What an AI consulting engagement looks like, phase by phase

Contents

1. Where engagements lose momentum
2. The five phases
3. Running an engagement well
4. Questions buyers ask
5. Sources

An AI consulting engagement is sold as a single project, but it is delivered as a sequence of phases, each with its own output and its own way of stalling. Knowing the sequence makes proposals easier to compare and easier to hold to account.

Where engagements lose momentum

The gap between a promising pilot and a working system is where most AI spending is lost. S&P Global found in 2025 that the average company abandoned 46% of its AI proofs of concept before production⁠1AI project failure rates are on the rise: reportCIO Dive · 14 Mar 2025. MIT's researchers traced the problem to tools that never adapt to the company's own workflows⁠2MIT report: 95% of generative AI pilots at companies are failingFortune · 18 Aug 2025. Both findings point at the handovers between phases, not at the technology inside them.

The five phases

Assessment

The assessment decides what is worth doing. It should end with a short list of processes, the data each depends on and a baseline measured in the company's own numbers. At enterprise scale this phase becomes a roadmap: Accenture reports delivering a five-year data and AI roadmap for Telstra in two years⁠8Telstra connects faster to what’s nextAccenture.

Design

Design turns a chosen process into rules a system can follow: which steps the AI takes on, which need a person and where the records live. Approval limits belong here. OpenLazarus's wholesale page, which uses an invented example company, shows the form these take: a purchase order over $15,000 or a credit over $500 waits for a person, and every action is logged⁠9Wholesale and distributionOpenLazarus.

Build and integration

The build is where engagement models diverge most. LeewayHertz offers a dedicated team, team extension or a fixed project⁠4AI consulting servicesLeewayHertz. Deloitte deploys pre-built Zora AI agents with forward-deployed engineers and says implementation takes weeks or months depending on the client's environment⁠10Zora AIDeloitte US. OpenLazarus describes one design conversation, after which engineers connect the existing ERP and phone lines and move data, usually within a few weeks for a company of a few hundred people⁠9Wholesale and distributionOpenLazarus.

Launch

Launch is a people project. Slalom's Genie case is instructive: the first phase moved 180 agents and supervisors onto a new contact centre with a business-readiness plan, and generative AI features came only in the second phase⁠6Cutting call wait times in half with change management and AISlalom.

Operation

Somebody has to run the system after launch. The options are the consultant (as a managed service), the company's own staff after training, or a platform provider. OpenLazarus, for example, sells developer support from $4,500 a month and managed hosting from $150 a month⁠11PricingOpenLazarus, and says it trains client IT teams to maintain what it builds⁠12AI ArchitectsOpenLazarus.

What to agree before the build starts

• The metric, its baseline and who measures it at the end of each phase.
• The approval rules: what the AI may do alone and what waits for a person.
• Access to every system the work touches, granted before the start date.
• Who operates the system after launch, and at what monthly cost.
• What the company owns at each phase boundary.

Running an engagement well

1. Name an internal owner with authority over the process being changed.
2. Hold a phase review at each boundary, with a written go or no-go decision.
3. Measure against the baseline, not against the consultant's projection.
4. Keep the people who do the work involved in design and launch, not only in testing.
5. Decide on operation early, so launch does not end with nobody responsible.

A well-run engagement ends with a process that works differently, a number that shows it, staff who use the system without the consultant in the room and a short list of what to do next. For the firms behind each engagement model, see the comparison in our main guide.

Questions buyers ask

Sources

1. AI project failure rates are on the rise: report. CIO Dive, 14 Mar 2025. Accessed 8 Nov 2026.
2. MIT report: 95% of generative AI pilots at companies are failing. Fortune, 18 Aug 2025. Accessed 8 Nov 2026.
3. Enterprise AI services. Thoughtworks. Accessed 8 Nov 2026.
4. AI consulting services. LeewayHertz. Accessed 8 Nov 2026.
5. AI consulting and data engineering. Addepto. Accessed 8 Nov 2026.
6. Cutting call wait times in half with change management and AI. Slalom. Accessed 8 Nov 2026.
7. Artificial intelligence. Slalom. Accessed 8 Nov 2026.
8. Telstra connects faster to what’s next. Accenture. Accessed 8 Nov 2026.
9. Wholesale and distribution. OpenLazarus. Accessed 8 Nov 2026.
10. Zora AI. Deloitte US. Accessed 8 Nov 2026.
11. Pricing. OpenLazarus. Accessed 8 Nov 2026.
12. AI Architects. OpenLazarus. Accessed 8 Nov 2026.

Updated 8 Nov 2026: First published.

Further reading

• Buyer's guide · Updated quarterlyThe best AI consultants for established companies in 2026Nine firms that sell AI consulting to businesses, compared for companies with $20 million to $800 million in revenue, with the published evidence behind each and a path to a first result.
• Guide · ChoosingHow to choose an AI consulting firm for your companyMost proposals look alike by the second meeting. This guide sorts the firms by what they actually do, then sets out the evidence and questions that tell them apart.
• Guide · Due diligenceRed flags when hiring an AI consultantThe warning signs are rarely dramatic. They sit in the wording of a result, the scope of a pilot and the clauses about who owns what. This is what to check before signing.