Agentic AI Won’t Close Your Bids

Roughly 40% of medicines in Europe are now bought through public procurement procedures. In hospital settings, 22 of 30 countries rely on bidding. When those contracts are awarded, 62% turn primarily on price, against only 24% that use most-economically-advantageous-contract criteria, and in classes such as antineoplastic agents the share of price-only awards reaches 84%.

For a widening share of the portfolio, the contract is not a channel. It is the commercial moment. Several years of revenue, and most of the margin attached to it, are settled by a number a team enters under deadline, often without systematic reference to what comparable contracts actually cleared at.

That is the moment the current wave of AI investment has to answer for. Most of it is answering a different question.

The life-sciences AI market has entered its agent-count era. Platform launches arrive with agent inventories attached. Adoption is reported as the number of agents deployed and the number of use cases covered. The counts climb every quarter, and they climb fast.

The underlying shift is real. Agentic AI has moved out of the laboratory and into enterprise workflows, and the infrastructure layer and the domain layer are converging faster than most commercial functions can absorb. The risk is narrower than the hype and more expensive: that commercial leaders read agent availability as commercial execution.

The market is already learning where that confusion leads. Gartner expects more than 40% of agentic AI projects to be canceled by the end of 2027, citing escalating costs, unclear business value and inadequate risk controls, and estimates that only around 130 of the thousands of vendors marketing agentic AI are genuinely building it. It calls the gap “agent washing”: rebranding assistants, robotic process automation and chatbots without substantial agentic capability. In May 2026 it applied the term to supply-chain planning technology and added a blunter warning, that any vendor promising end-to-end autonomy before 2027 is overstating what is possible.

Look at where commercial agents are actually being pointed and the gap becomes concrete. The commercial agent portfolios now on the market cluster around market insight, field-force targeting, commercial planning, literature review and data cataloguing: brand strategy, market intelligence, HCP engagement, data governance. That is useful work, and it sits adjacent to the contract. None of it is a contract-response engine. This is not a criticism of any platform. It is an observation about how much of the commercial P&L currently sits outside the places agents are being aimed.

An agent can retrieve data. It can summarize a contract pack, generate a draft and route tasks to other agents. None of that, on its own, answers the questions that decide whether a bid creates profitable growth:

• • Should we pursue this opportunity?
  • Which approved product meets each line-item requirement?
  • Can we supply it for the full contract term?
  • What is the right net price after rebates, indexation, logistics and service obligations?
  • Which claims are supported by current, approved evidence?
  • Which exceptions require commercial, legal, regulatory or quality approval?
  • What should the outcome teach us before the next bid?

Until AI can move those decisions through a governed workflow and into the systems where commercial work actually happens, it is assisting activity rather than executing commerce.

A bid is a chain of commercial commitments

A contract response is not one task. It is a linked set of commitments connecting buyer need, product eligibility, regulatory evidence, supply capacity, service obligations, price and contractual risk.

One weak link can make the entire submission non-compliant, uncompetitive or unprofitable, and the failure rarely announces itself at submission. A pack size mis-mapped in column D becomes an unfulfillable delivery obligation in year two. An expired claim buried in an annexe becomes a compliance finding. A price entered without an indexation mechanism becomes four years of quiet erosion.

Given how much of the European portfolio now moves through procurement, that exposure has outgrown the bid desk. It is a P&L exposure owned by the whole commercial organization, whether or not anyone has been made accountable for it.

This is why commercial-execution AI needs more than a collection of agents. It needs shared commercial context, domain-specific data models, deterministic controls, system integrations and an auditable understanding of who can approve what.

Gartner’s buyer guidance, written for supply-chain planning rather than bidding but reading across cleanly, asks for much the same: unified real-time data, robust integration across planning and execution systems, and transparent governance with clear guardrails, human hand-off points and audit mechanisms. The differentiator is not the agent. It is everything the agent has to stand on.

In practice, it needs to do five things well.

1. Find the right opportunity before the clock starts

Generic AI can monitor sources and summarize notices. Commercial-execution AI has to go further: ingest public and private opportunities, structure documents in multiple languages, identify deadlines and eligibility rules, and match buyer requirements against the supplier’s actual portfolio.

The output should not be another alert. It should be a qualified opportunity, with strategic fit, historical context, delivery constraints, risks, owners and a defensible bid/no-bid rationale already attached.

The first margin decision is often whether to bid at all.

2. Resolve the product truth line by line

Buyer files rarely speak the same language as supplier catalogues. They combine free-text descriptions, legacy identifiers, competitor references, distributor codes, pack sizes and units of measure. A single response may require hundreds or thousands of line-item decisions.

Commercial-execution AI has to normalize that language, cross-reference it against approved SKUs and substitutes, and show the evidence and confidence behind every match. High-confidence mappings can move quickly. Ambiguous or high-consequence matches should be escalated to a human reviewer.

This is product intelligence applied at the precise point where revenue is either captured or left blank.

3. Put margin inside the workflow

A bid can be compliant and still destroy value.

Where price carries as much of the award decision as the European data shows it does, the margin outcome is largely settled at the moment of pricing, and largely settled blind. Award history, competitive behavior and cost-to-serve tend to sit in different systems from the one the bid is being assembled in, when they are captured at all.

Commercial-execution AI has to connect opportunity data with award history, competitive behavior, cost-to-serve, supply risk, rebates, volume breaks, currency exposure, indexation and contract duration. It should model scenarios, estimate the trade-off between win probability and profitability, and enforce price corridors, floors and exception rules.

Humans should remain accountable for commercial strategy and walk-away decisions. AI’s role is to make the economics visible early enough to influence the bid, not to report the erosion after the contract is signed.

4. Build the response around proof

Fluent prose is easy. Defensible commitments are difficult.

A genuine execution system should convert the contract pack into a structured compliance matrix, retrieve approved answers, claims and evidence, flag gaps or outdated artefacts, assign actions across commercial, legal, quality and regulatory teams, and preserve version control, approval history and source traceability.

It should know when it can reuse an approved response, when an expert must validate it, and when the system must stop. In regulated markets, human oversight is not a concession to weak AI. It is how accountability becomes scalable.

5. Learn from the commercial result

Most bid tools stop at submission. That prevents the system from learning the part that matters, and the part that matters is measurable.

World Commerce & Contracting puts average contract value leakage at 5.4%, and found that 70% of respondents acknowledged a disconnect between how contracts are managed and how financial performance is overseen. On a €500 million contracted revenue base, that is €27 million a year going missing in a place nobody owns.

Commercial-execution AI should capture the award decision, winning price, competitor behavior, buyer feedback, contract performance and subsequent leakage. Those outcomes should then improve future opportunity matching, bid/no-bid scoring, pricing recommendations and response strategy.

Without that closed loop, agents can repeat the same intelligent-looking mistake at greater speed.

The scoreboard has to change

The number of agents deployed is an adoption metric. Documents generated, prompts answered and hours saved are productivity metrics. They are useful. They are not the commercial outcome.

For bid and contracting teams, the harder measures are:

• • opportunities qualified early enough to act;
  • line-item coverage and response completeness;
  • decision, approval and submission cycle time;
  • win rate and contribution margin;
  • contract leakage and performance after award; and
  • whether teams actually trust and use the system.

One data point, offered with the disclosure that it is our own and self-reported. In a 12-month Vamstar deployment with a top-10 global pharmaceutical company, anonymized under NDA, generics and biosimilars, active in 165 markets, the programme went live across 25 high-impact markets and delivered a 73% improvement in contracting-process efficiency, a 17% increase in win rate across deployed markets and 9% margin expansion on contract revenue. Local-team adoption reached 86% inside the year, which matters more than it first sounds: it followed an internal programme that had absorbed $6 million and four years without reaching scale.

Those results should not be generalized to every programme. The point is narrower and harder. Commercial AI should be accountable to business outcomes, not to the size of its agent catalogue.

Infrastructure is the starting line

Agentic AI will become part of the enterprise technology stack. Horizontal platforms are making agents easier to build and govern. Domain-specific platforms are proving the value of industry intelligence across life sciences. Both matter, and both are prerequisites rather than destinations.

Commercial advantage will be created one level deeper, where AI understands the contract, the product, the price, the evidence, the approval policy and the outcome as one connected workflow.

Commercial teams do not need the next hundred agents. The count will keep moving, which is precisely why it makes a poor scoreboard. What teams need is a system that can keep a bid commercially coherent from opportunity to award, and that knows when to act, when to ask and when to stop.

The winners will not be the companies with the most agents. They will be the companies with fewer missed opportunities, faster decisions, more defensible commitments and higher-quality revenue.

Agentic AI can automate tasks. Commercial-execution AI has to change the P&L.

Sources: Medicines for Europe, position paper on the EU Public Procurement Directive (10 February 2026); Gartner press release, “Gartner Predicts Over 40% of Agentic AI Projects Will Be Canceled by End of 2027” (25 June 2025); Gartner press release, “Gartner Warns of Agent Washing Risks in Supply Chain Planning Technology Market” (20 May 2026); World Commerce & Contracting, “Smarter Contracts, Better Margins” (22 September 2025); Vamstar case study, “Redefining Contracting Strategy: AI Enables Pharma Cost Optimization” (accessed 12 August 2026).