Newsletter · · Ashutosh Agarwal

Buyers Are Coached to Demand Five Consultants Instead of Fifty - IT Services vs AI - Week of September 12, 2026

IT Services vs AI for the week of September 12, 2026: an independent ERP adviser laid out the negotiating tactic that is the whole bear case, asking integrators why a project needs 50 people instead of 5, a former SAP executive quantified the last-mile pool at roughly 10x what customers pay SAP, forecasts put AI implementation services growing from $18 billion to $76 billion by 2031, and a 700-person bank cut its third-party software count for the first time ever.

IT Services vs AI

Week of September 12, 2026: Buyers Are Coached to Demand Five Consultants Instead of Fifty

TL;DR

  • The disruption argument got specific this week. An independent ERP consultant (Eric Kimberling of Third Stage) told buyers to look a big system integrator in the eye and ask why a project needs 50 people instead of 5, and to demand the savings, because the big firms would rather sell you their pricey AI tools than pass the savings on. That is the whole bear case in one negotiating tactic. (Transformation Ground Control, Sep 9)
  • But the counter-number is real, too. One podcast cited forecasts that the AI "consulting and implementation services" market grows from $18 billion last year to $76 billion by 2031, the services pie getting bigger, not smaller. The bull and bear can both point to hard figures right now. (AI to ROI, Sep 9)
  • The thing that actually attacks integrator revenue, the expensive "last mile" of customizing packaged software, got a name and a number this week: for every $1 a company pays SAP, it spends roughly 10x to make the software fit, and a former JP Morgan and ex-SAP executive is now selling a tool built to collapse that cost. (WBSRocks, Sep 9)

What's new

The five developments most likely to move numbers or the thesis, ranked by how actionable they are for a book. Worth noting up front: this was a strong read-through week, lots of people describing exactly how AI eats consulting work, but, for the eighth week running, no podcast directly discussed Accenture, IBM, Infosys, Wipro, TCS or Cognizant's own business. Everything below is the mechanism, not the name.

1. A consultant tells clients exactly how to squeeze the big integrators, and names the trick they'll use to fight back.

The single most useful thing said all week came from Eric Kimberling, CEO of Third Stage Consulting, an independent (vendor-agnostic) ERP advisory firm, on Transformation Ground Control (Sep 9). He walked buyers through the negotiation directly:

"You propose 50 people on this project. Do I really need 50 people? Do I need those 20 people in India that are doing development work? Couldn't I get that down to five with AI? Why not? If you can't, then why not? Maybe I should look at other options."

Two things make this matter for the integrator names. First, he described where the cuts land: the offshore development pools and the "PowerPoint jockeys." On his own work he said, "I literally spend zero time now doing PowerPoint work," the slide-and-deck labor and the offshore build labor are precisely the pyramid layers that Accenture, the Big Four and the Indian IT firms staff most heavily.

Second, and this is the part a bull should not skip, he named the counter-move the big firms are already making. Rather than cut headcount and pass the savings to clients, he says the large integrators are pitching their own AI tools:

"Our AI tool set might actually be more expensive than using all these people. So that's the commercial model conflict I think we're going to have, people will try to find a way to replace the revenue savings or the efficiency gain of AI through other means."

Why it moves the thesis: if integrators can re-package the AI productivity gain as a new, high-margin proprietary-tool line, revenue-per-project may hold even as headcount falls, that is the bull path to margin expansion. If clients instead do what Kimberling is coaching them to do (demand 5 people, not 50), it is the bear path to revenue deflation. This week we got the clearest articulation yet of the fork. He also repeated the industry's dirty statistic, transformation project failure rates of "80 to 85% or more," the pain that makes buyers willing to try something new.

2. The "last mile," the real profit pool for integrators, finally got quantified and directly targeted.

Sam Ramji, a former JP Morgan executive and long-time SAP veteran, now SVP of Product at mid-market ERP vendor Everest, laid out the economics that actually drive systems-integrator revenue, on WBSRocks (Sep 9):

"For like every dollar that SAP gives you to provide most of the best-practice capability, you're spending 10x of that to close that last mile."

The "last mile" is all the work to make generic software fit a specific company: the configuration, the custom code, the industry and regional and regulatory tweaks. That 10x, he says, is exactly what flows to "the system implementer route." Everest has productized a tool ("AI Specify") aimed at that last mile, taking work that "used to take weeks and months" down to "a version one in a matter of hours or days." Notably, he framed it as not vibe-coding: it runs a formal software-development lifecycle, requirements, specifications, quality checks, to produce "production-ready enterprise-grade code."

He also gave a clean macro read on IT budgets: CIOs face "pressure to show impact, but not a lot of new investments," so AI is being funded out of savings ("can we save money from a SaaS perspective?") rather than fresh budget. That is soft, not strong, discretionary demand.

Why it matters: the bear case for integrators has always been vague ("AI will eat consulting"). This puts a number on the target, the 10x last-mile pool, and shows a credible operator building a product to compress it. If that compression is real and repeatable, it hits the highest-value, stickiest part of an integrator's book.

3. Two more forecasts landed, and they cut in opposite directions.

On AI to ROI (Sep 9), hosts cited two market forecasts worth writing down:

  • Bull data point: the AI "consulting and implementation services" market is projected to grow from $18 billion last year to $76 billion by 2031 (attributed to Insight Partners and Demand Sage), roughly 27% a year. This is the cleanest evidence yet for the "AI grows the services pie" argument: someone still has to put these tools into production at a hospital or a bank.
  • Bear and disruptor data point: the agentic coding-tools market is projected to grow from $4 billion last year to $30 billion by 2030 (attributed to MarketsandMarkets). That is the budget line moving from "hire people to write code" toward "buy a tool that writes code."

The same episode added a structural warning for the integrators: both OpenAI and Anthropic are "building forward-deployed engineering arms in partnership with private equity companies." In plain terms, the model makers are standing up their own consulting-and-deployment teams, trying to capture the implementation dollars themselves rather than leave them to partners. (Note: both forecasts are third-party numbers relayed on a podcast, not primary company disclosures, so treat as directional.)

4. The build-vs-buy line moved, a 700-person bank cut its third-party software count for the first time ever.

On The Artificial Intelligence Show #238 (Sep 10), the CTO of PBAC Private (a roughly 700-person bank) described a genuine reversal:

"This was the first year that the number of third-party technologies our bank used actually decreased from the prior year... AI development is really changing that dynamic."

The concrete example: an account-review process they were about to buy would have cost "$375,000 plus a year"; instead their engineers built it with OpenAI's Codex, and each review now takes "50% less time." They have also deployed "digital employees," one, "Alex," monitors ServiceNow tickets and is "closing roughly half the tickets of a full-time employee"; a support chatbot, "Penny," cut a support team's tickets 60% over four months.

The nuance that matters (and tempers the bear): this small bank still could not do the biggest job alone, it "partnered with an AI company, Verapath" to tackle a 55-step loan-closing redesign. So even in a build-vs-buy reversal, complex transformation still pulls in an outside implementation partner. The work doesn't vanish; it shifts to a different, often smaller and cheaper, kind of provider. That is the whole ballgame for the integrators, do they become that partner, or get skipped?

5. A Big-3 consultant says the CRM-implementation consultant is going to "shrink and shift."

On The So What from BCG (Sep 9), Brian Gouch, who leads Commercial Tech at BCG, gave a candid read on his own industry. His view: AI agents replace "the screen," the interface, not the governed customer database underneath, so the CRM itself survives. But the people who configure it do not fare as well:

"We're going to see a shrink and a shift for those consultants who provide services for the customers of Salesforce.com, Dynamics, the other CRM providers."

His bet is that consulting moves back up the value chain, "actually consulting organizations on what they should and should not be doing... and the agents are going to be doing the work," and that contracts move "towards outcomes," which he called "a tectonic shift." He also dropped two useful cost facts: agentic AI consumes "5 to 30 times more tokens per task than a standard chatbot" (attributed to Gartner), and there is a severe shortage of people who can actually build this (the Stanford AI Index counted roughly 90,000 agentic-AI jobs in 2025; BCG thinks the real need is many multiples higher). That skills shortage is, quietly, a bull argument: someone scarce and expensive still has to do the work.

The debate

Both sides had a genuinely good week. Here is the steel-man of each.

Bull, AI grows the pie and lifts margins:

  • The implementation-services market is forecast to quadruple to $76B by 2031 (AI to ROI). More AI means more projects to deploy it.
  • Most companies simply cannot do this themselves. A widely cited McKinsey figure this week: 88% of organizations use AI but only 6% generate meaningful profit from it (Everyday AI #860, Sep 11). The capability exists; the ability to capture value does not, and that gap is what consultants sell into.
  • The talent to build agentic systems is scarce (BCG: the shortage is many multiples of the roughly 90,000 jobs counted), which supports pricing power for whoever has it.
  • Integrators may re-capture the AI productivity gain as a proprietary-tool line rather than hand it back. Kimberling's feared "commercial model conflict" is, from the seller's chair, a margin opportunity.
  • Rates for the right people are rising: staffing-firm CEO Maruf Ahmed said AI specialists and forward-deployed engineers command "far higher rates," even as he sees no wage depression in standard developer roles (Human Cloud, Sep 8).

Bear, AI absorbs the billable work and breaks the headcount model:

  • The buyer playbook is now explicit: cut the project from 50 people to 5, and demand the savings (Transformation Ground Control).
  • The high-value "last mile," the 10x pool that funds integrators, is being directly targeted by purpose-built tools (WBSRocks).
  • The people who configure enterprise software will "shrink and shift," per a Big-3 consultant himself (BCG).
  • Build-vs-buy is reversing even at a 700-person bank, saving $375K on one process alone (The Artificial Intelligence Show).
  • The model makers (OpenAI, Anthropic) are building their own deployment arms, disintermediation from the top (AI to ROI).
  • The bottom of the labor pyramid is thinning: Ahmed reports "less demand for the junior-level positions" and says the "BPO business is in trouble... getting automated far more with the AI agents" (Human Cloud).

The crux, in one line: everyone agrees AI is compressing the hours. The fight is over whether integrators can convert that compression into higher-margin, tool-and-outcome revenue faster than clients can claw the savings back.

Stocks in play

None of the four primary names were discussed directly this week, so the read-through below is built from the mechanisms above, and each carries citations.

Accenture (ACN)

  • Bull: Accenture is the archetype of the firm that could turn AI productivity into a proprietary-tool-and-outcome business, exactly the "commercial model conflict" Kimberling says the big integrators are already pushing (Transformation Ground Control). The $18B to $76B implementation-services forecast is its addressable pie (AI to ROI).
  • Bear: Accenture was named by name, twice, as the kind of firm buyers are being coached to negotiate down ("Accenture, Deloitte, KPMG... just show up on day one and start doing stuff"), and its offshore-dev and slide-heavy delivery layers are the ones AI hits first (Transformation Ground Control).
  • Next catalyst / number to watch: Accenture's fiscal Q4 print, due late September, this is THE event. Watch new bookings and, specifically, the GenAI bookings run-rate; the headcount trajectory (is the pyramid still growing?); and any language on pricing of AI-augmented work. This is the fundamental follow-through the stock's summer strength still needs.

IBM (IBM)

  • Bull: IBM Software's own CTO of that unit, Anant Jhingran, spent an episode this week on enterprise data infrastructure, federation and hybrid deployment, the unglamorous plumbing that agent projects sit on top of, and a genuine IBM strength (AI Radicals, Sep 9). The scarcity of people who can wire agents into governed enterprise data (BCG's point) plays to IBM Consulting.
  • Bear: the episode pointedly did not touch IBM Consulting, watsonx-for-services, deal wins, or how AI changes IBM's services revenue, so there is no fresh evidence its consulting engine is capturing the shift. The generic "consultants shrink" read-through applies to IBM's services arm as much as anyone's.
  • Next catalyst / number to watch: IBM Consulting bookings and book-to-bill, and any signal on "GenAI book of business" run-rate at the next quarterly update. Watch whether the software and infrastructure mix cushions services softness.

Infosys (INFY)

  • Bull: none surfaced directly this week.
  • Bear: the read-through is squarely negative for the offshore model. Buyers are being told to question "those 20 people in India that are doing development work" (Transformation Ground Control); a staffing CEO says the Indian outsourcers running BPO have "a lot of angst" and that junior and entry-level roles are the ones AI is thinning (Human Cloud). This is the fresher-hiring and labor-pyramid theme playing out against Infosys's core model.
  • Next catalyst / number to watch: fresher (entry-level) hiring plans and the headcount-vs-revenue gap in the next quarterly update; any commentary on pricing of AI-delivered work. Caveat: Indian IT has had no direct English-language podcast coverage for eight straight weeks, so treat the lack of direct evidence as a data gap, not a verdict.

Wipro (WIT)

  • Bull: none surfaced directly this week.
  • Bear: same offshore-model read-through as Infosys, the "last mile" compression (WBSRocks) and the 50 to 5 staffing pressure (Transformation Ground Control) hit Wipro's book directly, and Wipro carries the added burden of weaker relative deal momentum going in.
  • Next catalyst / number to watch: large-deal total contract value and book-to-bill; margin defense as pricing on AI-augmented work comes under pressure.

Read-throughs

TCS, Cognizant (CTSH), Capgemini, EPAM: no direct coverage. The offshore-development and ERP/CRM-implementation read-through applies to all four. Capgemini and EPAM (heavier in custom engineering) sit most directly in the path of coding-tool substitution; the $4B to $30B agentic-coding-tools forecast (AI to ROI) is the budget line moving against custom-build revenue.

Salesforce / Agentforce (CRM): BCG's core message is reassuring for the platform and cautionary for the implementers: agents replace the interface, not the governed data backbone, so CRM survives, but "consultants who provide services for the customers of Salesforce.com" will shrink and shift (The So What from BCG). Also notable: Salesforce's own reported AI spend, "$300 million on Anthropic" against a "$6 billion a year" engineering budget (roughly 5%), per a figure Marc Benioff gave and relayed on 20VC (20VC, Sep 7).

ServiceNow (NOW): appears this week only as plumbing that gets automated on top of, the 700-person bank's "Alex" digital employee monitors and closes ServiceNow tickets at "half the rate of a full-time employee" (The Artificial Intelligence Show). Read that two ways: agents are being built around the ServiceNow workflow (sticky), but the human labor that used to service those tickets is exactly what's being removed.

Workday (WDAY), SAP / Joule (SAP): no dedicated coverage this week, but the ERP "last-mile" attack (WBSRocks) is aimed straight at the customization-and-configuration revenue that surrounds SAP deployments, a continuation of last week's Workday-agents disintermediation thread, now on the ERP side.

Microsoft / GitHub Copilot (MSFT): Copilot came up mainly as a share-loser in the coding wars, one enterprise "shifted 70% of its coding work to Claude Code" and switchers "moved off Copilot once Copilot went to usage-based pricing" (AI to ROI). The category is exploding, but Copilot is no longer the default winner.

In-house build vs. buy: the strongest signal of the week. A 700-person bank reversing its build-vs-buy ratio for the first time ever (The Artificial Intelligence Show) tells you the "buy the tool, build the last mile yourself" muscle is now within reach of mid-size enterprises, the exact customers integrators rely on for volume. The offset: those same enterprises still reach for an outside partner (Verapath, in this case) on the genuinely hard jobs.

Service-desk and MSP automation (a live analog): MSP-software firm Syncro put dated numbers on record (Business of Tech, Sep 7): on track for 30% of level-1 and level-2 technician work handled autonomously this year, 50% by 2027; 15% to 25% of tickets currently resolved autonomously; ticket summaries alone are resolving tickets "about 25% faster." Different corner of IT services, but the same physics, routine support labor is being automated out, and pricing is moving to a per-ticket and credit model.

What changed vs last week

Last week's issue was built around the Cursor and SpaceX headline (SpaceX acquiring Cursor and OpenAI cutting off Cursor's model access, all podcast-sourced, not primary-confirmed). Here is what moved:

  • The Cursor and OpenAI story firmed slightly, but is still not primary-confirmed. A second, independent podcast this week repeated that OpenAI is "taking their foundational model away from Cursor now that Elon Musk and SpaceX owns it," adding color that it is "5% of Cursor's model value" and that Claude remains "the main squeeze" inside Cursor (AI to ROI). Two independent podcasts now assert it; still no primary confirmation, so keep the caveat.
  • The "model labs go into services" thread extended. Last week it was Eno Reyes's "sovereign intelligence" idea (labs going after every industry). This week it hardened into a concrete claim: OpenAI and Anthropic are "building forward-deployed engineering arms in partnership with private equity companies" (AI to ROI), and Anthropic is shipping named vertical products (Claude for legal, financial services, life sciences) that compete with its own partners.
  • The disintermediation story rotated vendors. Last week's cleanest number was Workday (22,000 agents built by non-engineers). This week it moved to ERP's last mile (Everest/Ramji) and CRM implementation (BCG), same theme, new evidence.
  • The labor debate got more balanced. Last week's fresh bull counter-evidence was Revelio/Ramp (AI-heavy adopters hiring faster, a "J-curve"). This week's labor read from Dexian's CEO is more mixed: rates are not falling and staffing is "even more important," but the BPO model is "in trouble" and junior and entry-level demand is slowing (Human Cloud). Net: the top of the labor market is fine; the bottom of the pyramid, the fresher and junior layer the Indian IT model is built on, is where the pressure is real.
  • New bull evidence appeared for the first time in a while: the $18B to $76B implementation-services forecast is the most explicit "AI grows the services pie" data point we've logged.
  • Accenture's fiscal Q4 is still ahead. The catalyst we flagged last week (late-September ACN fiscal Q4) has not yet arrived. It remains the single most important near-term event.