Your bill grew.
Your traffic didn’t.
Summit Integration cuts Dynatrace bills and then keeps them cut. Send us last month’s consumption statement and a usage export. Within 48 hours you get a written breakdown of where the money goes and roughly what’s recoverable.
Free, and there’s nothing to sign.
No call needed before we look.
Dynatrace specialists. We also work on Datadog, Elastic and Splunk.
Nobody decided to spend this.
Dynatrace consumption doesn’t climb because somebody made a bad decision. It climbs because of four things that each look harmless on the day they happen, then compound quietly between statements.
Monitoring mode set once, never revisited
OneAgent goes on and every host gets Full-Stack, because that is the setting that makes the demo work. Two years later the batch nodes, the jump boxes and the internal tooling fleet are all still billing at Full-Stack rates for code-level detail nobody has opened. On most estates we look at, this single setting is the largest number on the statement.
Log volume outpacing traffic
Debug logging switched on during an incident and never switched off. A library upgrade that got chattier. A new sidecar on every pod. Log volume tracks the number of things emitting logs, not the number of customers you serve, so it climbs in quarters when revenue does not.
One blanket retention across every bucket
Grail lets you set retention per bucket and per data type. Most estates set it once, at the top, and inherit it everywhere. So the debug logs nobody queries after a fortnight are kept as long as the audit data that genuinely has to be, and both are paid for at the same rate.
Nobody owns the budget
The platform team owns the contract. The application teams create the consumption. Neither has a view of the other’s side, and the statement lands with finance, who cannot tell which team caused what. So the conversation happens once a year, at renewal, when it is too late to be about engineering.
Priced in the open.
Three steps. You can stop after any of them, and most of the value of the first is that it tells you whether the rest is worth buying.
Bill review
Send last month’s consumption statement and a usage export. We come back in writing with where the money is going, which levers apply to your estate, and a rough figure for what is recoverable.
- No call required before we look
- You keep the analysis whether or not you buy anything
Remediation
We work through the estate and implement the cuts: monitoring modes host by host, OpenPipeline filtering, bucket layout and retention. One price, whatever we find, and it does not move once we have agreed it.
- A fixed fee, not an hourly rate and not a percentage of what we save you
- Every change reversible, and staged so you can watch consumption respond
- You get the written analysis as well, so your own team can maintain it
Cost governance
The layer that stops the savings evaporating. Without it, consumption re-inflates within two or three quarters as new services and new teams are onboarded.
- Consumption budgets per team, with the numbers visible to the teams themselves
- Spend anomaly alerting, so you find out in days rather than at renewal
- Monthly review, quarterly re-optimisation
- OpenPipeline guardrails, so the default is cheap
- Cost review of every new service before it starts emitting
We do not do gain-share. Taking a percentage of savings turns every month into an argument about what the baseline was, and it caps our upside at exactly the point the estate becomes well-run, which is the point we would rather be useful. Fixed fees keep the incentives boring.
What we actually touch.
Not a methodology. These are the specific settings, pipelines and buckets where the money is.
Dynatrace
Where we spend most of our time, and the platform two of us work on daily. Under DPS almost every lever is either a monitoring mode you can downgrade or data you can drop at the edge before it is billed.
Full-Stack vs. Infrastructure Monitoring per host usually the largest lever
Code-level detail where you need it, infrastructure-only where you don’t. Set host by host rather than as an estate-wide default. We work through the fleet with the owners, because the argument is never technical: it is about who is willing to give up a view they never use.
Grail retention buckets per data type
Retention set per bucket and per data type instead of one blanket period. Debug logs get a fortnight, audit data gets what compliance actually requires, and the two stop costing the same.
OpenPipeline filtering, masking and routing
Drop at the edge, before ingest. Data you will never query costs nothing if it never arrives, and masking at the same point solves a compliance problem while it is solving a cost one.
Log ingest rules and bucket assignment
What gets ingested at all, and which bucket it lands in once it does, which decides what it costs to keep for the next year.
Custom metrics and events consumption
Consumption driven by custom metrics and events, under DPS or under classic DDU licensing. The units differ between the two models, the discipline does not, and we establish which one you are actually on before quoting a saving.
RUM and Session Replay sampling
Session Replay is billed separately from RUM and is captured, in our experience, at a rate well above the rate anyone watches it back.
Other platforms
Dynatrace is the focus, but the work transfers. If your estate is on one of these, or on more than one, the engagement is the same and so is the price list.
Datadog
- Custom metric cardinality. Billed per unique timeseries and usually the largest line. We find the tags doing the multiplying and cut them at source, or keep the dimension queryable with Metrics without Limits rather than billable.
- Ingest vs. index. Exclusion filters so you ingest everything and index only what gets queried, plus Flex Logs for the retention you keep for compliance rather than debugging.
- APM spans and host counting. Retention filters that keep the traces worth keeping, how hosts count against your commitment, and synthetic tests still running against endpoints that were retired.
Elastic
- ILM and tiering. Hot, warm, cold and frozen with the transitions set from how the data is actually queried, and searchable snapshots so frozen data is served from object storage.
- Shard and mapping hygiene. Over-sharding that costs you nodes the workload does not need, and dynamic mapping turning every new field into an indexed field permanently.
- Data stream retention. Set deliberately on the stream rather than inherited from whatever the index template said when it was written.
Splunk
- Index-time filtering. Routing noise to nullQueue so it never counts against the licence. This has to happen where events are parsed, on a heavy forwarder or the indexer, which is the part most teams get wrong when they try it on universal forwarders.
- Ingest Actions and SmartStore. Filter, mask and route at ingest, including straight to S3 for data you keep but never search, with storage decoupled from compute so retention stops being priced as indexing capacity.
- Pricing model first. Which levers pay off depends on whether you are on ingest-volume or workload pricing. We establish that before quoting anything.
Cutting the bill is a project. Keeping it cut is a subscription.
Two or three quarters after a good remediation, the number is climbing again. Not because the work was wrong, but because the estate kept moving. New services ship with OneAgent on default settings. A team adds a dimension that looks harmless. Somebody raises retention during an incident and never lowers it again.
Nobody is watching the meter between statements, so the drift is invisible until it isn’t. Within a year you are roughly back where you started, and the audit you paid for has become a document about the past.
That is the honest reason the retainer exists, and we would rather write it down than dress it up. If your estate were static you would buy the remediation once and never need to speak to us again. It is not static, so the savings need somebody to defend them: budgets per team, alerting on consumption the way you already alert on latency, and a cost review of every new service before it starts emitting.
Four engineers. No client list yet.
Summit Integration is new, and we would rather say so than imply a history the company does not have. What isn’t new is the experience: our engineers ran observability platforms at enterprise scale before founding Summit, at previous employers.
That experience belongs to the people, not to the company, and we will be specific about it on a call: which estates, which platforms, which scale. What you will not find on this site is a logo wall we have not earned or a case study we cannot put a reference behind.
Send us last month’s bill.
Last month’s consumption statement and a usage export are enough to start. Within 48 hours you get back a written breakdown: where the money is going, which levers apply to your estate, and a rough figure for what is recoverable.
It is free, there is nothing to sign, and you keep the analysis whether or not you buy anything afterwards. If the answer is that your estate is already well-run, we will tell you that. It is a short email and it saves us both a call.
- Turnaround
- 48 hours, in writing
- Primary platform
- Dynatrace. Also Datadog, Elastic, Splunk.
- Send the statement
- info@summit-integration.com
- Company
- Summit Integration Group LLC
Start the review
Six fields. We reply within 48 hours.
We’ll be in touch within 48 hours.
Next step: email last month’s consumption statement and a usage export to info@summit-integration.com. The review starts when that arrives.