AI theatre
Going through the motions is not adoption
Most firms trialling AI in 2025 did the same thing. Open ChatGPT. Paste in some text. Copy out the answer. Switch to Excel. Start again from scratch. Same workflow, same friction, same hours lost.
That is not AI adoption. That is AI theatre. The gap between firms doing AI properly and firms going through the motions is widening — and the reason is not intelligence or budget. It is infrastructure.
Last month, Anthropic’s Head of Americas opened an enterprise briefing with a striking acknowledgement: 2025 was meant to be the year AI agents transformed the enterprise, but the results were mostly premature. That is a vendor admitting, about their own technology, that the industry got ahead of itself. The promise of AI in professional services is not chat interfaces and copy-paste — it is continuous, connected workflows that carry context from one tool to the next without human intervention. Three things have now changed.
The first shift
Excel into PowerPoint, without starting over
The most common delivery workflow in a consultancy goes like this: data arrives, an analyst works through it in Excel, then manually rebuilds the narrative in PowerPoint by hand. Every time. That is where the hours go, where errors creep in, and where you lose the thread between the numbers and the story.
Claude now analyses data across multiple Excel tabs in a single session and, within the same workflow, builds a presentation from that analysis in PowerPoint. It picks up the context from Excel and builds the deck — no export, no copy-paste, one continuous workflow. This is not theoretical. Wall Street Prep ran the benchmark earlier this year: Claude scored 5.5 out of 6 for financial analysis tasks, Microsoft Copilot came in at 4.4, ChatGPT at 2.5. The gap shows up in practice, not just in demos. It is currently a research preview available on all paid Claude plans.
What this means for your firm. If your analysts are still rebuilding narratives in PowerPoint from Excel data they just finished analysing, you are paying senior people to do work that no longer needs a human in the loop. The Excel-to-PowerPoint workflow is the highest-value starting point for most consultancies right now — it is the workflow everyone has and the one that saves the most time immediately. Two caveats matter:
- This works best when your spreadsheets are tidy and well-organised. Messy data gives messy results.
- Check outputs against your normal working method before removing manual review entirely. The tool is strong, not infallible.
Claude also now supports live data connections — MSCI and LSEG are available — but these need to be set up separately through the connector directory.
The second shift
From individual experiments to firm-wide infrastructure
Here is the real problem with how most firms deploy AI right now. Every person sets up their own Claude account, their own integrations, their own prompts. The results vary wildly. Knowledge stays with the individual. When they leave, it goes with them. This is the Workflow layer of the KWA framework stalling at the individual level when you need it operating at the firm level.
The private plugin marketplace changes that. A plugin bundles skills, connections, commands, and instructions into one deployable package. An administrator creates it once, and the whole team gets it automatically. Critically, your organisation owns the plugin file — it is portable, not locked to Anthropic’s servers.
ChatGPT is a tool. A well-built plugin on Claude is infrastructure.
Anthropic demonstrated one complete workflow at their business presentation in February: run the analysis, build the PowerPoint slides, check it legally with Thomson Reuters, and file everything automatically. PWC announced a working partnership with Anthropic on the same day, focusing specifically on helping finance directors accelerate their work. That timing was not a coincidence — it signals where enterprise deployment is heading.
The honest constraint. The private plugin marketplace is only as powerful as the person building and maintaining the plugin, and most consultancies do not have that person yet. That leaves two paths: build the capability now, get ahead of your competitors, and start compounding the advantage — or wait six months for the ecosystem to mature and watch the gap between you and early movers widen.
The readiness check
Where does your firm actually sit?
Before deciding which path to take, be honest about your starting point. The 5 Steps for AI Leadership framework starts with Align and Activate for a reason. Infrastructure built on shaky foundations moves fast in the wrong direction.
Three signals you are ready to move now:
- Your data is clean and in structured tools — HubSpot, Excel, SharePoint — not scattered across emails and personal drives.
- You have someone who can own the plugin build. It does not have to be a developer, but it has to be a named person who will be responsible for it.
- Your team already uses Claude individually and trusts the output. Adoption at firm level starts with trust at individual level.
Two signals you may need to wait:
- Your core workflows are not documented yet. AI cannot systematise chaos — it produces it faster.
- Leadership has not bought in. You cannot build firm-level infrastructure without a firm-level mandate. This sits at the Govern step of the 5 Steps framework, and skipping it is why most rollouts stall at the pilot stage.
The compounding advantage
Start with the workflow everyone has
The firms building infrastructure are not just saving time today. They are creating a compounding advantage over every competitor still copy-pasting between tabs. Each workflow automated, each plugin deployed, each hour freed from admin is a head start that accumulates.
The Excel-to-PowerPoint workflow is the right place to start for most consultancies. It is the one everyone has, and the one that delivers the clearest time saving, fastest. Get that working, own the plugin, and build from there.