Data procurement leaders - MCP, pricing, governance, AI

We held our most recent Data Procurement Leaders Roundtable this week. The conversation covered a lot of ground, but it kept circling back to one theme: governance, and specifically the gap between what AI and MCP promise and what they actually cost to deliver.

Governance is the thread running through everything

Governance came up repeatedly, both in relation to MCP specifically and to data use and storage more generally. A lot of vendors are building licensing terms around AI that don't hold up well in practice, particularly for organisations already running tools like Copilot. Vendors want to place restrictions that simply don't map onto how those tools are actually used day to day.

Vendors are selling possibility to users, and procurement inherits the bill

One of the more consistent frustrations: vendors talking directly to end users about everything AI and MCP could do, without making clear that any of it comes at extra cost. The user gets excited about the possibility, then that conversation lands on procurement's desk, and stalls the moment the price tag becomes clear. It's a pattern that repeats often enough to be a genuine structural problem, not a one-off misunderstanding.

Data teams and innovation teams aren't always rowing in the same direction

AI and innovation teams frequently don't come from a data background, and don't have the same depth of experience negotiating licensing. That creates real budget tension with data and procurement teams, especially when both are negotiating with the same vendors from different starting points. In some cases, innovation teams end up building entirely new solutions for problems a simple search of an existing, well-established database would already solve. Reinventing the wheel, at a cost.

Vendors are mostly good to work with, just inconsistent

Worth saying plainly: most vendors are collaborative, and the instinct shouldn't be to default to saying no. The real problem is consistency. There's no common approach across the market, so every negotiation effectively starts from scratch, and outcomes depend heavily on who you're dealing with rather than any shared standard.

The real cost of MCP doesn't always add up

Several in the room are seeing MCP uplift requests running into the hundreds of thousands. At that price, it's often cheaper, and far more productive, to hire additional corporate finance analysts to do the work directly.

This wasn't a single anecdote. It came up independently, several times, from different organisations comparing notes with each other. The consistency of it is the real finding here: for a meaningful share of use cases, hiring people is not just the cheaper option on paper, it's also the one that delivers more actual output and business impact for the money. That's a hard thing for a vendor to argue against, and it's increasingly the comparison procurement teams are being asked to make.

That comparison lands badly with internal stakeholders, who understandably struggle to see why accessing effectively the same data through a different piece of technology should cost significantly more.

“AI” is sometimes doing a lot of pricing work it hasn't earned

There's a pattern of vendors raising prices and pointing to AI investment as the justification, even when no AI is named anywhere in the contract. In practice, what's often being delivered is a fix to search or product issues that were overdue regardless, not a genuine AI-driven improvement. It's a distinction worth pushing on in any renewal conversation.

MCP doesn't always mean the full picture

A recurring source of confusion: users assume that if a product has an MCP integration, that gives them the complete view of that vendor's data. Often it doesn't. Not all of a vendor's data is necessarily exposed through MCP, so users still end up going back to the vendor's own portal to get the full picture. That gap undermines the case for paying a premium for MCP access in the first place, and it's catching people out.