Five key reasons agentic AI fails before it starts

Business value beyond the boardroom

More than two-thirds of agentic AI pilots will fail, and many are doomed from the start. Max Rice examines why.

Gartner estimates that more than 40% of agentic AI projects will be cancelled by the end of 2027, due to “escalating costs, unclear business value or inadequate risk controls.” Many of these initiatives were doomed before they left the drawing-board.

Those that succeed will have moved with purpose; clarity of vision leading to greater efficiency and an ability to forecast value and prove viability in a data-driven way. Without an effective machine for ideation and solutioning, being able to deliver effectively becomes irrelevant.

In this article, I will address five prominent failures in the ideation and solutioning phases of agentic AI projects and how your organisation can avoid the same early pitfalls.

#1: Putting the solution before the problem

Inspired by the AI trend, some implementations have proceeded regardless of a specific business need or without care to ensure the solution being delivered is efficient, effective and appropriate.

The risks of letting these projects run are notable:

These risks can be avoided by:

  1. Creating open environments in which ideas flow organically inwards, by building trust across departments and educating on what agentic is and is not;
  2. Partnering with colleagues to assess internal and external pain points; identifying root-causes and defining the areas where leveraging technology can have the greatest impact;
  3. Building appropriate governance mechanisms for swift triage of potential candidates for investment, with messaging around rejected ideas handled with care.

#2: Failure to uncover the numbers

Finding relevant datapoints can be complex and time-consuming.

Understandably, IT leaders rarely have the micro-level expertise of individual processes operating outside their technical domain. Taking a data-led approach typically requires stakeholder inputs, and the ease of obtaining such information will often depend upon strength of relationships, and how persuasive and audience-relevant a value-case is.

Similarly, identifying the right metrics can be difficult: number of events, people involved, financial cost of time, drop-out rates, CSAT scores, etc. – there are countless options. Distinguishing between anecdote and fact can be uncomfortable, but using estimates like ‘it takes roughly an hour’ can kill ROI.

Navigate this by:

  1. Assessing the availability of existing datasets, taking care to ensure relevance, accuracy, completeness and significance;
  2. Enforce and support colleagues in data-led approaches – assess current data availability and build missing datasets;
  3. Leverage the time and enthusiasm of those closest to the associated benefits, to build the value-case.

#3: Confusing scope and scale

It’s easy to mistake broader scope for greater scale, i.e. failing to apply a refined solution to a larger dataset. If the numbers on a proposed pilot only stack up when additional functionality is added, then the pilot isn’t addressing the right points.

Steps to success:

  1. Ensure pilot implementations add value in their own right, while providing the necessary confidence results can be repeated across similar scenarios;
  2. When calculating scale, ensure the scope of functionality remains broadly the same;
  3. Avoid excessive scope as part of the pilot, to ensure costs are contained while viability and forecast accuracy is assessed. Crucially, this will also give you the opportunity to embed governance infrastructure before tackling more scope.

#4: Ignoring scenario viability during assessment

Digital agents are increasingly competent at handling complexity, but some scenarios – particularly high-risk environments such as healthcare – still require human intervention. When calculating value, if not removed, these scenarios can lead to inflation of projected value.

Failing to undertake scenario-based assessment up front also leads to an incomplete view when assessing agentic as a solution against alternatives. Accurate calculation of one-off and ongoing costs also becomes impossible.

Avoid this by:

  1. Connecting your team’s Solution Architects to business stakeholders to whittle down these points with a technical lens;
  2. Assessing non-agentic solution approaches as alternatives. In many cases this can lead to a similar outcome at a lower cost;
  3. Factoring in one-off and transactional costs across the implementation and ongoing use. Note that testing and optimisation, including in Production, are likely to be a significantly larger than usual cost and in some cases an ongoing investment;
  4. Critically, being sure to assess your datapoints and data quality early on, along with any remedial work necessary;
  5. Reviewing the end-to-end process to ensure appropriate hand-offs, including between systems – ensure viability of connecting with legacy systems early on and otherwise consider how data can be shared.

#5: Communicate the value proposition with impact, relevance and quantitative grounding

Repeating the same message to every audience using the same language and supporting datapoints can be counter-productive – many will hear ‘change’ and hence ‘pain’. This can negatively impact investment and user adoption.

Build confidence and interest in your initiatives by:

  1. Identifying key metrics that can become value KPIs – estimate these up-front based on real data and measure against them later to determine success;
  2. Tying your messaging to your organisation’s strategic goals;
  3. Creating a well-indexed baseline pack of information that will be of interest to anyone looking to understand the detail, and customise a top sheet introduction that draws out audience-relevant datapoints or features;
  4. Create a succinct description of the projected value-add to capture attention.

IT leadership should set the tone as change-drivers. Their teams have a crucial role to play in ensuring rigorous triage and supporting teams in obtaining the right information. This doesn’t necessarily mean building the value-case themselves, but rather being clear in expectations and providing guidance to those needing it.

For some organisations, agentic will be a tickbox exercise – "We did it, and it wasn’t for us." For others - those who invest in up-front planning and value definition (along with appropriate delivery models and governance) - agentic will mark a shift in how their businesses operate, leading to greater margins and higher customer satisfaction.

Max Rice is a 10x Salesforce Certified Programme Delivery Lead, dedicated to bridging the gap between commercial and technology strategies. With experience driving value at global and regional scales in a range of industries – including manufacturing, software, energy, non-profit and education – Max helps organisations turn complex technology investments into measurable business outcomes.

Views herein are the author’s and may not represent those of his employer.