This autumn SAP put agents into the heart of its transformation toolchain. At its Transformation Excellence Summit, SAP announced a Process Consulting Agent in Signavio that mines processes across SAP and non-SAP systems to show where agents can help, an AI Agent Hub that maps agents, models and MCP servers to capabilities and owners, LeanIX assistants for architecture and EU AI Act risk classification, UI-native agents in WalkMe, and Cloud ALM assistants that support analysis, configuration, testing and rollout, then track what agents do in production.
The tools are real, and they will change how ERP programmes are designed, built and adopted. Which means that, very soon, every proposal you receive will call itself "AI-first". Gartner has a name for the version that isn't: agent washing. Its June 2025 research put more than 40% of agentic AI projects on course for cancellation by the end of 2027, for three reasons that will sound familiar to anyone who has evaluated an ERP bid: escalating costs, unclear business value and inadequate risk controls.
Where "AI-first" shows up in a bid today
Across 30+ years of SAP and enterprise deals, we have learned to read a proposal from the back. Today, AI usually appears in three places: adjectives in the executive summary, an "accelerators" appendix, and a slide of logos. It rarely appears in the one place that matters, the effort model. If agents really turn weeks of process analysis into hours and generate test cases on demand, the plan should be shorter in those phases, the team smaller, and the price should show it, or the bidder should explain exactly where the saving went.
AI-first is not a claim. It is a difference you can find in the plan, the price and the contract.
Where agents should change the plan
Select a phase to see the SAP tooling now in play, what should look different in a credible proposal, the evidence to ask for, and who must own the output.
Seven places AI should show up in an ERP programme
Tooling reflects SAP's 2026 announcements. Savings will vary by programme, so ask for evidence, not adjectives.
Does the price show the saving?
This is the simplest test, and the one most often missed. If a bidder says agents cut effort by a quarter, a buyer should be able to see roughly where that quarter went. Some of it may fairly go to tool licences, a smarter team or extra assurance. None of it should simply disappear.
Where did the AI saving go?
Enter the figures from a proposal. The defaults are illustrative.
AI-first, AI-assisted, or AI bolted on?
Use these ten questions on any proposal that calls itself AI-first. They work equally well as an evaluation sub-criterion for buyers and as a red-team checklist for bidders.
Score a proposal in five minutes
Answer Yes, Partly or No for the proposal in front of you.
What to write into the contract
Most of the risk Gartner describes, unclear value and weak risk controls, is cheaper to settle before signature than after go-live. In our experience the clauses that matter are short:
- Named tools and licence ownership. Which agents and platforms are used, who pays for them during the programme, and who pays after it.
- Human sign-off. Every agent output that becomes a deliverable, from a process design to a test result, has a named reviewer on each side.
- An agent register. What each agent does, which data it can reach, and who owns it, kept current from the first sprint.
- Regulatory responsibility. Who classifies AI use under the EU AI Act and keeps the documentation.
- A measured baseline. The effort and timeline the savings are measured against, and how any savings beyond the plan are shared.
- A fallback. If an agent underperforms, the work is done the conventional way at no extra cost, not through a change request.
A note for the partners bidding
If you are an SAP partner with a genuine AI-first delivery model, this is your moment, and the way to win it is with evidence rather than volume. Show the effort model with and without the agents, phase by phase. Price the licences openly. Name who reviews agent output. Bring one comparable programme with measured results. Evaluators facing five proposals that all say "AI-first" will score the one that proves it.
Download the playbook
Our free AI-First ERP Evidence Playbook puts this on paper: the seven-phase map with tooling, plan changes, evidence and ownership; the savings worksheet; the ten-question scorecard laid out for two bidders; the contract clause checklist; and a bid checklist for partners. Every statistic is sourced.
The AI-First ERP Evidence Playbook
8 pages. No sign-up, no email required.
Buying ERP, or bidding for it? For buyers, our buy-side tender support builds AI evidence into your requirement and scoring, so every claim is comparable. For partners, our Red team reviews and Bid Director projects test your AI story the way an evaluator will. Book a free discovery call or email sales@dealcraftadvisory.ie.
Sources
- Futurum Group, SAP Links AI Agents to Business Processes and Oversight, on SAP's announcements at the Transformation Excellence Summit, September 2026. Customer time savings quoted there are not independently verified.
- SAP News, Navigating the Transition from SAP Solution Manager to SAP Cloud ALM, June 2026.
- Gartner, prediction that over 40% of agentic AI projects will be cancelled by the end of 2027, June 2025, as reported by W.Media.
- SAPinsider, reporting Xmateria's research on SAP data migration delays, July 2026.
The phase map, scorecard thresholds and contract checklist reflect our practitioner experience, not a published standard. SAP, Signavio, LeanIX, WalkMe and Cloud ALM are trademarks of SAP SE or its affiliates; Dealcraft Advisory is not affiliated with SAP.