Overview
Most small technology companies make growth decisions on reported numbers. A campaign looks like it worked, so it gets more budget. LIFTS asks a harder question first: what did the decision actually cause, compared with what would have happened anyway?
The framework turns that question into a repeatable cycle with five stages. Each stage has a working template, so a team moves from "why isn't this product growing?" to a documented, evidence-based decision. LIFTS is the operational name and structure of the author's AI-Enabled Commercialization Decision Methodology.
Who it is for. Small and emerging U.S. technology companies making high-stakes growth decisions without dedicated growth, analytics, or experimentation teams. The level of rigor scales to the data a company actually has.
The cycle
Each stage produces one completed template, and each decision starts the next cycle.
Locate the constraint
Find the one barrier holding the product back, based on evidence.
ProducesConstraint Diagnostic
Investigate and design the test
Turn the constraint into a testable question, with decision rules set in advance.
ProducesExperiment Protocol
Fit the intervention
Choose the response, and use AI only if it beats a non-AI option.
ProducesUse-Case Assessment
Test incremental impact
Measure what the response caused, net of side effects.
ProducesImpact Assessment
Scale or stop
Act on the result using the rules agreed before the test.
ProducesDecision Matrix
Core rules
Five rules apply to every cycle. They make results comparable across cycles, products, and teams.
- R1
One primary constraint per cycle
A cycle that addresses several constraints at once cannot attribute its result to any one of them.
- R2
Decision rules before data
Success thresholds and the action for each possible result are recorded before the test begins, and are not changed once results arrive.
- R3
AI must earn its place
Choose an AI option only when there is good reason to expect it to beat a named non-AI option on results, cost, effort, risk, and human oversight. Then measure whether it did.
- R4
Realized impact, not gross
Results are reported net of offsetting effects elsewhere in the business.
- R5
Every decision has an owner and a review date
A decision without both is recorded as open, not made.
Stage gates
A stage is not complete until its gate is met, and the next stage does not begin until then.
| Stage | Entry condition | Required output | Exit gate |
|---|---|---|---|
| L · Locate | A product with a stated growth goal | Constraint Diagnostic naming one primary constraint, with evidence | Decision owner agrees the primary constraint |
| I · Investigate | An agreed primary constraint | Experiment Protocol (hypothesis, metrics, baseline, design, decision rules), with the Decision Matrix filled in | Hypothesis, measurement plan, thresholds, and provisional decision rules approved before testing begins |
| F · Fit | A signed-off test plan | Use-Case Assessment comparing the AI option with a non-AI alternative, including risks and compliance review | Intervention chosen and implemented as specified |
| T · Test | Intervention live for the planned period | Impact Assessment with gross and realized impact, against the pre-set threshold and evidence tier | Result classified against the decision rules |
| S · Scale or stop | A classified result | Decision recorded in the Decision Matrix with an owner and a review date | Scale, iterate, or stop; the decision feeds the next cycle |
Evidence tiers
Every result is labeled with the tier it rests on. The tier sets how much weight the final decision can carry.
| Tier | Design | When to use | Confidence |
|---|---|---|---|
| Tier 1 | Randomized holdout: users, accounts, or markets randomly assigned to test and control | Default whenever the channel or platform allows random assignment | High, if the test is well designed, large enough, and run as planned. Can support scaling when the effect size, uncertainty, economic value, and rollout risks justify it. |
| Tier 2 | Geo split or synthetic control: matched or modeled comparison markets | When randomization is not possible, such as market-wide campaigns, partnerships, or offline channels | Moderate. Can support scaling when comparison markets tracked closely before the test; confirm with a second test where possible. |
| Tier 3 | Structured before-and-after: comparison to a pre-set baseline, with seasonality controlled | Low-cost, reversible decisions, or as a screen before a stronger test | Low confidence that the change caused the result. Enough to keep investigating or make small, reversible calls; not enough to scale. |
The toolkit
Five instruments, one per stage. Each has a template to fill in and an illustrative example that follows one hypothetical company, an appointment-scheduling startup for independent clinics, through the full cycle.
Commercialization Constraint Diagnostic
What is holding this product back, and which barrier matters most?
| Section | What to capture | Your response |
|---|---|---|
| Product and stage | Product, target customer, market, stage (pre-launch, launch, scaling) | Fill in |
| Growth goal | The outcome the company needs, with a number and date | Fill in |
| Current performance | Key metrics today: acquisition, conversion, retention, cost per customer | Fill in |
| Candidate barriers | Unclear product-market fit; low trust or understanding; inefficient acquisition spend; missing measurement; compliance friction; manual workflow on a small team; other | Fill in |
| Evidence for each barrier | Data, customer feedback, or past tests that support or rule out each candidate | Fill in |
| Primary constraint | The single barrier to address first, and why it outranks the others | Fill in |
| Cost of inaction | What continuing without addressing it costs, in money, time, or customers | Fill in |
| Unknowns | What the team does not yet know, to carry into Stage 2 | Fill in |
- Rank by evidence, not instinct. A barrier with no supporting data becomes an unknown, not the primary constraint.
- For an early-stage company, the most common primary constraints are unclear product-market fit and low trust.
| Product and stage | Appointment-scheduling software for independent clinics; U.S.; launched, early growth |
|---|---|
| Growth goal | Double monthly paying clinics within two quarters without raising acquisition cost |
| Current performance | Steady free-trial sign-ups across acquisition channels, but most trials end without a paid plan |
| Evidence | Exit surveys and sales calls show trial users worry about moving patient records; acquisition costs are stable across channels |
| Primary constraint | Low trust: clinics are not confident moving their patient records to a new provider |
| Cost of inaction | Continued spend on trials that rarely convert, and a growth goal that slips |
| Unknowns | Whether reassurance changes conversion, or whether price is the real barrier |
Learning Agenda and Experiment Protocol
What must be proven, how will it be tested, and what will each result trigger?
| Section | What to capture | Your response |
|---|---|---|
| Constraint addressed | The primary constraint from Stage 1 | Fill in |
| Hypothesis | "If we [change], then [outcome] will [move by how much], because [reason]." | Fill in |
| Primary metrics | The one to three metrics that decide the result | Fill in |
| Secondary metrics | Metrics that explain why the result happened or flag side effects | Fill in |
| Baseline | Current values of the primary metrics and the period they cover | Fill in |
| Test design | Holdout, market split, synthetic control, or structured before-and-after, matched to available data | Fill in |
| Measurement limitations | Known data gaps, delays, or biases, and how each will be handled | Fill in |
| Success threshold | The result that counts as success, stated as a number | Fill in |
| Execution rules | What stays fixed during the test (pricing, budgets, channel mix, messaging), dates, approvals | Fill in |
| Decision for each outcome | Positive, neutral, and negative ranges, and the action each triggers | Fill in |
- Match rigor to data. A company with limited volume can use a structured before-and-after comparison, as long as its limitations are written down.
- Keep other conditions as steady as you can, and write down anything that changes. If something cannot be controlled, account for it in the design or analysis.
| Hypothesis | If trial users are offered guided data migration, trial-to-paid conversion will rise by at least 20% (relative), because it removes the main switching risk |
|---|---|
| Primary metrics | Trial-to-paid conversion; cost per paying clinic |
| Test design | New trial sign-ups randomly split 50/50 between the migration offer and the standard trial (Tier 1) |
| Measurement limitations | Weekly volumes are small, so the test runs until each group has enough trials to detect a 20% lift |
| Success threshold | A relative lift in trial-to-paid conversion of 20% or more |
| Execution rules | Pricing, channel mix, and onboarding held fixed; the founder approves the launch |
| Decision for each outcome | 20% or more: roll out to all trials. 5% to 20%: keep it and test a lighter version. Under 5%: revisit price as the constraint in a new cycle |
AI-Enabled Use-Case Assessment
Does an existing AI tool outperform a non-AI alternative for this specific problem?
| Section | What to capture | Your response |
|---|---|---|
| Constraint addressed | The primary constraint from Stage 1 | Fill in |
| Response options | The AI option and at least one non-AI option (research, process change, analytics, repositioning) | Fill in |
| Users and workflow | Who uses the tool, at which step, and how often | Fill in |
| Expected benefit | The measurable improvement expected (time, cost, speed, accuracy), with a number | Fill in |
| Comparison | Why the AI option beats the non-AI option, or why it does not | Fill in |
| Data and access | The data the tool needs, where it lives, and who must approve access | Fill in |
| Human oversight | Which decisions stay with people, and who reviews the tool's output | Fill in |
| Risks and limitations | Accuracy, data quality, privacy, compliance, cost | Fill in |
| Success criteria | How the benefit will be measured after adoption (carried into Stage 4) | Fill in |
| Recommendation | Adopt, pilot, or decline | Fill in |
- Start with tools the company already has. LIFTS applies existing AI tools; it does not require building models.
- In regulated industries, keep compliance review in the workflow; use AI to speed that review up, not to replace it.
| Constraint addressed | Low trust; offering migration help to every trial user is too slow for a three-person team |
|---|---|
| Response options | AI option: an assistant built on an existing AI tool that maps a clinic's exported records to the new format and drafts a migration checklist. Non-AI option: the founder runs each migration manually on a call |
| Users and workflow | Customer success lead, within 24 hours of a trial starting |
| Expected benefit | Migration prep cut from about two hours to 20 minutes per clinic, so every trial user can be offered help |
| Human oversight | Staff review every mapping before any data moves; no patient data leaves approved tools |
| Outcome | In a two-week pilot, the AI option met the time target; the manual option could not keep up with trial volume |
| Recommendation | Adopt |
Incremental Impact Assessment
Did the response cause a real improvement, or would the result have happened anyway?
| Section | What to capture | Your response |
|---|---|---|
| Test reference | The Stage 2 protocol and the actual test dates | Fill in |
| Baseline vs result | Primary metrics before and after, with the period for each | Fill in |
| Method used | Holdout, market split, synthetic control, or before-and-after, and any deviation from the plan | Fill in |
| Incremental effect | The change attributable to the response, net of what would have happened anyway | Fill in |
| Attributed vs incremental | Where reported or attributed results differ from the incremental estimate, and by how much | Fill in |
| Economic impact | Gross and realized effect on spend, revenue, profit, or time, after side effects | Fill in |
| Side effects | Changes in other channels, teams, or metrics | Fill in |
| Confidence and caveats | Statistical significance where available, data gaps, and what the test could not show | Fill in |
| Verdict | Positive, neutral, or negative against the pre-set success threshold | Fill in |
- State what the test could not show. An honest limit is more useful than an overstated result.
| Method used | Randomized holdout (Tier 1) |
|---|---|
| Baseline vs result | Trial-to-paid conversion of 8% in the control group vs 10% with the migration offer |
| Incremental effect | A 25% relative lift in conversion caused by the offer |
| Economic impact | Cost per paying clinic down about 20% gross; about 15% realized after staff review time |
| Side effects | Support tickets rose slightly in the first month as new clinics onboarded faster |
| Confidence and caveats | Significant at the planned sample size; the test did not show whether the effect lasts beyond the first billing cycle |
| Verdict | Positive: above the 20% threshold |
Commercialization Decision Matrix
Given the evidence, what should happen next?
| Result range | What it means | Action | Owner | Review date |
|---|---|---|---|---|
| Positive (above threshold) | Fill in | Scale, expand to new markets or teams, or keep the change | Name | Date |
| Neutral (between thresholds) | Fill in | Adjust and retest, or keep partially | Name | Date |
| Negative (below threshold) | Fill in | Stop and revert, or return to Stage 1 | Name | Date |
| Inconclusive (data too weak) | Fill in | Extend the test or redesign it | Name | Date |
- Also record the decision taken, any deviation from the matrix and why, how the change will be monitored, and the next question for a new cycle.
- The same process can produce opposite decisions in different situations. That is the point: the evidence decides, not the plan.
Decision 1: migration offer. The result was positive, so the team rolled the offer out to all trial users, with a monthly review of conversion and support load owned by the customer success lead.
Decision 2: channel investment, the next cycle. The team tested whether its largest acquisition channel was driving incremental growth. Pausing it in half of its markets lost paying clinics that had cost about $90 each to acquire, well below their first-year value. Rather than cut, the team set an investment matrix in advance:
| Marginal cost per new paying clinic | Action |
|---|---|
| Over $300 | Reduce channel investment by 30% |
| $150 to $300 | Adjust investment by 15% to find the next point on the curve |
| Under $150 | Increase channel investment by 30% |
Defined terms
These terms mean the same thing everywhere in the toolkit.
- Commercialization constraint
- The barrier that most limits a product's growth or adoption at a given time, such as unclear fit, low trust, inefficient acquisition spend, or weak conversion.
- Primary constraint
- The single constraint a LIFTS cycle addresses, chosen in Stage 1 on the strength of evidence.
- Counterfactual
- What would have happened without the change. It can be estimated through an experiment or a careful comparison, and it is only as reliable as the design and data behind it.
- Incrementality
- The share of an observed outcome caused by the intervention rather than by factors that would have produced it anyway.
- Lift
- The measured difference between the test group and the counterfactual, as an absolute or percentage change.
- Decision rule
- A pre-agreed statement of what the team will do for each possible result, recorded before the test starts.
- Realized impact
- Gross impact net of offsetting effects elsewhere, such as volume or cost shifted to another channel.
- AI-enabled intervention
- A change that uses an existing AI tool for analysis, targeting, content, or automation, evaluated against a non-AI alternative.
Scope and limitations
Intended use
Commercialization decisions in small and emerging technology companies, including go-to-market spend, acquisition channels, conversion, retention, and the adoption of existing AI tools in those workflows.
Outside scope
Building or training AI models; brand or creative strategy that cannot be measured against a counterfactual; legal, regulatory, or financial advice.
Minimum conditions
A measurable outcome metric, enough volume to detect the expected effect, and a named decision owner. Where volume is too low for a Tier 1 or Tier 2 design, the toolkit still applies and the result is labeled Tier 3.
Limitations
Results are specific to the product, market, and period tested, and should be re-validated before they are applied elsewhere. The toolkit structures judgment; it does not replace it.
About and sources
Efearue Uduigwomen is a growth and commercialization practitioner with a master's degree in marketing analytics. LIFTS formalizes decision processes she developed and applied through her work at Risevest, Remitly, and Inkitt between 2020 and 2026. The integrated framework was formally documented in October 2026, and the examples on this page use a hypothetical company.
Foundations
LIFTS draws on established improvement, experimentation, and growth methods and combines selected principles into one decision framework built for small and emerging technology companies.
- Theory of Constraints (Goldratt, The Goal, 1984): the focus on one primary constraint before any response.
- DMAIC (Six Sigma, Motorola, 1980s): the five-stage structure and control step, rebuilt around causal testing.
- PDCA (Deming, building on Shewhart): the repeating loop, with "check" replaced by an incrementality test.
- Build-Measure-Learn (Ries, The Lean Startup, 2011): measuring what was caused rather than only what happened.
- G.R.O.W.S. (Ellis and Brown, Hacking Growth, 2017): structured experimentation, started from a diagnosed constraint.
References
- Abadie, A., Diamond, A., & Hainmueller, J. (2010). Synthetic control methods for comparative case studies. Journal of the American Statistical Association, 105(490), 493–505.
- Deming, W. E. (1986). Out of the Crisis. MIT Press.
- Ellis, S., & Brown, M. (2017). Hacking Growth. Crown Business.
- Goldratt, E. M., & Cox, J. (1984). The Goal: A Process of Ongoing Improvement. North River Press.
- Kohavi, R., Tang, D., & Xu, Y. (2020). Trustworthy Online Controlled Experiments. Cambridge University Press.
- Pyzdek, T., & Keller, P. A. (2014). The Six Sigma Handbook (4th ed.). McGraw-Hill Education.
- Ries, E. (2011). The Lean Startup. Crown Business.
- Shewhart, W. A. (1939). Statistical Method from the Viewpoint of Quality Control. Graduate School of the U.S. Department of Agriculture.