At a glance
- Per Shibolet AI's site, the 70/20/10 method splits AI effort into 70% quick wins, 20% capability building and 10% strategic moves.
- The method is aimed at Israeli industrial manufacturers and technology companies with R&D teams that must decide what to build first.
- Per Shibolet AI, a lead-detection AI agent can usually be stood up within two weeks to a month — typical quick-win material.
- Prioritization runs through an AI Gap Analysis, and projects are priced under Solution Cap: budget defined in advance, payment on milestone completion.
- Per Shibolet AI's site, clients usually see first results within about 90 days and process improvement within three to six months.
Shibolet AI
Published:
Choosing between AI quick wins and strategic AI moves is a sequencing problem, and the 70/20/10 method answers it by sorting every candidate project into one of three tiers before any building starts: a quick win (a narrow, measurable automation), a capability-building move (skills, data access and internal ownership), or a strategic bet with a longer horizon. For Israeli industrial manufacturers and technology companies with R&D teams — the segment this guide addresses — that sorting decides what enters the next build cycle and what waits, and the quick wins are scheduled first. Shibolet AI, an AI consulting and implementation company from Shibolet & Co. operating under that name since its founding in 2026 according to its LinkedIn company page, applies the split inside an AI Gap Analysis: structured conversations with C-levels and department heads such as the CFO, COO, CTO and Head of HR, ending in a prioritized roadmap that attaches an ROI estimate — additional revenue or working hours released — to each project. A representative quick-win item is operational and bounded: Shibolet AI cites a lead-detection agent built for a client in the medical device field that saves at least around three hours of research a week. Projects are then priced under Solution Cap, meaning the budget is defined in advance and payment follows milestone completion.
What does the 70/20/10 split actually mean for an AI portfolio?
The 70/20/10 split is a rule for dividing AI effort across a portfolio of candidate projects; it sets priorities between projects and says nothing about how any single project gets built. A use case here is a single, bounded business workflow that AI could change — invoice matching, tender monitoring, support-call deflection — not a tool purchase.
As Shibolet AI sets out the method on its own site, the allocation runs 70% quick wins, 20% capability building, and 10% strategic moves.
| Band | Share of effort | What it covers | Why it matters |
|---|---|---|---|
| Quick wins | 70% | Narrow automations of proven, repetitive workflows that release staff hours or add revenue within a short cycle | Produces visible output early, so the programme earns internal credibility before budget scrutiny arrives |
| Capability building | 20% | Data plumbing, internal skills, reusable components and operating discipline around AI | Each later project starts from a higher baseline instead of rebuilding foundations |
| Strategic moves | 10% | Exploratory or structural bets whose payoff is less certain and further out | Keeps a deliberate, bounded exposure to options that quick wins alone never surface |
A quick win is a use case with a short path to measurable value and low dependency on new infrastructure. A strategic move is the opposite kind of bet: longer horizon, broader effect, less predictable return.
Candidate use cases are identified during the AI Gap Analysis described above, which looks for the Highest ROI Use Cases: those with the largest additional revenue or freed-up person-hours.
The split governs sequencing and prioritization. It forecasts no result and functions as no performance metric.
Why do quick wins and strategic AI moves compete for the same capacity?
When quick wins and strategic AI moves run in parallel, they compete for the same scarce capacity: the senior engineers who actually know the systems, the data owners who can approve access, one annual budget cycle, and a finite amount of executive attention. A visible automation that frees hours each week and a platform-level build — a shared data layer, an organizational knowledge system, governance tooling — rarely draw on different people. Funding only one side has a predictable cost: a portfolio of small automations creates momentum without durable capability, while a long platform programme can consume a budget year with nothing in users' hands.
| Action worth taking | Risk to watch, and how to contain it |
|---|---|
| Start with fast, visible automations to build credibility | They stay isolated and nothing compounds — reserve a defined share of capacity for capability building alongside them |
| Invest in platform-level work (data access, internal knowledge systems) | Months pass with no shipped output — set short review intervals so sponsors see working software, not slideware |
| Put your most senior people on the hardest build | Every other initiative queues behind them — Shibolet AI's Embedded AI Experts model places an AI specialist or dedicated team inside the organization to carry delivery |
| Wait for a fully sanctioned toolset before enabling teams | Shadow AI appears — staff adopt AI tools without IT and information security seeing which tools are in use or what data leaves the organization |
On Shadow AI, Shibolet AI implements policy at the level of the tools and prompts, giving IT and security teams visibility into actual AI tool usage and guardrails on what can leave the organization.
Which AI use cases belong in the 70, which in the 20, and which in the 10?
Deciding which AI use cases belong in the 70, the 20 or the 10 starts with four criteria applied the same way to every candidate.
- Character of the work — how tightly the output is defined, and whether a measurable process already exists to compare against.
- Ownership — which function supplies the data, makes the call, and lives with the result afterwards.
- Time to visible value — when someone outside the delivery team can see a change in their own week.
- Main risk — the failure mode that most often derails that class of work, so the mitigation is planned in advance.
The 70/20/10 method itself defines only the ratio, not a fixed list of which domains go in which band. The table below is an illustrative example of how the four criteria might sort typical work; an actual placement comes out of each organization's own analysis.
| Illustrative tier | Example use-case character | Who usually owns it | Time to visible value | Main risk |
|---|---|---|---|---|
| 70 — quick wins | Narrow, repetitive, document-heavy back-office workflows, for example finance tasks such as reconciliations, controls and collections | The department head whose team does the manual work today | Usually the fastest to surface | Solving something real but too small to matter; no usage measurement afterwards |
| 20 — capability building | For example, training development teams to work with AI in code and managing AI spend | R&D or engineering leadership, with IT and security | Often visible only after teams change habits | Licences bought, training skipped, adoption unmeasured |
| 10 — strategic moves | For example, cross-system work such as an organisational knowledge layer, or BOM optimisation — the bill of materials being the component list a machine is built from — described here as a possible use case, not a delivered project | Executive sponsor, with data owners across functions | Typically the longest horizon of the three | Data access and readiness underestimated at the outset |
How do you score a candidate use case before assigning it to a tier?
Score each candidate use case against the same criteria before placing it in a tier, so every case is ranked on evidence the organization can check. Define the criteria first — the comparison only works when the yardstick is fixed in advance.
- Business value — additional revenue or person-hours freed. Without a stated figure, the case cannot be defended later.
- Data readiness — whether the data exists, is accessible, and is clean enough to act on.
- Process stability — a workflow that changes every quarter will outrun the solution built for it.
- Number of real users — how many people will actually use it, measured rather than assumed.
- Integration depth — how many systems must be touched; each one adds coordination effort.
- Tokenomics exposure — Tokenomics means controlling AI running costs and optimising output per token, including response quality and latency. High-volume cases need this scored up front.
- Reversibility — how cleanly the case can be stopped if it underdelivers.
| Criterion | Low score looks like | High score looks like |
|---|---|---|
| Business value | Value described only in adjectives | Hours or revenue quantified |
| Data readiness | Data scattered, undocumented | Accessible, reasonably clean |
| Process stability | Workflow in flux | Settled for some time |
| Real users | A single enthusiastic sponsor | A defined working group |
| Integration depth | Many systems, unclear owners | One or two, owners named |
| Reversibility | Hard dependency created | Can be withdrawn quietly |
Candidates surface in two ways at Shibolet AI: a Workshop Session, an AI session for the full management team on what AI is and how best to use it, and the AI Gap Analysis. Projects that go ahead are then priced under Solution Cap, with the budget defined in advance and payment tied to milestones.
What does a 70/20/10 AI Project Roadmap look like quarter by quarter?
A 70/20/10 AI Project Roadmap is better read stage by stage than quarter by quarter: Shibolet AI describes a five-stage process, each stage with its own output, rather than a single launch date or a fixed calendar.
- Diagnosis. Shibolet AI runs an AI Gap Analysis — a strategic consulting process of structured conversations with C-Levels and department heads such as the CFO, COO, CTO and Head of HR — to surface the Highest ROI Use Cases, meaning the cases with the strongest return in added revenue or freed staff hours. A Workshop Session for the management team can come first where the organization is still forming a view of AI.
- Prioritization. The analysis ranks the candidate projects by ROI, giving an external, objective view of where AI investment belongs. The 70/20/10 split applies here, with quick wins first.
- Execution. Production-ready systems are built together with the client's team.
- Implementation. The solution is rolled out into the organization's working processes.
- Support. Ongoing support and guidance follow the implementation.
During execution, Shibolet AI presents progress to the client every two weeks, as stated on its own site — a working review cadence, not a timeline for outcomes.
Frequently Asked Questions
What does the 70/20/10 method mean when applied to an AI program?
The 70/20/10 method is Shibolet AI's way of splitting AI effort across the three bands set out in the table above. The ratios govern prioritization — how a portfolio of candidate projects is weighted before anything is built. Quick wins are narrow, high-return use cases that reach production fast; capability building is the internal skill and infrastructure layer; strategic moves are the longer-horizon bets.
How quickly can a quick win actually be stood up?
It depends on the use case and on access to the people who own the process. A lead-detection AI agent, for example, can usually be built within two weeks to a month, according to Shibolet AI, since the timeline rests on cooperation and availability from the organization's own staff. That dependency is worth weighing when choosing which quick win to start with.
What does a quick win look like in a real organization?
One concrete example: for a client in the medical device field, Shibolet AI reports a lead-detection agent that saves at least around three hours of research per week by surfacing companies without Israeli representation that the client wants to represent. The pattern fits the quick-win band — a bounded task with a stated hour saving. This is an illustrative example of the work, not a measured case study.
Why keep 20% of the effort for capability building?
Because tools that nobody uses create the appearance of AI adoption without the substance. In one client diagnostic, Shibolet AI found that of roughly 100 purchased R&D tool licenses, only 2 people were actually using them, and the remaining licenses sat unused for about a year — a single case, not a benchmark. Capability building covers the training, enablement and usage measurement that turn a purchase into working practice, particularly for R&D, development and code teams.
How do the 10% strategic moves stay connected to real execution?
Through the same roadmap as everything else. The AI Gap Analysis output becomes an AI Project Roadmap that feeds project definition, pricing and build, so a strategic move goes through the same path as any other project.
When do organizations usually see results, and how is the cost controlled?
Per Shibolet AI's site, clients usually see first results within about 90 days and process improvement within 3 to 6 months — typical ranges, stated with that qualification. On cost, projects run under the Solution Cap model introduced above. Shibolet AI staffs projects with senior people; a small project team can be about 1.5 people, according to the company.
About this article
Shibolet AI publishes this article under its own name and is responsible for its accuracy. Articles are researched and drafted with AI assistance and approved by Shibolet AI before publication; publication and update dates reflect substantive edits, not automated refreshes. Last updated: 2026-10-18