Share

AI-Powered CapEx Process

Written by: Richard Frykberg

An AI CapEx boom is underway as platform providers build out the infrastructure and organizations seek out applications for this new technology. Fortunately, with the right training, AI can also help you make these capital expenditure decisions.

Smarter CapEx choices will help you to achieve your strategic goals sooner, at lowest cost and least risk. AI can provide you with invaluable assistance in the CapEx process by enhancing the quality of business cases, facilitating more reliable estimates and forecasts, conducting more rigorous risk assessments, and providing rich portfolio insights and control.

For your AI-powered CapEx assistant to be effective, it needs to be well trained. AI has a voracious appetite for data, and the more reliable data it ingests, the more valuable the results it provides. The critical data you should provide for your AI model is identified in the sections below.

The Current CapEx Process versus an AI-Powered CapEx Process

Quality data is essential for leveraging AI to improve CapEx decision-making. The infographic below illustrates how the traditional CapEx process is influenced by human input, which can lead to biased outcomes. In contrast, an AI-powered CapEx process uses machine learning to analyze both ideas and historical results, generating weighted, objective insights that lead to more informed decisions.

Current CapEx Process versus an AI-Powered CapEx Process

Strategic Context for AI

Generative AI models are becoming increasingly sophisticated. Rather than needing to prompt multiple times to elicit a useful response, these models are increasingly adopting internal train-of-thought evaluations to solve more complex problems.

The most critical dimension to any high-value, long-term, capital investment is strategic alignment. A good project for one organization may be a waste or misapplication of resources in another if it is not strategically aligned. Clearly identifying the targeted strategic objectives of any project will help your AI-assistant learn what worked, and what didn’t.

The more formalized the definition of corporate and departmental strategies, and the clearer the relationship between these strategies and your project portfolio, the greater the learning and the more effective your AI-assistant’s recommendations.

Risk Assessments for AI

All businesses operate in a risky environment, under the expectation that higher returns will compensate for this exposure to risk. Every future investment must consider this trade-off between risk and return.

Humans, however, are inherently biased by our personalities. We all perceive risk, and assess risk, constrained by our own limited experience and motivations. When prompted to consider worst, best, and most likely outcomes we are all likely to provide varying answers depending on our optimistic or pessimistic inclinations.

AI-assistants are not subjective. With sufficient empirical evidence they are much more likely to provide realistic assessments of project outcomes. Knowing how humans ‘like you’ typically estimate ‘projects like these,’ and how the results vary ‘in reality,’ AI-assistants can provide valuable insight into real implementation risks.

These risks relate to the uncertainty in both project investment and return. Investment risk relates to the unpredictability of costs and schedules. These inherent risks are often clearly identified by comparing previous projects planned and actual results. Successfulness of returns are often not as clearly recorded. Were the expected cost savings realized? Was the incremental revenue achieved? Did we cut our emissions and improve safety outcomes?

For an AI assistant to provide effective investment risk assessments, it is important to keep a reliable record of both project investment and return outcomes, in a format conducive to machine learning.

Capital Budgeting with Spreadsheets ebook cover

Get your eBook

The Spreadsheet Trap

An eBook that reveals the hidden costs of spreadsheets and what you can do about it.

Planning Data for AI

The frame of reference for evaluating likely outcomes is the project plan. Historically, this has been prepared by humans and reflects the planned investment and anticipated returns. For your AI-assistant to predict the likely outcome of your next project, it needs to know how previous projects ended up.

The more granular the planning and actual data, the better your AI-assistant can help identify planning errors. The planning data should identify planning assumptions and parameters. These are the key values, like quantities and rates, which are fed into the cost model to produce your cost estimates.

The planning data should also include estimated activity durations. Variances in both cost and timing can have a significant impact on both actual investment and the timing and value of future returns.

Many legacy project planning systems do not systemize this data. It becomes locked away in standalone project plans and spreadsheets. For planning data to be a useful input to machine learning, it needs to be centralized, validated, and standardized.

Business Case Scoring and Ranking for AI

Project plans are typically incorporated into business case submissions for budget allocation and approval. All organizations are constrained by available time, resources, and capital. Therefore, to facilitate decision making, it is common to apply a scoring and ranking methodology to candidate projects. This helps to select the most urgent and valuable projects.

Scoring mechanisms usually encompass both qualitative and quantitative assessments. Qualitative scoring may include, for example, quality, safety, or goodwill evaluations. Quantitative scores are typically financial such as payback period, Net Present Value (NPV) or Internal Rate of Return (IRR). Other non-financial quantitative metrics such as CO2 emissions, energy consumption, or headcount impacts may also be considered.

The project evaluation scores are usually very important features in machine learning. Actual project outcomes can be evaluated with reference to these scores to help predict future project outcomes, as well as to help improve the scoring methodology.

Too often, project business cases are retained in stand-alone documents. This gravely impacts the accessibility of standardized information to help train your AI assistant. To achieve better AI-powered outcomes, project business cases should be fully digitized into a centralized repository, scoring and ranking metrics standardized, and combined with actual project outcomes to facilitate effective machine learning.

Laptop with demo of Capital Project Scoring and Ranking functionality

4 Minute Demo

How to Score and Rank Projects

See how Stratex Online can identify, assess, and prioritize your projects.

Portfolio Selection and Substitution

In practice, an effective CapEx management process focuses not just on individual business case submissions, but on the entire capital project portfolio. In order to optimize returns within resource and risk constraints, it is necessary to select a balanced and appropriately diversified project portfolio.

Active strategic portfolio management involves meaningful classification of candidate projects. This classification will typically include investment reason (e.g. replacement, growth, or environmental), asset type (e.g. land, fixed equipment, or intangibles), business area, scale and risk.

Considering these classifications, and based on linear programming, or Monte-Carlo simulations, an optimally balanced projects portfolio can be selected within time, resource, and risk-exposure constraints.

Project portfolio selection and budgeting are normally conducted for a fiscal year based on known information at the time. During the year, as circumstances change, project substitutions occur as more urgent important initiatives are identified.

However, the effectiveness of project portfolio selections is seldom validated. We can often clearly identity the projects we wish we hadn’t undertaken. But are we sure we’re achieving an optimal mix?

Keeping track of project portfolio selections, substitutions and achieved outcomes is essential for machine learning. Based on this evidence, AI assistants can help provide useful analysis. By simulating alternative portfolio selections and statistically likely outcomes of each, capital planning AI can assess the effectiveness of your historical project selections, and help you make better portfolio selections in the future.

Capital Expenditure and Operational Expenditure Data for AI

Tracking actual expenditure alongside planned monthly activity expenditure provides a crucial data set for training of AI models.

Does our split of CapEx and OpEx align with the plan? Was expenditure in line with our work breakdown structure and estimation? Was the expected work allocation between external and external resources realistic?

For many organizations, planning is performed at a detailed level and then costs are tracked at a summary level in the financial system. In other organizations, budgets are allocated at a project level, and actuals are tracked at detailed cost element level (e.g. external services) but provide no useful break-down by cost component (e.g. walls or roof).

However, to leverage the predictive power of AI assistants it is imperative to consolidate planning and actual expenditure information at the same level of detail. Typically, this requires integration with procurement, financial and resource management systems to combine useful training data into a single CapEx Planning and Budgeting solution.

Scheduling Data for AI

The frustration of many executives is not just the inaccuracy of capital project expenditure estimates, but also the unpredictability of timing. Whilst arguably made worse in recent times because of the pandemic, war, and technology upheavals, accurately estimating project delivery schedules has never been easy.

To train your AI model to estimate project delivery durations more accurately, it is essential to record original activity duration estimates as well as actual activity timelines. Empowered with historical evidence of similar projects, AI assistants can provide more objective project schedules.

Forecast Data for AI

From a purely economic perspective, once initiated, all project investment is a sunk cost. What is critical is an ongoing assessment of likely future investment requirements and return expectations.

In a dynamic world, unexpected factors can derail both. Technical complexities can cause investment costs to inflate and take longer than expected. Changing market dynamics can invalidate cost saving or revenue projections.

Hence, it is vitally important to continuously forecast likely project outcomes. When the fundamentals no longer stack up, projects should be stopped. When alternative projects offer a higher return, project substitutions should be triggered.

Unfortunately, humans are very biased forecasters. Sandbagging, the shark-fin or snow-plough effect reflect this wishful thinking. Unwilling to release allocated budgets, project managers will often simply shift the forecast into later months, desperately wishing that they can recover lost time.

The evidence doesn’t lie. With access to similar projects, historical cost projections, and actual historical expenditure profiles, a project that is over-budget and overdue at half-way, is more than likely to be over-budget and delayed in the end. Your AI assistance can provide invaluable predictive forecasting with the right data set.

In addition to actual project expenditure, the monthly ‘human’ forecasts help establish the ‘typical’ pattern. By identifying the relevant features of project cost and schedule over-runs, future variances can be more reliably forecast, and more effective mitigating action taken sooner.

Financial Analysis Data for AI

Predictive analytic AI machine learning models are data intensive. The key figures that should be included in the monitoring and control of capital project include plan, budget, actual, forecast, and commitments.

This data should be accumulated at the most granular level available, and ideally by project work-breakdown structure element.

Many project execution solutions handle the detailed scheduling, allocation and status of project delivery. However, they do not typically incorporate commitment, actual costs, and real cashflows.

In order to have access to complete, reliable, and standardized analytical data, disparate data sources, and especially spreadsheets and stand-alone project plans, should be replaced with an integrated solution that handles the capital expenditure management process from ideation to successful project delivery.

Benefit Realization Data for AI

Legacy CapEx management solutions do not typically include benefit realization analysis. This is the whole point of any capital investment project: to generate future benefits. These may be in relation to replacement of productive assets, savings or growth initiatives.

Tracking the outcome of any capital project is essential to training any AI model. This is the ultimate measure of project success and the key objective of training any AI assistant.

By identifying the relevant features of successful projects in the past will help you identify and choose, with AI assistance, the projects that will matter most in the future.

External Data for AI

Not all data available to make better project decisions is available in-house. A key advantage of AI-powered systems is that they’re able to be trained on a broader set of data.

CapEx management solution that leverages cumulative experience on a large number of projects is able to perform better than models trained exclusively on your own internal experience.

In addition, with reference to the entire internet, AI models are able to provide objective insight into the validity and reasonableness of cost and duration estimates. Anomalous data can be readily identified and corrected before executives make inappropriate decisions on misleading information.

Enable your AI-Powered CapEx Process with CapEx Software

Critical capital expenditure decisions will always have humans in the loop. However, AI is becoming an indispensable assistant in the identification, evaluation, prioritization, and control of capital projects. To be able to apply the power of machine learning to your CapEx process, you need to collect and curate high quality training data.

For many organizations, the CapEx process is fractured into various disconnected data repositories, spreadsheets, and manual processes. Digital transformation of the CapEx process involves consistent data aggregation, standardization, process automation, integration, and validation. This means connecting up the entire capital expenditure process including demand management, business case evaluation, strategic portfolio management, project forecasting and financial analysis.

Whilst you could build your own solution utilizing generic database, workflow, and user-interaction tools, Stratex Online provides end-to-end digital transformation of the CapEx process as a ready-to-run SaaS solution. Furthermore, Stratex Online provides an AI-assistant that leverages your data and project experience to help you make smarter project selections and achieve your strategic goals sooner, and at lowest cost and least risk.