Mercanis's new Agentic Skill Studio offers procurement teams a way to turn working methods into reusable AI instructions. Its business significance depends on whether those instructions preserve sound judgement, expose assumptions and improve sourcing decisions rather than simply accelerate preparation
Two suppliers can submit prices that look comparable while offering very different commitments. One quotation may include delivery, another may assume a larger order, and a third may depend on a condition that is easy to overlook. Comparing them is therefore an exercise in judgement as well as arithmetic. This illustrative purchasing problem provides a useful way to examine Mercanis’s launch of Agentic Skill Studio, described in its 11 November blog. The company says procurement teams can write their methods as plain-language Skills for its Mercu assistant to apply across source-to-pay tasks. A separate launch announcement followed on 12 November.
The proposition is to give AI access to how a team works, rather than ask it to approach each case with generic instructions. Mercanis describes Skills as containing steps, rules, templates and permitted tools, with changes recorded in version history. These are supplier descriptions of the product, not independent proof that the system consistently follows every instruction or improves commercial outcomes. The useful business question is narrower than whether agents can perform procurement. It is whether an organisation can express its working knowledge clearly enough for AI to apply it, and then check that the resulting preparation supports a better decision.
The comparison needs a method before it needs an answer
In the quotation example, a team first needs to decide what it is comparing. The relevant unit might be a delivered order, a period of service or a specified production requirement. A model can help extract information, but the business must establish which differences matter. An illustrative comparison could separate confirmed terms from assumptions, flag missing information and preserve each supplier’s original wording. That is a possible application to validate, not a capability established by a demonstration. Its value would come from helping the buyer see where offers differ, rather than compressing them prematurely into a single recommendation.
Writing a reusable method can reveal disagreements that were previously hidden in individual practice. One buyer may prioritise short-term price while another places greater weight on supply continuity. Both may believe they are applying the company’s usual approach. Turning that approach into instructions creates an opportunity to make the choice explicit. It also creates a risk: an undocumented preference can acquire the appearance of policy once it is placed in a shared Skill. The team should therefore distinguish approved requirements from judgement calls and contextual guidance. Reusability is useful only when the method being reused deserves that status.
Negotiation preparation depends on the source of a claim
A supplier’s explanation for a price increase can involve market conditions, input costs and changes specific to the relationship. An assistant could help organise those claims and retrieve relevant records, while the buyer decides how much weight to give them. An illustrative analysis might show the requested increase alongside the contractual pricing basis, previous changes and any evidence the supplier has provided. It should leave gaps visible. Inferring a supplier’s capacity, margin or willingness to concede from incomplete information would be a different and less defensible exercise. Fluent preparation is useful when it clarifies evidence, not when it fills missing commercial facts with a plausible story.
Mercanis’s launch explanation presents negotiation preparation as an example and says buyers retain the decision. That framing is important: a counterproposal can change the relationship, create expectations or exchange one concession for another. The organisation needs to decide who can approve those choices. A reusable instruction should not silently convert a past negotiation outcome into a universal rule. Conditions may have changed, and a concession that made sense in one relationship could be inappropriate in another. The role of the assistant should be defined around the preparation it can support and the judgement that remains with the responsible buyer.
Skills and purchasing authority are separate layers
The supplier describes fixed workflows as holding approval thresholds and audit requirements, while Skills guide how the assistant works within them. It also says Mercu uses the same platform permissions as a buyer. These are implementation claims to test during adoption, rather than controls this article has independently audited. The distinction nevertheless expresses a useful design principle. A method for evaluating an offer is not permission to issue an order. A well-written explanation of what to do cannot replace an enforced boundary around what a system is allowed to change or communicate.
An illustrative acceptance test could give the assistant an incomplete request or a request outside a test user’s authority and observe whether it stops at the intended point. Another could check that an apparently attractive recommendation still reaches the required approver. These are proposed tests of the operating arrangement, not evidence of a particular product’s failure. The purpose is to ensure that instructions and permissions reinforce each other. If a task involves external communication, the team should also define which drafts can be prepared automatically and which commitments require review. That boundary should remain understandable to the people accountable for the supplier relationship.
A maintained method is an organisational asset
Reusable instructions create ongoing work. A policy can change, a template can become obsolete and a once-useful source can stop reflecting the relevant market. Version history helps identify a change, but the organisation still needs an owner who decides when a revision is necessary and how it should be checked. An illustrative procurement team could associate each shared method with its purpose, approved sources and review responsibility. That would make the method easier to assess when circumstances change. Without ownership, an instruction can remain readable while becoming commercially outdated, allowing consistency to reproduce an old assumption across many new cases.
There is a data foundation beneath the method as well. Supplier identities, contract versions and quotation units need to be sufficiently reliable for comparisons to make sense. A reusable Skill cannot resolve a contradiction that the organisation itself has not addressed. Earlier BCG research on agentic AI in technology procurement, based on a survey of more than 200 leaders and buyers plus interviews, identifies inconsistent data, integration difficulties and governance constraints among the obstacles to progress. This is wider contextual research, not an independent evaluation of Mercanis. It supports treating deployment as a change in working arrangements rather than only a software installation.
A small business can capture one decision process
A large purchasing function may have specialist teams and an extensive systems environment. A small manufacturer or independent service business may have one person who buys supplies alongside other responsibilities. The transferable idea is to identify a recurring decision and write down the evidence needed to prepare it. An illustrative small firm could start with comparing quotations for a familiar category, requiring the output to distinguish like-for-like terms, unresolved questions and assumptions. That is a possible adaptation to investigate, not a recommendation to purchase an enterprise platform or upload confidential supplier information to an unrestricted tool.
The exercise may produce value before automation. A clear comparison method could improve consistency between colleagues or expose information that is routinely missing from supplier responses. If AI is added, the business would then have a better basis for judging its contribution. It could check whether the assistant captures the relevant terms and whether the comparison remains traceable to the source offers. The scale of the method should match the business. An elaborate system may be difficult to maintain when purchasing volume is modest, while a bounded, well-understood process may offer a more realistic starting point for learning what assistance is useful.
Measure preparation and commercial outcomes distinctly
A quicker sourcing summary and a better supplier agreement are different results. The first may help the second, but market conditions, negotiation choices and the supplier’s alternatives also affect the agreement. BCG’s research distinguishes internal operational benefits from supplier-facing commercial outcomes and reports that the former tend to appear earlier. That is an observation from its study, not a guaranteed sequence for every firm. It gives managers a reason to avoid claiming a saving solely because an assistant produced a comparison quickly. The evaluation should ask what changed in the work and what evidence supports the claimed benefit.
A practical pilot could compare similar sourcing exercises, examining completeness, errors and the effort needed to reach an approved position. Commercial results could be assessed separately, with care over attribution and changing conditions. These are proposals, not reported returns from the new launch. The business should also examine whether the method suppresses an important exception or encourages buyers to accept a recommendation too readily. The point of reusable knowledge is not to make every case look the same. It is to bring relevant experience into the next case while preserving the ability to notice when that experience no longer applies.
Agentic Skill Studio is a concrete new example of an idea gaining importance in business AI: the organisation’s methods can become part of the system’s instructions. The launch establishes an offer to customers; its effectiveness remains something to evaluate in actual use. For procurement leaders, the deeper opportunity is to make judgement more explicit, shareable and open to challenge. That requires decisions about evidence, ownership and authority as well as model capability. An assistant becomes useful when it helps a buyer understand the available choices more clearly, leaving the organisation able to explain why the eventual purchase was the right one.
Image credit: People collaborating around a table by Dylan Gillis, used under the Unsplash License. Illustrative photograph unrelated to Mercanis or its customers. Resized for this site; WordPress may crop display versions.
