Should we build our customer data platform in-house or buy one?
A sample decision room with illustrative data for a fictional company, not a client engagement. A $40M-revenue B2B services firm needs a customer data platform to unify records across three acquired business units. Engineering has proposed building it; a vendor has quoted for the same scope. The CFO must recommend one path to the board.
- Decision owner
- Chief Financial Officer
- Deadline
- Board meeting, 12 weeks out
- At stake
- ≈$1.9M of three-year spend, plus 9 months of engineering capacity
- Alternatives
- 3 considered
Decision brief
Buy the vendor platform. Over three years it costs $126.3K less than building, on the assumptions recorded below.
Build — 3-year total
$921.3K
Buy — 3-year total
$795.0K
Buying saves
$126.3K
The assumption this turns on
Annual maintenance is assumed at 22% of the build cost. The two paths cost the same at 14.4%. Below that, building is cheaper. That figure is an industry range rather than this team’s measured history — which makes it the first number to replace with evidence.
Change the assumption. See whether the decision changes.
This is the model a customer receives — not a screenshot of one. Move the slider and every figure on this page recalculates.
Break-even at 14.4%
At 22% maintenance, buying is cheaper over three years by $126.3K.
Three cases, not one forecast
A single number hides the disagreement. These three make it explicit — and they do not all point the same way.
Quoted plan
Engineering's estimate and the vendor's signed quote, taken at face value.
Buy cheaper by $126.3K
Build runs long
The build takes 45% longer than estimated and carries a heavier maintenance load — the most common failure mode for in-house platform work.
Buy cheaper by $620.6K
Lean build
A smaller team ships a narrower first release on time. The optimistic case engineering is arguing for.
Build cheaper by $311.3K
Assumption ledger
Every number carries where it came from, who owns it, when it was set, and how far we trust it. An observed fact and someone’s estimate are not the same kind of number, and a model that renders them identically cannot be challenged properly.
| Assumption | Value | Type | Source | Owner | As of | Confidence |
|---|---|---|---|---|---|---|
| Engineers on the build Named individuals are already allocated in the capacity plan. | 4 FTE | Customer assumption | Engineering capacity plan, v3 | VP Engineering | 2026-07-14 | high |
| Loaded cost per engineer | $185,000 | Observed fact | 2026 payroll actuals plus 28% benefits and overhead load | Finance | 2026-07-02 | high |
| Months to first production release Raise confidence by benchmarking against the last two platform builds and their original estimates. | 9 months | Customer assumption | Engineering estimate; not validated against a comparable delivered project | VP Engineering | 2026-07-14 | low |
| Annual maintenance, as % of build cost This is the assumption the recommendation turns on. Replace with the team's own measured run cost on an existing internal service. | 22% | External estimate | Vendor-neutral industry range of 15–25%; midpoint plus 2pts for multi-unit data complexity | GradeCircle analyst | 2026-07-20 | medium |
| Vendor implementation (one-time) | $75,000 | Observed fact | Signed vendor quote, valid 90 days | Procurement | 2026-07-09 | high |
| Vendor licence (annual) Pricing beyond year 3 is not fixed and is excluded from this model. | $240,000 | Observed fact | Signed vendor quote, 3-year term, flat pricing | Procurement | 2026-07-09 | high |
Which assumption actually moves the answer
Monte Carlo, 500 samples, perturbing every input within its plausible range and correlating each against the three-year cost difference. Figures shift slightly between runs — that is sampling, and we would rather show it than hide it.
Running analysis…
Risks on the record
- The 9-month build estimate has not been validated against a comparable delivered project.high
- Maintenance share is an industry range, not this team's measured history.high
- The vendor quote is valid for 90 days; renewal terms after year 3 are not fixed.medium
- Four engineers on this build are four engineers not on the product roadmap.medium
What this model excludes
- Taxes, cost of capital, and discounting — figures are undiscounted pre-tax cash
- Strategic option value of owning the codebase
- Switching cost if the vendor is replaced in year 4+
- Revenue upside from either path — this compares cost to deliver the same scope
Verification status
This model sits at Sample: the structure is real and the arithmetic runs live, but the figures are illustrative and no analyst has signed them off against a real company’s data. A customer engagement moves up this ladder — source-verified inputs, analyst review against a published checklist, then a named approver. Outcome-measured comes later, when actuals are compared against what the model predicted.
That was our example. What's your decision?
Yours will not look like this one — different options, different criteria, different numbers. Describe it in a sentence and we will show you the inputs we would need, the model we would build, and the timeline, before you commit to anything.
Take this further
- Run this with your own numbers — the free build-vs-buy calculator: 3-year TCO for both paths and the crossover point, in about two minutes.
- Check your customer data first — the decision above turns on how fragmented the data already is, and that is measurable before anyone builds or buys.
- Have us model it properly — criteria weighted openly, vendors scored, the choice defensible to a board.
Prefer to explore first? Try the market-entry model or browse all free tools.