Services
Enterprise AI consulting services, from first question to running system.
Six kinds of work, one standard: every system is scoped to a measurable outcome and built so your own team can understand, run and extend it. For how an engagement is structured, read the boutique AI consultancy approach.
AI readiness and discovery sprints
A discovery sprint is two to three weeks of focused work: we map one process as it actually runs, look at the data it produces and find the places where automation pays for itself. You get a concrete design, measurable targets and an honest recommendation, including, when the numbers say so, the recommendation not to build.
Worth doing when you suspect AI could remove real cost but nobody can point at the exact spot. Not worth doing when the process still changes weekly or the data does not exist yet; we say that in the first call, for free.
Document understanding and data extraction
Invoices, contracts, delivery notes, emails with attachments: most enterprise processes begin with a document a person has to read. We build document understanding pipelines that read, classify and extract structured data from those documents, validate it against your master data and hand uncertain cases to a person instead of guessing.
Worth doing when a team spends hours every day retyping or checking documents. Not worth doing at low volume; below a few hundred documents a month, a good template and a shared inbox beat a pipeline.
Enterprise knowledge and retrieval systems
Retrieval augmented generation applied to company knowledge: manuals, contracts, tickets, wikis and structured data behind one question-and-answer interface. Answers cite their sources, respect existing permissions and say "not found" instead of inventing. This is the LLM consulting work we do most often, because it creates value without touching core transactions.
Worth doing when institutional knowledge lives in a few heads or thousands of unread files. Not worth doing when the underlying documents are wrong or stale; retrieval makes bad documentation faster, not better.
Process automation with human checkpoints
We automate multi-step processes end to end, with a person approving exactly the steps where a wrong answer is expensive. Plain rules where rules are enough, models only where they earn their place, and an audit trail behind every decision either way.
Worth doing for repetitive, document-heavy processes with clear rules and tiring exceptions. Not worth doing when the process itself is broken; automating a broken process only produces mistakes faster. That is a redesign conversation first, and we say so.
SAP AI consulting and integration
AI that produces a record nobody posts into SAP is a demo. We connect document pipelines, retrieval systems and automation agents to SAP through the interfaces your Basis team already governs: OData, BAPIs, IDocs and custom ABAP where needed, built by someone who has spent years inside SAP landscapes. The same discipline applies to other ERPs and internal tools.
Worth doing when SAP is your system of record and a manual step sits between the AI output and the posting. Not a fit when policy forbids external systems touching SAP; then we stay at the edge, working on documents and data exports.
Managed operation or full handover
Every engagement ends one of two ways. Either your team takes the system over, with operator-grade documentation, training and a defined period where we stay reachable, or we keep operating it: monitoring quality, updating models and adjusting rules as your process changes, at a fixed monthly rhythm.
Handover is right when you have engineers who want to own the system. Managed operation fits when you would rather buy an outcome than build a capability. Both are fine; the one thing we do not do is disappear after go-live.
FAQ
Questions buyers actually ask.
What does an AI discovery sprint actually produce?
Three deliverables: a map of the process as it actually runs, an analysis of the data it produces, and a build recommendation with measurable targets and cost estimates. If the numbers do not justify a build, the recommendation says so. The sprint takes two to three weeks and the output is the client's to keep, whoever builds next.
How long does an enterprise AI project take?
Discovery takes two to three weeks. A first working version on the client's own documents typically lands within the first month of the build. Most systems reach production in three to six months, depending on integration depth and approval cycles on the client side. Anything promised in days is a demo, not a system.
When should a company not use AI?
When a plain rule can do the job, when the process still changes every week, when the data does not exist yet, or when nobody would be allowed to act on the output. In those cases Elpis Technology recommends the simpler fix and skips the build; that is cheaper for the client and better work.
How does it integrate with our SAP system?
Through the interfaces the client's Basis team already governs: OData services, BAPIs, IDocs and custom ABAP where necessary. Nothing bypasses the authorization concept, and every posting keeps a traceable origin. Elpis Technology's background is SAP development, so integration is designed at the start, not bolted on at the end.
How is AI consulting priced?
The discovery sprint is a fixed-scope, fixed-price package quoted before it starts. Builds are priced per project after discovery, once scope and targets are concrete. Managed operation runs at a fixed monthly rate. Elpis Technology does not sell open-ended day rates without a defined outcome.
Do you work with companies outside Turkey?
Yes. Elpis Technology is based in Istanbul and runs engagements remotely for clients across Europe and beyond, in English or Turkish. Collaboration runs through shared repositories, regular calls and short written updates, and workshops or go-lives happen on site when they earn the trip.
Who runs the system after handover?
The client's team does, and the handover is designed for that: documentation written for operators, training sessions and a defined period of reachability for questions. Clients who prefer not to own the system have Elpis Technology operate it at a fixed monthly rhythm, with the option to take it over later. Code and documentation belong to the client either way.
Not sure which of these you need?
Bring the process that annoys you. The discovery sprint exists to answer exactly that question.