Methodologies / Dodone — AI Engineering ROI v1

Dodone — AI Engineering ROI v1

DODONE-MOD-001 · Dodone Model v1 · Version 1Engine 0.1.3Not independently audited

A transparent, reproducible scenario calculator for the incremental economic effect of introducing an AI coding tool to an engineering workforce over one 12-month period, compared with the same organisation without the tool.

How Dodone works — five-step guide

Download the controlled methodology PDF (Version 1, 26 pages, SHA-256 2bc18cca7deef2b6e278429c61486d2fba02c236401af3e7436e059c0da5b85d).

Question, counterfactual and scope

For a given company and one 12-month period: what is the incremental annual economic value of deploying an AI coding tool to an identified engineering workforce, compared with the same workforce and operations without it?

The business-as-usual case excludes the assessed tool but includes expected ordinary growth, existing output, margins and compensation.
It applies to a new tool against no tool. Replacing or expanding an existing AI tool needs a separately defined incremental baseline.
Annual-average FTE, adoption and costs are used; ramp-up, multi-year effects and NPV are out of scope.

Two alternative valuations — never added

Capacity-based and revenue-linked values are two alternative valuations of the same output gain. Never add them.
Neither valuation is a guaranteed minimum or maximum; which one is higher depends on the inputs.
Capacity value is the cost-equivalent of redeployed engineer time, not a payroll or cash saving.
Revenue-linked value is a hypothetical scenario based on stated assumptions, not a causal sales forecast.

Formulas

All percentages are stored as decimals. Productivity is net of prompting, review, testing, rework, security and integration, so no separate quality penalty is applied.

Workforce and cost

N=\sum_i n_i,\qquad W=\sum_i n_i w_i,\qquad \bar w=\frac{W}{N}\;(N>0)

Engineers are annual-average FTE in location × seniority cohorts; w is the fully loaded annual cost. Existing payroll is never an AI cost.

EQ-3.1 · methodology §3.1

w_i = w_i^{\text{original}} \times \text{FX}_i

Converted at a stored, dated rate before aggregation; the original salary and currency are kept.

EQ-3.4 · methodology §3.4

C_L = 12\,L\,p_L,\qquad C_T = 12\sum_i n_i a_i T_i,\qquad C = C_L + C_T + I

Licences, extra usage per active engineer and optional implementation spend. I = 0 means excluded, not free.

EQ-4.1 – 4.3 · methodology §4

Productivity

r_i = f_i\,s_i,\qquad g_i = \frac{r_i}{1-r_i}

Mode A: eligible work f × net time reduction s gives released time r; g is the equivalent output uplift. A 20% task saving is not a 20% team uplift.

EQ-5.1 · methodology §5.1

r_i = \frac{g_i}{1+g_i}

Mode B: a measured end-to-end uplift g. It is never multiplied by the eligible fraction again.

EQ-5.2 · methodology §5.1

r_i^{+}=\max(r_i,0),\quad r_i^{-}=\min(r_i,0)\quad(\text{likewise } g)

Gains and losses are separate channels so a positive-only discount can never hide an adverse outcome.

EQ-6.1 · methodology §6

Capacity-based value

B_C = \sum_i n_i\,w_i\,a_i\,\bigl(u_i\,r_i^{+} + r_i^{-}\bigr)

Cost-equivalent of released time actually redeployed (u) by active engineers (a). Losses apply in full. Not a cash saving.

EQ-7.1 – 7.2 · methodology §7

Revenue-linked contribution value

R = R_0\,(1+b)

Business-as-usual revenue for the analysis year, without the AI tool.

EQ-8.1 · methodology §8.1

Q_{+} = \frac{1}{N}\sum_i n_i a_i u_i g_i^{+},\qquad Q_{-} = \frac{1}{N}\sum_i n_i a_i g_i^{-}

Realised output change across the whole workforce.

EQ-8.4 · methodology §8.4

\Delta R = R\,e\,\bigl[Q_{+}(1-d) + Q_{-}\,t_{-}\bigr],\qquad B_R = \Delta R \times m

e: share of revenue assumed to respond to engineering output; d: downstream loss on gains; t−: downside transmission; m: one incremental contribution margin.

EQ-8.5 · methodology §8.5

Returns

V = B - C,\qquad \text{ROI} = \frac{B - C}{C}\;\;(C>0)

Computed separately for each valuation. With zero cost, ROI is not defined; benefits and net value are still shown.

EQ-9.1 – 9.3 · methodology §9

B_R > B_C \iff R > \frac{W\,r}{m\,e\,g\,(1-d)}

For uniform positive productivity: which valuation is higher depends on the inputs, so neither is a fixed bound.

EQ-11.2 · methodology §11.2

Defaults — illustrative, not benchmarks

The calculator opens with a fictional UK B2B SaaS company: £50m prior-year revenue, 0% BAU growth, 100 engineering FTE (20% junior, 50% mid, 30% senior), 70% licensed, 60% effectively active, £25 per seat per month, £15 extra usage per active engineer per month, 50% eligible work, 20% net time reduction, 50% redeployment, 100% downside transmission and implementation cost excluded. Replace them with your own figures.

Industry presets
Industry	Incremental margin proxy	Commercial response e	Diffusion loss d	Eligible work f
B2B Software / SaaS	75%	65%	25%	50%
IT Consulting / Professional Services	35%	40%	45%	50%
Financial Services	40%	30%	55%	40%
Manufacturing / Industrial	30%	20%	65%	35%
Retail / E-commerce	25%	15%	65%	40%
Other / Custom	40%	30%	50%	40%

Illustrative fully loaded annual costs
Location	Junior	Mid-level	Senior
United Kingdom	£52,000.00	£78,000.00	£117,000.00
United States	US$124,000.00	US$182,000.00	US$247,000.00
Poland	PLN 120,000.00	PLN 225,000.00	PLN 325,000.00
India	₹875,000.00	₹2,250,000.00	₹3,750,000.00

Salaries are artificial starting values for product testing, not compensation survey statistics. Other locations are entered manually. Dodone does not supply exchange rates: you enter a dated rate for each foreign currency.

Reference worked example

The default company, calculated live by the same engine the calculator uses (methodology chapter 11):

Reference example results
Weighted loaded cost	£84,500.00
Total AI cost C	£31,800.00
Released time per active engineer r	10.0000%
Output uplift per active engineer g	11.1111%
Realised workforce output uplift	3.3333%
Capacity benefit B_C	£253,500.00
Capacity economic ROI	697.17%
Incremental revenue scenario ΔR	£812,500.00
Commercial contribution benefit B_R	£609,375.00
Commercial scenario ROI	1,816.27%

Sensitivity

Varying only the net time saved on eligible work by ±10 points. These are sensitivity tests, not confidence bounds or percentiles.

Sensitivity of the reference example
Scenario	Capacity benefit	Capacity ROI	Commercial benefit	Commercial ROI
Conservative	£126,750.00	298.58%	£288,651.32	807.71%
Base	£253,500.00	697.17%	£609,375.00	1,816.27%
Optimistic	£380,250.00	1,095.75%	£967,830.88	2,943.49%

With all other defaults fixed, the two valuations are equal at about £20,800,000.00 of BAU revenue. Below that, the revenue-linked value is lower than the capacity value.

Key limitations

L01. Historic revenue does not show how many sales extra engineering output would unlock; the commercial response e is a scenario assumption.
L02. Industry values for e, d, m and f are illustrative and not empirically benchmarked.
L03. Both valuations use the same output gain, so they must never be added.
L04. Time saved is not necessarily avoided payroll or new cash flow.
L07. Revenue response is treated as locally linear; large modelled changes may overstate reality.
L08. One year only: lagged revenue and recurring-revenue persistence are not modelled.
L10. With implementation cost left at 0, rollout, platform and governance costs are excluded.
L16. Only valid for a new tool against no tool, not for replacing or expanding an existing AI tool.
L19. A cheap subscription can produce a very large ROI; read the monetary values and sensitivity, not ROI alone.

Even a fully reproduced result is not a validated causal estimate. Finance-grade use needs firm-specific measured productivity, actual vendor spend, validated incremental margins and, ideally, observed revenue evidence.

Versions and status

2026-10-31 — Methodology DODONE-MOD-001 Version 1 — controlled draft for independent review
2026-10-31 — Engine 0.1.1 — reproduces every chapter 11 reference value; Mode B sensitivity ±5 points of released time
2026-11-02 — Engine 0.1.2 — adds the “rounded from reported figures” input label for curated examples; no change to any calculation
2026-11-01 — Methodology Version 1 approved for publication; official PDF released. Version label corrected (was “1.1” in two places); no change to any equation. Engine 0.1.3 records the new document checksum only

The methodology is a controlled draft prepared for independent methodological audit. No audit opinion has been issued. Source document SHA-256 12FE50158B0E92F5…

Open the calculator