flat quote. The firms that win the next decade will be the ones that stopped treating evidence as optional; the buy-in starts with one well-appraised decision.