Data Engineering & Custom IT Automation Services

Data engineering and custom IT automation are converging into a single purchase. The buyer is the same executive under the same mandate — get the data estate ready for AI, then automate the processes sitting on top of it — and both categories are now sold against the same objection: we ran a pilot and it never reached production.

Both markets have pulled apart into a barbell. At one end, platform-certified specialists command $150–$275 an hour and are being acquired at speed by global integrators and private equity. At the other, offshore and no-code delivery has collapsed the floor to $40–$75 an hour — and in the automation tier, to $8.50. The independent mid-market generalist is the position being squeezed out.

Five findings

Both markets sort into the same three tiers. What separates them is not capability claims — those are near-identical across the field — but price, minimum deal size, and whether the firm owns delivery IP.

A directory analysis of 86 firms puts the global blended average at roughly $165 an hour, with a full range of $50–$275 and typical projects running three to nine months. That average conceals the barbell: the same directory's median across 64 Databricks-focused firms is about $88, dragged down by offshore delivery, while the US specialist tier holds $150–$275.

Who competes where

Rates follow the stack, not the seniority label

Onshore rate cards ladder by role, but the sharper premium comes from the technology on the engagement. Against a $140–160 baseline, dbt adds about 5%, SnowPro certification 10%, Databricks 15% — while GenAI and RAG work commands 40–60% more, at $220–300 an hour, with vector-database specialists quoted up to $400. A US-based "senior" priced below $130 is treated in buyer guides as a red flag.

Demand drivers

• AI readiness is the engine. 62% of organizations report significant AI-readiness gaps in data infrastructure, and only about one in five are satisfied with their data accuracy and completeness — while 88% now use AI in at least one function. Every acquisition in the category is framed as "data for AI".
• Migration waves keep renewing. Synapse-to-Fabric is the distinct 2025–26 wave, with Fabric architect demand up roughly 180% year over year and T-SQL incompatibility driving validation hours. Meanwhile 75% of cloud migrations exceed budget.
• FinOps became a standing line item. Snowflake and Databricks spend governance now sells as its own engagement, with vendors citing up to 70% waste reduction and 50–70% compute savings from moving batch work to cheaper compute tiers.

The automation tier splits on a single economic fact: whether the firm resells platform licences. A UiPath partner running a 25-bot client carries roughly $236,000 a year in licensing — $200,000 in licences plus $36,000 support — as pass-through annuity, with unattended robots at $6,000–$8,000 each per year. A Make or n8n shop has effectively none, which is why the boutique tier monetises through retainers instead, and why the two tiers behave like different businesses despite both selling "automation".

Who competes where

Engagement economics

Project pricing is remarkably consistent across the boutique and specialist tiers, which makes it a usable benchmark when qualifying an incoming quote.

The agentic transition

UiPath launched its agentic platform in FY2026 and returned to GAAP profitability in Q3 FY26 on revenue of $411.1M, up 16%, with ARR of $1.72B — while partnering with Deloitte, HCLTech, SAP and Anthropic. Automation Anywhere rebranded around "agentic process automation" and acquired Aisera in November 2025; SS&C Blue Prism now markets itself as an agentic automation company. Analyst consensus is that agents extend RPA rather than replace it.

The buy-side has moved faster than the marketing. Agentic AI is a top technology priority, up 31.5% year over year, but CFOs have imposed budget discipline: revenue and profit as the primary ROI metric nearly doubled to 21.7%, while "productivity gains" fell 5.8 points. Governance maturity is now a decisive procurement filter, and Gartner expects more than 40% of agentic AI projects to be cancelled by 2027.

Six positioning patterns recur across both categories. Most firms occupy two or three at once, which is why so much of the category's marketing reads identically.

1. Platform-tier depth. Climb one vendor's certification ladder and market the badge. UiPath's Diamond, Platinum and Gold hierarchy rewards certification counts, active deployments, a functioning centre of excellence and quarterly business reviews; Snowflake and Databricks run equivalent elite tiers.
2. Deliberate vendor neutrality. The counter-position — Analytics8 markets an explicit refusal of exclusive platform allegiance. It is strengthening as a differentiator precisely because platform vendors now hold equity in their partners.
3. Proprietary accelerators as moat. Reveal's RoboSuite, Roboyo's six internal platforms, WonderBotz prebuilt bundles, Lovelytics' Snowflake-to-Databricks migrator. Delivery IP shortens projects and defends premium rates against commoditisation.
4. Delivery-cost arbitrage. Nearshore is now strategy rather than discount: Aimpoint in Medellín, Indicium in Brazil, Auxis in Costa Rica and Colombia, Thoughtworks across LatAm.
5. Productized fixed-scope offers. Rittman's £7,500 two-week sprint, Pragmatiq's £6,200 five-day block, fixed-cost pods at $12,000–18,000 a month marketed as 20–30% cheaper than hourly staff augmentation.
6. Outcome and consumption pricing. Still a minority but growing — roughly 23% of large enterprises prefer value-based models for automation, and Gartner projects 40% of enterprise software spend shifting to usage or outcome models by 2030.

The single most useful distinction in this market is between a use case and a lifecycle stage. Almost every crowded offer is a use case — build me a chatbot, build me an outbound agent, process my invoices. Almost every underserved offer is a stage that comes after the build: evaluate it, govern it, secure it, integrate it, maintain it, review its output.

That asymmetry exists because building is the part that demos well and the part a generalist can learn in a weekend. Gartner estimates only around 130 of the thousands of self-described agentic AI vendors are substantive; the rest is rebadged chatbots and RPA. Competing on the build means competing with all of them. Competing on what happens after the build means competing with almost nobody — while selling to a buyer who has already spent the money and cannot walk away.

Ground that is already crowded

Where the demand is real and the supply is thin

Each of the wedges below is evidenced by the same pattern: a large share of buyers have already adopted agents, and a much smaller share have the control in place. The distance between those two numbers is the market.

1. Agent observability and evaluation. The strongest supply gap found. Roughly 85% of production generative-AI deployments run without observability, and while 89% of teams claim they have it, only 52% run offline evaluations and 37% run online evaluations. The buyer already has agents live, so there is no ROI case to argue — it sells as insurance on money already spent. Deliverables: evaluation suites, golden datasets, regression gates on agent releases, drift dashboards, observability retrofits.
2. Governance, audit and compliance attestation. About 74% of leaders expect meaningful agentic use within two years, but only 21% have a mature governance model, and 62% name security and risk — not model capability — as the primary blocker to scaling. The EU AI Act's high-risk obligations bind 2 August 2026, more than half of organizations lack a systematic AI inventory, and ISO 42001 certification is becoming a procurement gate while the certified pool is still counted in the low hundreds. This is a dated, funded demand event with countable competition.
3. Agent identity, access and security. 92% say legacy identity management cannot handle non-human identities, and half report no clear ownership of agent credentials at all. On PyPI, agent-framework downloads outpace agent-security tooling by roughly 83 to 1, a gap that widened 41% in five months.
4. Agent inventory and sprawl control. Over three million agents are estimated to be running inside companies with only about 47% monitored or secured. Nobody is selling a census-and-decommission service to the mid-market.
5. Integration where no API exists. The most quietly damning number in this report: enterprises run an average of 1,061 applications and have integrated 29% of them. The other 71% is invisible to any agent without custom wiring. Legacy surfaces (IBM i, mainframe, on-premise ERP) remain a specialist market with real scarcity value.
6. The human review layer. Human-in-the-loop tooling is growing at roughly 25% a year, 70% of practitioners say reliability comes from AI paired with human review, and 64% expect oversight needs to increase. Everyone sells the agent; almost nobody sells the reviewed escalation tier around it.
7. Maintenance and drift remediation. Organizations budget to build agents and almost never budget to maintain them; degradation typically surfaces only when customers complain. Forrester expects process intelligence to rescue 30% of failed AI projects — an explicitly sized remediation market, and the most under-sold recurring revenue line in the category.
8. Data and AI readiness. 94% of IT leaders name data quality as the determinant of AI project success and nearly 80% say AI is held back by data access. Only 16% currently treat unstructured data as a strategic priority while 60% plan to increase that spend within eighteen months — a twelve-to-eighteen month window to be positioned before it becomes obvious.

Research was conducted across public sources in July 2026: vendor and consultancy websites, partner-award announcements, merger and acquisition releases, analyst summaries (Gartner, Everest, HFS and Forrester as reported in public materials), pricing directories, earnings coverage and practitioner community discussion. Every figure is attributed to the publisher that reported it.

Confidence notes

Firm-level rate cards drawn from consulting directories are self-reported and should be treated as indicative rather than audited. Headcounts come from third-party trackers (Revelio, CB Insights, LeadIQ) and drift between refreshes. Market-size projections originate in vendor and analyst press releases whose underlying methodology is generally paywalled, and growth rates from different publishers are not directly comparable. Where a claim rests on a single source, it is stated as reported rather than as fact.